S1E3: An Architect's Verdict on OpenSpec
spec-driven-developmentopenspecai-agentsagentic-engineeringsoftware-architectureclaude-codedecision-records
Verdict: Michael Carducci went from joking about spec-driven development to using OpenSpec on a production SaaS, valuing its repeatable process and its record of the reasons behind decisions, which Michael frames through Naur's Programming as Theory Building
TL;DR
- Tim Berglund talks with the software architect Michael Carducci (recorded Sept 24, 2026) about OpenSpec, a command-line tool and set of agent skills for spec-driven development. Tim gave a talk on it at UberConf in July. Michael, who had joked about spec-driven development before, learned it from a colleague who attended Tim's session.
- Michael's verdict: OpenSpec gave a consistent, repeatable process and consistent, repeatable results. Michael says it keeps better memory of the project than a CLAUDE.md file or loose Markdown files, and has used it on a production SaaS with paying customers and SLAs ever since.
- The workflow, as Tim describes it at about the 25-minute mark: explore or propose a change, review its proposal, design document, task list and specs, apply it to write the code, optionally verify with a different model, then archive it. The specs stay in the repository and merge into a master spec.
- Michael's main argument comes from Peter Naur's "Programming as Theory Building": the program is more than the code, the code is lossy, and OpenSpec records the context behind decisions that would otherwise stay as tribal knowledge.
- The specs work in both directions. Tim, building a Lightroom replacement as a hobby, uses them to stay on track; for Michael, their value is letting an agent detour around a roadblock and return to the plan, the way GPS recalculates.
Full summary
From skeptic to user
Tim introduces Michael Carducci as an old friend: an architect, speaker, writer and magician. The trigger for the conversation was Tim's OpenSpec talk at UberConf in July, a No Fluff Just Stuff conference in Westminster, Colorado, where Michael spoke in nearly every slot. Tim jokes that filling every slot used to be called pulling a Venkat and is now pulling a Carducci.
Michael opens with a confession. Hearing "spec-driven development", Michael mapped it onto the functional and technical specifications of the 1990s, where the whole implementation was planned before any code, and even made jokes about it on a panel. Tim's LinkedIn post answering such jokes made Michael pause. Then Michael sent a colleague to UberConf, the colleague attended Tim's session, and the two spent a day "pair spec-driven developing". That, Michael says, is when everything changed.
Tim points to a semantic diffusion problem: people call many things spec-driven development, from writing a giant prompt to opinionated tool-driven approaches like OpenSpec. YouTube videos asking "Did agile teach us nothing? Why are we writing specs again?" sounded, to Tim, nothing like the actual practice.
Consistent process, consistent results
Michael found two advantages over anything tried before: a consistent, repeatable process and consistent, repeatable results. The side effect was that working with the agent kept getting better. Before, every session with Claude felt like the same conversations and reminders, like "this new person who was a very capable coder had just joined the team", even with edits to CLAUDE.md. OpenSpec keeps its own memory of the project, and Michael says it does this better than a CLAUDE.md file or a pile of Markdown files because it is part of a tool-driven process.
To anyone who dislikes the approach, Tim points out that writing software changed radically about nine or ten months before the recording, and figuring out what we are doing is a three-to-five or five-to-ten year problem. Agile took 10 to 15 years to settle, and teams did microservices badly for years. Michael goes further: in 15 years of architecture consulting, Michael has seen teams survive microservices but never personally seen one succeed with them. What changes now, Michael argues, is that humans are being taken out of the loop, so the broader ideas can't be skipped. Tim keeps the tent big: there will be good AI workflows that have nothing to do with OpenSpec.
On track and off track
Tim works on hobby projects, since a developer relations leader's job is managing people. The current one is a replacement for Adobe Lightroom, with Tim's own storage and import rules, face recognition and semantic image search. When a new feature is one prompt away, Tim needs to be kept on track, and a long conversation with Claude produced a set of proposals and explorations to work through instead of "a random walk through the feature space."
Michael has the opposite experience: OpenSpec lets Michael go off track. Michael runs a SaaS with paying customers and SLAs, works interactively, and stays away from unattended agent loops and "dark software factories". Before OpenSpec, the routine was to ideate with Fable, produce a Markdown file and hand it to Opus to build. OpenSpec's explore step makes Michael feel like an architect again, talking through constraints, what must be true, and the reasons behind decisions such as semantic data or a modular monolith. A routine dependency bump turned into a three-day yak shave, which Michael blames mostly on buggy Oracle drivers. The plan absorbed it. Michael compares it to GPS directions: when you hit a roadblock or miss an exit, the GPS recalculates and gets you back on track.
How OpenSpec works
Around the half-hour mark Tim explains the tool, noting it is unsponsored and that Tim isn't a contributor. OpenSpec is a command-line tool plus a set of skills. You install it with Homebrew, install the skills into your coding harness's directory (it supports a couple dozen harnesses), and follow a workflow:
- Propose a change, from a sentence to paragraphs and bullet points. The agent may ask what you mean or suggest a spike.
- Explore first if you don't understand the change well enough. Tim says the questions are often challenging, real architecture work.
- A change in flight has a proposal, a design document, a task list and specs in formal "when ... then ... shall" language, all in Markdown. Michael notes the specs have shades of BDD.
- Apply writes the code and runs whatever verification the agent can do; some checks may be manual. A newer verify command can bring in a different model, which Tim admits not having tried.
- Archive ends the change, but its documents stay in the OpenSpec directory, and the specs merge into a master spec for the application.
Michael adds that developers don't build software, they change it, in nested change cycles, and important context lives in previous decisions and commitments that the code doesn't show. Tim sees this in the Lightroom project, which has 50 or 60 planned changes. Sometimes the agent points out that a new change contradicts an older spec and asks what to do.
Programming as theory building
Michael's main argument rests on Peter Naur's paper "Programming as Theory Building". The program is more than the code: the code reifies the part of the program that runs in the machine, but software exists to deliver value outside the computer, so looking only at code means seeing only part of the program. Naur calls code lossy. Michael says using OpenSpec as a consistent process, even a ritual, expands the theory along with the code: its Markdown files hold the kind of learning that happens in a team. Michael's example is being sent to make a simple change and taking production down because of a load-bearing bug in a class nobody touches, knowledge that doesn't fit in comments. OpenSpec gives a way to capture "the context behind decisions, not just the code that resulted," and makes the agent part of the team.
Tim calls it guardrails, and also state about the project that the agent and harness can read. Michael says some of it turns into architectural decision records: given this context and these criteria, these options were weighed and this one chosen. The next agent won't blindly undo a decision because it is inconvenient, and a decision that no longer holds gets surfaced for a human, not a stochastic chatbot, to decide.
Small lessons that stick
Michael's examples of this operational learning are small. On Windows, Michael runs OpenSpec in Zsh inside WSL. Claude couldn't find it, then looked in Bash, and once told about Zsh, every later session knew. The agent also kept creating worktrees and branches; Michael, a trunk-based, continuous-delivery developer, told it the team does trunk-based development, and that went into OpenSpec's config.yaml. Michael's closing claim is that the agent got smarter and so did the team, because it flagged inconsistencies in their approaches that they hadn't known about.
Key moments
- [01:23] UberConf โ Tim's OpenSpec talk in July, and "pulling a Carducci" for speaking in every slot. frame_0003.jpg
- [04:27] The confession โ Michael had joked about spec-driven development, imagining 1990s functional specs. frame_0005.jpg
- [10:22] Two advantages โ a consistent process and consistent results, and an agent that stops starting from scratch. frame_0015.jpg
- [12:42] Nobody knows yet โ Tim calls AI-era practice a three-to-ten year problem, like agile and microservices.
- [16:15] A Lightroom replacement โ Tim uses OpenSpec to stay on track on a hobby project. frame_0017.jpg
- [19:01] Production SaaS โ Michael explores as an architect and lets the plan absorb a three-day yak shave.
- [24:06] GPS recalculates โ roadblocks and missed exits don't get the work lost.
- [24:57] What OpenSpec is โ CLI plus skills; propose or explore, apply, verify, archive. frame_0025.jpg
- [31:41] Programming as theory building โ Naur's idea that code is lossy, and OpenSpec as the record of the theory. frame_0032.jpg
- [36:05] Trunk-based development โ Zsh in WSL and no branches, learned once and remembered. frame_0040.jpg
Hook microscope (0-10s)
-
Frames: 20 at 2 fps
Read these hook frames with the Read tool:
hook_frames/frame_0001.jpg(t=0.0s)hook_frames/frame_0002.jpg(t=0.5s)hook_frames/frame_0003.jpg(t=1.0s)hook_frames/frame_0004.jpg(t=1.5s)hook_frames/frame_0005.jpg(t=2.0s)hook_frames/frame_0006.jpg(t=2.5s)hook_frames/frame_0007.jpg(t=3.0s)hook_frames/frame_0008.jpg(t=3.5s)hook_frames/frame_0009.jpg(t=4.0s)hook_frames/frame_0010.jpg(t=4.5s)hook_frames/frame_0011.jpg(t=5.0s)hook_frames/frame_0012.jpg(t=5.5s)hook_frames/frame_0013.jpg(t=6.0s)hook_frames/frame_0014.jpg(t=6.5s)hook_frames/frame_0015.jpg(t=7.0s)hook_frames/frame_0016.jpg(t=7.5s)hook_frames/frame_0017.jpg(t=8.0s)hook_frames/frame_0018.jpg(t=8.5s)hook_frames/frame_0019.jpg(t=9.0s)hook_frames/frame_0020.jpg(t=9.5s)
-
Word-level transcript (18 words):
[ 0.09s] Michael
[ 0.49s] Carducci
[ 1.05s] is
[ 1.19s] an
[ 1.33s] old
[ 1.57s] friend
[ 2.00s] and
[ 2.16s] a
[ 2.21s] multi-talented
[ 3.00s] software
[ 4.00s] engineer,
[ 5.00s] architect,
[ 5.55s] speaker,
[ 6.00s] communicator,
[ 7.11s] writer
[ 7.59s] with
[ 8.00s] magical
[ 9.07s] powers.
- 0.0-10s: One static shot: Tim, in a blue polo at a podcast mic in a home studio with a vintage Macintosh and purple lighting, talking straight to camera: "Michael Carducci is an old friend and a multi-talented software engineer, architect, speaker, communicator, writer with magical powers. I'm not kidding."
Pattern: a guest introduction with a curiosity gap (an architect "with magical powers"). The first ten seconds have no visual change; the hook is the line itself, and the cold open then promises that OpenSpec gets explained at about the 30-minute mark.
Editorial profile
- Shots: 40
- Cuts/min: 1.0
- Mean shot length: 59.91s
- Median shot length: 41.7s
- Talking-head ratio: n/a (opencv not installed)
Remote two-person conversation: a solo intro to camera, then side-by-side webcams that cut to a full-frame shot of Tim during longer passages; no slides, screen shares or B-roll, so the words carry the episode.
Quotable moments
- [11:56] "it was always felt like this new person who was a very capable coder had just joined the team and you're trying to catch it up on all of the things."
- [17:33] "I don't know that that a random walk through the feature space of what a Lightroom replacement might be is all that good of an idea."
- [24:30] "the value that I observed early on was the ability to absorb meandering."
- [30:32] "we don't build software and we have not built software for a very long time. Developers change software."
- [33:57] "it gives you a mechanism for capturing the context behind decisions, not just the code that resulted."
Entities mentioned
- People: tim-berglund, michael-carducci, Peter Naur, Venkat Subramaniam
- Companies: Oracle, Adobe, Netflix
- Tools / products: openspec, Claude, Claude Code, Fable, Opus, CLAUDE.md, AGENTS.md, Adobe Lightroom, Apple Photos, Homebrew, WSL, Zsh, Bash, UberConf (No Fluff Just Stuff)
Concepts surfaced
- spec-driven-development: changes are written as proposals, designs, tasks and formal specs before the agent writes code, and the specs stay in the repo. The term suffers from semantic diffusion, covering everything from one giant prompt to tool-driven workflows like OpenSpec.
- programming-as-theory-building: Naur's idea that the program is the team's theory of the problem, and the code is a lossy part of it; OpenSpec keeps the rest in Markdown.
- Nested change cycles: developers work by changing existing software, and the context for each change sits in earlier decisions and commitments that the code doesn't carry.
- Absorbing meandering: a written plan lets an agent detour around a roadblock and return to the route, like GPS recalculating.
- Decision records as agent context: recorded trade-offs stop the next agent from undoing a decision because it is inconvenient, and surface stale decisions to a human.
Frame timeline
What's said while each frame is on screen.
[00:00]

Michael Karduchi is an old friend and a multi-talented software engineer, architect, speaker, communicator, writer with magical powers. I'm not kidding. He is an outstanding magician. I've seen him perform many times and he never ceases to blow my mind. Today, he and I didn't talk about the dark arts, but rather OpenSpec and some experiences he had resulting from a talk I gave on OpenSpec at a conference last July. If you're new to the tool or to spec driven development, we...
[01:23]

>> There you go. link in the show notes to Dev Random. >> Indeed, >> I gave a talk at Uber Conf in July. That's uh if dear listener, if you don't know, that has nothing to do with Uber, the uh ride service. Uh that is a talk in the no fluff just stuff family of conferences takes place in Westminster every summer, Westminster, Colorado, not Westminster, London. And I gave a talk on OpenSpec uh a tool that uh supports spectrum development of in the context...
[01:59]

>> I was not. >> No, we don't usually go to each other's talks. >> We don't normally have that opportunity because at Uber I was speaking the entire conference. So >> every slot, >> every slot. Well, modulo say give or take one I think. >> Sure. Sure. Just just for the bleeding in your throat to stop. I am not that cool anymore. I had opposite end of the spectrum just that one talk. And uh there was a year I think it was 2011 where I did...
[04:41]

>> Because I I' I've done that thing that a lot of people tend to do. We hear ideas, we hear words, and we start to connect those words, these new buzzwords, these new ideas in the industry. We connect them to the concepts we're familiar with. Yeah. >> And it's very very easy to connect all those concepts together and then basically quickly parse it in your mind. Build a little mental model. Hey, it parses and then it just becomes fact in our head. I I deride...
[06:38]

you were thinking of that kind of spec." >> I Yeah. Well, yes. And I think really what what I imagined when I heard the term spectrum development, uh, and and I'm I'm not I'm not afraid to say there are a lot of things in the world that I don't understand or that I'm wrong about because what I don't know could fill a warehouse. But I had done some AI assisted coding and I'd been doing some back and forth, but I'd never really had a good repeatable, consistently useful...
[07:14]

I could see the value of sitting down and doing the kinds of things that we did. I'm of a certain age that I remember writing functional specifications and technical specifications and planning out the entire implementation before I even wrote a line of code. And that mindset sort of fit both my experience of working with AI to produce code, but also this this label of spec because like I had this label of spec in my mind and I associated this thing that I started doing...
[08:44]

way to very opinionated very uh tool-driven approaches like openspec. >> Yeah. Like the the one I was I was advocating in that talk. >> Um >> and I was kind of set off by some YouTube discourse. Um, and that that in July was the first time I'd given that talk and I I thought, you know, wait, what are people saying about this? Maybe I should go check. I think it's great. I've been using it, but what's going on in the world? And I saw uninformed arguments...
[10:09]

usually I go to my happy place during the panel. I can't hear any words. >> Um, your your team put this to practice at at work. What What was the thing that it turned out to be that you saw? What talk to me about? Oh, okay. Well, like I said, I had I had somebody
[10:22]

talk to me about? Oh, okay. Well, like I said, I had I had somebody who was working with me uh on one of my projects, and I said, um I said, I'm going to send you to Uber. He goes, he comes back, he says, I learned about spectriven development. And I said, okay, explain this to me. So, we sat down afterwards and just spent the day um pair pair programming is not the right word here, but we were pair specri developing, whatever that is. and just kind of walking down and...
[12:03]

and and it was always felt like this new person who was a very capable coder had just joined the team and you're trying to catch it up on all of the things. >> Right. Right. Um, but Spectre of Development has its own memory of the meta that it maintains, which is enormously valuable. And it does that better than what I was able to do with like a claw.md file or a bunch of markdown files or anything else because it's part of this more tool driven consistent approach....
[17:39]

through the feature space of what a Lightroom replacement might be is all that good of an idea. And so I I started with a conversation with Claude saying what I wanted to do and you know I want storage to work like this. I want import to work like this. I want to support these formats. Um I want you know face recognition and and uh semantic search of images and you know stuff that why do I not have these in Apple photos or any of these things? It's it's it's it's crazy. I...
[18:57]

allows me to go off track. >> Can I can I can I uh yang your yin for a moment? >> I I may know what you mean, but I I want to I want to hear. So, uh, I use OpenSpec in a slightly less hobbyish context. I built and maintain a SAS, >> which is good that there's a >> a responsible professional in the conversation here, at least one. >> I have I have customers, I have SLAs's um, you know, and paying customers, the rest of it. So, >> you know,...
[21:46]

everything else so it understands oh okay yep I understand what you're talking about semantic data or modular monolith or the rest of it and why and what the constraints are all of those things have been emerging beautifully and what what comes out the other end is is the change and so that's this that's the This didn't feel that different to what I'd already been doing. And I didn't realize how rigorous that approach had been because it had baked in as many surprises as...
[22:57]

and that turned into a threeday yak shaving extravaganza >> as it always does >> because of I'm going to hold my tongue because I have so many unkind things to say about Oracle right now. >> Oh boy. >> Um but it's mostly Oracle's fault. >> Hey, we're not we're not sponsored by Oracle here on this occasional podcast. So, you know, you say anything you want, >> but it it it just shipped a bunch of buggy drivers and then and then did nothing...
[23:50]

TomTom or whatever turnbyturn directions that that you use. >> TomTom. Okay. Boomer. >> Yeah. >> I just want to be inclusive. >> Please go on. >> Don't shame me for that. [laughter] >> There are dozens of them. Dozens. >> Dozens. So many. Yeah. But whatever you're using for your turn byturn directions, GPS navigation, sometimes you will hit a roadblock that isn't in the map that wasn't in the plan or sometimes you just make a...
[25:28]

we're doing. So it is fundamentally a command line tool. >> Um and a set of skills, very particular set of skills. Um and when you you know say on a Mac you'll brew install openspec and uh uh you have a new project you'll you'll install the skills into the dotcloud or agent or you know wherever they go in your particular coding harness. Um and it's got support for a couple dozen of those. Um then there's this workflow. Okay. So you got those skills, you got that...
[28:03]

language. That's all in markdown and on a perch change basis any >> but it does have shades of BDD that become not like an executable spec per se but like you can see it from there. >> You can you can get I'll get there. There's >> the specs don't just go into a folder to die and never be heard from again. um when you know development of something complex, new changes are going to interact with old specs and and and we'll get there. So um once you got a...
[28:54]

verification the agent can do. Some of them might be manual verification. Uh there's a new version of OpenSpec that has a verify step um a verify command that you can get a different model to go in and and and check things out. Actually, I'll confess I haven't done that yet. But um when you're done, everything works. Then you archive the change and it is no longer an active change. But the the design document, the proposal, the task list and the specs stay in the open spec...
[29:37]

all the modern harnesses and the the the contemporary models contemporary as of this recording do a really good job saying okay what's out there in the codebase what am I doing what do I need to look at and you know maybe if there's a little markdown file of whether it's a agent MD or claw MD or some other sort of advice you know they'll find those things but now explicitly there's this long list of specs and the specs are
[29:57]

things but now explicitly there's this long list of specs and the specs are merged into the master spec for the application. And this isn't part of the problem with writing functional specs in the bad old days was it was a focus on the software, the thing you were building and not the people who used it and what they were doing. >> Um, and it was very easy to just get absorbed and off the off into weird territory and and not thinking about the people who use your...
[31:51]

don't know how to pronounce his last name. I've only ever read it. Do you know who I'm talking about? >> Yes. uh he wrote a paper back I think in the 80s called uh uh it was like programming as theory building. So he has this concept of uh what he calls a theory but um essentially there is there is the program and the program is more than the code. The code is like a raification of the part of the program that happens in the machine. But there is there is something...
[32:24]

bigger that that exists outside of the machine because the reality is the software that we build exists to deliver value and where it delivers value ultimately is outside of the computer. The the problem is outside the solution is inside. And so by definition just looking at the code you're only looking at part of the program part of the part of the solution or part of the problem. He says that code is uh is lossy. And what I found is is continuing to use open spec um as a...
[33:09]

to expand not only the code but the the program as theory. It's it's the >> it's the theory building part because >> what it's maintaining in those markdown files and the opinionated structure that it adopts is basically the kind of learning that happens in a team. Like I remember one of my jobs as a coder I show up and they say, "Hey, Michael, go do this thing." And I I'm like, "Oh, okay. Yeah, we need to do this. I know how to do that." D. And all of a sudden,...
[36:18]

very small way because I'm on Windows for better or for worse. Um, and I know that puts me in the minority because your first instinct was to send the link to this podcast um, studio online to my phone. >> I I in fact texted it to you. Yeah. Not not helpful. >> I did. >> Yeah. >> Uh, I mean that works for you and you know, Slutton does the same thing and a bunch of, you know, all all of my Mac friends are like, "Wait, you're not on an Apple device."...
Transcript
Source: captions.
[00:03] Michael Karduchi is an old friend and a multi-talented software engineer,
[00:05] multi-talented software engineer, architect, speaker, communicator, writer
[00:08] architect, speaker, communicator, writer with magical powers. I'm not kidding. He
[00:11] with magical powers. I'm not kidding. He is an outstanding magician. I've seen
[00:12] is an outstanding magician. I've seen him perform many times and he never
[00:14] him perform many times and he never ceases to blow my mind. Today, he and I
[00:17] ceases to blow my mind. Today, he and I didn't talk about the dark arts, but
[00:19] didn't talk about the dark arts, but rather OpenSpec and some experiences he
[00:21] rather OpenSpec and some experiences he had resulting from a talk I gave on
[00:23] had resulting from a talk I gave on OpenSpec at a conference last July. If
[00:26] OpenSpec at a conference last July. If you're new to the tool or to spec driven
[00:28] you're new to the tool or to spec driven development, we don't actually describe
[00:30] development, we don't actually describe what it is and how it works until around
[00:31] what it is and how it works until around the 30 minute mark. Takes us a little
[00:33] the 30 minute mark. Takes us a little bit, but you know what? It's a fun
[00:34] bit, but you know what? It's a fun conversation like it always is with
[00:36] conversation like it always is with Michael. Let's listen in. Mr. Karduchi,
[00:40] Michael. Let's listen in. Mr. Karduchi, it is good to see you.
[00:42] it is good to see you. >> Tim, it has been too long. How the heck
[00:45] >> Tim, it has been too long. How the heck are you?
[00:46] are you? >> I am doing okay. I really am having a
[00:49] >> I am doing okay. I really am having a good time. Um it is uh I don't mind
[00:52] good time. Um it is uh I don't mind saying September the 24th as we record
[00:54] saying September the 24th as we record this 2026. Last time we saw each other
[00:57] this 2026. Last time we saw each other was like two months ago plus a little
[00:59] was like two months ago plus a little bit. That is I agree too long.
[01:01] bit. That is I agree too long. >> And um it's good to see you. Haven't
[01:03] >> And um it's good to see you. Haven't seen this this setup. This background
[01:05] seen this this setup. This background looks great. Little spooky red light
[01:07] looks great. Little spooky red light back there. I feel I feel like I should
[01:10] back there. I feel I feel like I should I should just pay heed uh to you.
[01:14] I should just pay heed uh to you. >> Well, we are in the Dev Random TV
[01:16] >> Well, we are in the Dev Random TV studio.
[01:18] studio. >> There you go. link in the show notes to
[01:22] >> There you go. link in the show notes to Dev Random.
[01:23] Dev Random. >> Indeed,
[01:24] >> Indeed, >> I gave a talk at Uber Conf in July.
[01:27] >> I gave a talk at Uber Conf in July. That's uh if dear listener, if you don't
[01:29] That's uh if dear listener, if you don't know, that has nothing to do with Uber,
[01:32] know, that has nothing to do with Uber, the uh ride service. Uh that is a talk
[01:36] the uh ride service. Uh that is a talk in the no fluff just stuff family of
[01:37] in the no fluff just stuff family of conferences takes place in Westminster
[01:39] conferences takes place in Westminster every summer, Westminster, Colorado, not
[01:42] every summer, Westminster, Colorado, not Westminster, London. And I gave a talk
[01:45] Westminster, London. And I gave a talk on OpenSpec uh a tool that uh supports
[01:48] on OpenSpec uh a tool that uh supports spectrum development of in the context
[01:52] spectrum development of in the context of aentic engineering.
[01:53] of aentic engineering. >> And you were not there which is okay.
[01:56] >> And you were not there which is okay. >> I was not.
[01:57] >> I was not. >> No, we don't usually go to each other's
[01:59] >> No, we don't usually go to each other's talks.
[02:00] talks. >> We don't normally have that opportunity
[02:03] >> We don't normally have that opportunity because at Uber I was speaking the
[02:07] because at Uber I was speaking the entire conference. So
[02:09] entire conference. So >> every slot,
[02:10] >> every slot, >> every slot. Well, modulo say give or
[02:13] >> every slot. Well, modulo say give or take one I think.
[02:14] take one I think. >> Sure. Sure. Just just for the bleeding
[02:16] >> Sure. Sure. Just just for the bleeding in your throat to stop. I am not that
[02:19] in your throat to stop. I am not that cool anymore. I had opposite end of the
[02:21] cool anymore. I had opposite end of the spectrum just that one talk. And uh
[02:25] spectrum just that one talk. And uh there was a year I think it was 2011
[02:27] there was a year I think it was 2011 where I did I didn't do every slot. I
[02:29] where I did I didn't do every slot. I did all six on Wednesday. There's again
[02:32] did all six on Wednesday. There's again folks uh if you don't know the
[02:33] folks uh if you don't know the conference there 90-minute sessions and
[02:35] conference there 90-minute sessions and on the first day of the conference after
[02:37] on the first day of the conference after the the workshops the first proper day
[02:39] the the workshops the first proper day there are six sessions that go from like
[02:40] there are six sessions that go from like 8:30 to 10:30 at night. It's insane. I
[02:42] 8:30 to 10:30 at night. It's insane. I did all six of those once and you're
[02:44] did all six of those once and you're just like we used to call this pulling a
[02:46] just like we used to call this pulling a venat but it's now pulling a Karduchi uh
[02:49] venat but it's now pulling a Karduchi uh filling every slot.
[02:50] filling every slot. >> Oh my I'm uh
[02:55] >> Oh my I'm uh I'm about to become
[02:57] I'm about to become whatever part of speech that is. My
[02:59] whatever part of speech that is. My brain is too tired today to figure that
[03:01] brain is too tired today to figure that out because
[03:03] out because >> a verb. No, it's still a noun, but I'm
[03:05] >> a verb. No, it's still a noun, but I'm already a noun. You know what I mean?
[03:06] already a noun. You know what I mean? >> Yeah, but I've always said that I'm
[03:09] >> Yeah, but I've always said that I'm >> Yeah,
[03:09] >> Yeah, >> like Xerox. I feel like Xerox.
[03:12] >> like Xerox. I feel like Xerox. >> Yeah, there you go. Xerox is a verb and
[03:13] >> Yeah, there you go. Xerox is a verb and a noun. But I've always said that I'm
[03:15] a noun. But I've always said that I'm more of a verb than a noun. You know,
[03:17] more of a verb than a noun. You know, people ask me uh what what what do you
[03:19] people ask me uh what what what do you do for a living? And I don't know. I do
[03:23] do for a living? And I don't know. I do fun and interesting things and
[03:24] fun and interesting things and occasionally people pay me. Um the only
[03:27] occasionally people pay me. Um the only which is a great way to live your life
[03:29] which is a great way to live your life except for the occasionally part but you
[03:32] except for the occasionally part but you know we learn to adapt.
[03:34] know we learn to adapt. >> There's you know some some ups and downs
[03:37] >> There's you know some some ups and downs and and things like that but you do get
[03:39] and and things like that but you do get to do fun and interesting things and do
[03:41] to do fun and interesting things and do things like give 15 talks at a
[03:43] things like give 15 talks at a conference 15 90 minute talks which uh
[03:46] conference 15 90 minute talks which uh >> yeah that's fantastic.
[03:48] >> yeah that's fantastic. >> It's more like type two fun than it is
[03:51] >> It's more like type two fun than it is type one fun.
[03:52] type one fun. >> It is. And folks, if you don't know what
[03:54] >> It is. And folks, if you don't know what type two fun is, that's fun that's fun
[03:56] type two fun is, that's fun that's fun after the fact. Like, uh,
[03:57] after the fact. Like, uh, >> but not certainly not in the middle.
[03:59] >> but not certainly not in the middle. >> Not during. And I'll usually describe
[04:01] >> Not during. And I'll usually describe difficult outdoor experiences like
[04:03] difficult outdoor experiences like getting caught in bad weather as type
[04:04] getting caught in bad weather as type two fun cuz it's miserable at the time.
[04:07] two fun cuz it's miserable at the time. Um, and afterward, like 30 minutes after
[04:09] Um, and afterward, like 30 minutes after you get warm, you have become the most
[04:12] you get warm, you have become the most awesome human being ever to exist.
[04:16] awesome human being ever to exist. >> And just listen to my story. I'm going
[04:18] >> And just listen to my story. I'm going to tell you how awesome I am. It becomes
[04:19] to tell you how awesome I am. It becomes the best thing ever. So yeah. Yeah, that
[04:21] the best thing ever. So yeah. Yeah, that that kind of that kind of load at a
[04:23] that kind of that kind of load at a conference can be uh that kind of thing.
[04:26] conference can be uh that kind of thing. So,
[04:27] So, >> but you gave this talk. I want to
[04:29] >> but you gave this talk. I want to preface this though because before we
[04:31] preface this though because before we get to this talk,
[04:32] get to this talk, >> I want to get to you calling me out
[04:34] >> I want to get to you calling me out whether you knew it or not. Okay?
[04:36] whether you knew it or not. Okay? >> Because I I' I've done that thing that a
[04:40] >> Because I I' I've done that thing that a lot of people tend to do. We hear ideas,
[04:42] lot of people tend to do. We hear ideas, we hear words, and we start to connect
[04:45] we hear words, and we start to connect those words, these new buzzwords, these
[04:47] those words, these new buzzwords, these new ideas in the industry. We connect
[04:49] new ideas in the industry. We connect them to the concepts we're familiar
[04:51] them to the concepts we're familiar with. Yeah.
[04:52] with. Yeah. >> And it's very very easy to connect all
[04:53] >> And it's very very easy to connect all those concepts together and then
[04:55] those concepts together and then basically quickly parse it in your mind.
[04:57] basically quickly parse it in your mind. Build a little mental model. Hey, it
[04:59] Build a little mental model. Hey, it parses and then it just becomes fact in
[05:02] parses and then it just becomes fact in our head. I I deride that behavior, but
[05:06] our head. I I deride that behavior, but I am human and I also do it and I'm
[05:08] I am human and I also do it and I'm trying to get better. I had heard this
[05:11] trying to get better. I had heard this talk of spec driven development and I
[05:15] talk of spec driven development and I let my own assumptions define it
[05:17] let my own assumptions define it independently of everything else. I'd
[05:20] independently of everything else. I'd even cracked a joke or two at the
[05:22] even cracked a joke or two at the expense of spec driven development
[05:25] expense of spec driven development >> in the context of a larger point that I
[05:27] >> in the context of a larger point that I was trying to make in a panel or in a
[05:29] was trying to make in a panel or in a talk or something like that. And then
[05:31] talk or something like that. And then you posted on LinkedIn. You said to the
[05:35] you posted on LinkedIn. You said to the people who are making these kinds of
[05:37] people who are making these kinds of jokes, first of all, scene.
[05:40] jokes, first of all, scene. Second of all, you followed it with
[05:42] Second of all, you followed it with >> Get wrecked, whoever you are. No, I
[05:44] >> Get wrecked, whoever you are. No, I don't think I said that. But
[05:45] don't think I said that. But >> no, I think you said, "Do you even lift,
[05:48] >> no, I think you said, "Do you even lift, bro?" was your exact
[05:50] bro?" was your exact >> That sounds like me.
[05:52] >> That sounds like me. >> Yeah.
[05:53] >> Yeah. >> And And so I paused and I just sort of
[05:56] >> And And so I paused and I just sort of accepted that maybe I don't know what I
[05:59] accepted that maybe I don't know what I think this thing is. Yeah.
[06:01] think this thing is. Yeah. >> And that was where it was resting. Now,
[06:05] >> And that was where it was resting. Now, I sent a colleague of mine to Uber Comp
[06:07] I sent a colleague of mine to Uber Comp and he went to your session
[06:11] and he went to your session and then he taught me what he learned
[06:14] and then he taught me what he learned from you and that's when everything
[06:16] from you and that's when everything changed.
[06:16] changed. >> Okay. So, what the narrative
[06:19] >> Okay. So, what the narrative >> did did you think Tim is back to the '9s
[06:22] >> did did you think Tim is back to the '9s and functional specifications and I was
[06:25] and functional specifications and I was working for IBM at the time just a few
[06:26] working for IBM at the time just a few months ago. You're thinking, "Wow, IBM
[06:28] months ago. You're thinking, "Wow, IBM has really had its impact on him. He
[06:30] has really had its impact on him. He used to be something and and now look, I
[06:32] used to be something and and now look, I mean, was it I'm overelling it, but you
[06:34] mean, was it I'm overelling it, but you you were thinking of that kind of spec."
[06:37] you were thinking of that kind of spec." >> I
[06:39] >> I Yeah. Well, yes. And I think really what
[06:44] Yeah. Well, yes. And I think really what what I imagined when I heard the term
[06:46] what I imagined when I heard the term spectrum development, uh, and and I'm
[06:49] spectrum development, uh, and and I'm I'm not I'm not afraid to say there are
[06:50] I'm not I'm not afraid to say there are a lot of things in the world that I
[06:51] a lot of things in the world that I don't understand or that I'm wrong about
[06:53] don't understand or that I'm wrong about because what I don't know could fill a
[06:54] because what I don't know could fill a warehouse. But I had done some AI
[06:58] warehouse. But I had done some AI assisted coding and I'd been doing some
[07:00] assisted coding and I'd been doing some back and forth, but I'd never really had
[07:02] back and forth, but I'd never really had a good repeatable, consistently useful
[07:07] a good repeatable, consistently useful approach and
[07:10] approach and I could see the value of sitting down
[07:12] I could see the value of sitting down and doing the kinds of things that we
[07:14] and doing the kinds of things that we did. I'm of a certain age that I
[07:17] did. I'm of a certain age that I remember writing functional
[07:18] remember writing functional specifications and technical
[07:20] specifications and technical specifications and planning out the
[07:22] specifications and planning out the entire implementation before I even
[07:24] entire implementation before I even wrote a line of code. And that mindset
[07:28] wrote a line of code. And that mindset sort of fit both my experience of
[07:32] sort of fit both my experience of working with AI to produce code, but
[07:36] working with AI to produce code, but also
[07:37] also this this label of spec because like I
[07:42] this this label of spec because like I had this label of spec in my mind and I
[07:44] had this label of spec in my mind and I associated this thing that I started
[07:46] associated this thing that I started doing back in the 1990s or we'll just
[07:49] doing back in the 1990s or we'll just say the 1900s because it might as well
[07:51] say the 1900s because it might as well be that far away. It is the last
[07:53] be that far away. It is the last century. And uh that was just sort of
[07:56] century. And uh that was just sort of where I landed. So I was certainly
[07:59] where I landed. So I was certainly associating this idea of specri
[08:03] associating this idea of specri developments as these kinds of
[08:05] developments as these kinds of specifications. I wasn't really hung up
[08:07] specifications. I wasn't really hung up on granularity,
[08:08] on granularity, but uh but it it it made intuitive sense
[08:11] but uh but it it it made intuitive sense to me because I figured writing more and
[08:15] to me because I figured writing more and more about the implementation in a
[08:18] more about the implementation in a specification that is essentially a
[08:21] specification that is essentially a giant prompt was uh what people were
[08:24] giant prompt was uh what people were talking about. Now there is a semantic
[08:26] talking about. Now there is a semantic diffusion problem with spectriven
[08:28] diffusion problem with spectriven development. I'm finding out because
[08:30] development. I'm finding out because people call a lot of things specri
[08:32] people call a lot of things specri development and some people are even
[08:35] development and some people are even doing what I was imagining and they're
[08:37] doing what I was imagining and they're calling that specri development all the
[08:39] calling that specri development all the way to very opinionated very uh
[08:42] way to very opinionated very uh tool-driven approaches like openspec.
[08:45] tool-driven approaches like openspec. >> Yeah. Like the the one I was I was
[08:47] >> Yeah. Like the the one I was I was advocating in that talk.
[08:49] advocating in that talk. >> Um
[08:51] >> Um >> and I was kind of set off by some
[08:54] >> and I was kind of set off by some YouTube discourse. Um, and that that in
[08:58] YouTube discourse. Um, and that that in July was the first time I'd given that
[08:59] July was the first time I'd given that talk and I I thought, you know, wait,
[09:01] talk and I I thought, you know, wait, what are people saying about this? Maybe
[09:03] what are people saying about this? Maybe I should go check. I think it's great.
[09:05] I should go check. I think it's great. I've been using it, but what's going on
[09:06] I've been using it, but what's going on in the world? And I saw uninformed
[09:11] in the world? And I saw uninformed arguments against it that that um and
[09:13] arguments against it that that um and and by the way, Michael, you uh called
[09:16] and by the way, Michael, you uh called yourself out because you heard a thing
[09:18] yourself out because you heard a thing and you interpreted it in terms of the
[09:20] and you interpreted it in terms of the mental scaffold scaffolding that you
[09:22] mental scaffold scaffolding that you had. um that's you don't have a choice.
[09:25] had. um that's you don't have a choice. There are too many individual things in
[09:26] There are too many individual things in the world. You you you have plausibility
[09:29] the world. You you you have plausibility structures and sort of a skeleton for
[09:31] structures and sort of a skeleton for knowledge and you get a new thing in
[09:33] knowledge and you get a new thing in there and you you got to put it in the
[09:35] there and you you got to put it in the skeleton somewhere until something comes
[09:38] skeleton somewhere until something comes along to disrupt that and you think,
[09:39] along to disrupt that and you think, "Oh, I need to rebuild some frameworks."
[09:41] "Oh, I need to rebuild some frameworks." There's nothing wrong with that, but
[09:43] There's nothing wrong with that, but these were guys going to the trouble of
[09:45] these were guys going to the trouble of making a whole YouTube video saying,
[09:47] making a whole YouTube video saying, "Did agile teach us nothing? Why are we
[09:49] "Did agile teach us nothing? Why are we writing specs again?" And I thought,
[09:50] writing specs again?" And I thought, "Okay, that just sounds nothing like the
[09:53] "Okay, that just sounds nothing like the actual practice here." So
[09:55] actual practice here." So >> Oh, wow. Okay. So, you weren't calling
[09:57] >> Oh, wow. Okay. So, you weren't calling me out because I I made one or two.
[10:01] me out because I I made one or two. >> No, I not that you were talking trash
[10:03] >> No, I not that you were talking trash about me. If you said it on a panel,
[10:05] about me. If you said it on a panel, usually I go to my happy place during
[10:06] usually I go to my happy place during the panel. I can't hear any words.
[10:09] the panel. I can't hear any words. >> Um, your your team put this to practice
[10:12] >> Um, your your team put this to practice at at work. What What was the thing that
[10:15] at at work. What What was the thing that it turned out to be that you saw? What
[10:18] it turned out to be that you saw? What talk to me about? Oh, okay.
[10:22] talk to me about? Oh, okay. Well, like I said, I had I had somebody
[10:24] Well, like I said, I had I had somebody who was working with me uh on one of my
[10:26] who was working with me uh on one of my projects, and I said, um I said, I'm
[10:28] projects, and I said, um I said, I'm going to send you to Uber. He goes, he
[10:30] going to send you to Uber. He goes, he comes back, he says, I learned about
[10:31] comes back, he says, I learned about spectriven development. And I said,
[10:34] spectriven development. And I said, okay, explain this to me. So, we sat
[10:36] okay, explain this to me. So, we sat down afterwards and just spent the day
[10:39] down afterwards and just spent the day um pair pair programming is not the
[10:43] um pair pair programming is not the right word here, but we were pair specri
[10:46] right word here, but we were pair specri developing, whatever that is. and just
[10:49] developing, whatever that is. and just kind of walking down and playing with
[10:51] kind of walking down and playing with it. And what I discovered is
[10:55] it. And what I discovered is this is
[10:57] this is an extremely valuable workflow that may
[11:00] an extremely valuable workflow that may not be the optimum workflow, but it is
[11:04] not be the optimum workflow, but it is the that has two advantages over
[11:06] the that has two advantages over anything else that I had personally
[11:07] anything else that I had personally done. Number one is I started it. it
[11:12] done. Number one is I started it. it gave me a very consistent repeatable
[11:15] gave me a very consistent repeatable process that I could follow and
[11:20] process that I could follow and two
[11:21] two it gave me consistent repeatable
[11:25] it gave me consistent repeatable results. Um and the side effect is my
[11:30] results. Um and the side effect is my experience with the agent kept getting
[11:33] experience with the agent kept getting better. So whereas every time I would
[11:36] better. So whereas every time I would sit down with Claude, it was like we
[11:38] sit down with Claude, it was like we were having the same conversations and I
[11:41] were having the same conversations and I was giving it the same reminders and oh
[11:44] was giving it the same reminders and oh no, no, we do these things this way. And
[11:45] no, no, we do these things this way. And yeah, I'd been going in there and
[11:47] yeah, I'd been going in there and editing things in my claw.md file and
[11:50] editing things in my claw.md file and doing other bits and pieces, but it just
[11:54] doing other bits and pieces, but it just always felt like we were from square one
[11:57] always felt like we were from square one and and it was always felt like this
[12:01] and and it was always felt like this new person who was a very capable coder
[12:03] new person who was a very capable coder had just joined the team and you're
[12:04] had just joined the team and you're trying to catch it up on all of the
[12:07] trying to catch it up on all of the things.
[12:07] things. >> Right. Right. Um, but Spectre of
[12:10] >> Right. Right. Um, but Spectre of Development
[12:12] Development has
[12:14] has its own memory
[12:16] its own memory of the meta that it maintains, which is
[12:19] of the meta that it maintains, which is enormously valuable. And it does that
[12:22] enormously valuable. And it does that better than what I was able to do with
[12:24] better than what I was able to do with like a claw.md file or a bunch of
[12:26] like a claw.md file or a bunch of markdown files or anything else because
[12:28] markdown files or anything else because it's part of this more tool driven
[12:32] it's part of this more tool driven consistent approach.
[12:34] consistent approach. And since then, I haven't gone anywhere
[12:37] And since then, I haven't gone anywhere without it. It's like my American
[12:40] without it. It's like my American Express.
[12:41] Express. >> Nice.
[12:43] >> Nice. Before we go on, I want to say to anyone
[12:47] Before we go on, I want to say to anyone listening who hates spectrum development
[12:48] listening who hates spectrum development or openspec or whatever thinks, you
[12:51] or openspec or whatever thinks, you know, if you don't have agent loops
[12:52] know, if you don't have agent loops running, then you're missing the G. What
[12:54] running, then you're missing the G. What you know, there's a lot of ideas. Um, we
[12:56] you know, there's a lot of ideas. Um, we don't the practice of writing software
[12:59] don't the practice of writing software has just radically changed for most of
[13:01] has just radically changed for most of us as of about nine or 10 months ago.
[13:04] us as of about nine or 10 months ago. Um,
[13:04] Um, >> and it's still radically changing ways
[13:07] >> and it's still radically changing ways >> and we don't really know what we're
[13:09] >> and we don't really know what we're doing now. I I think I'm getting great
[13:11] doing now. I I think I'm getting great results. A lot of people are getting
[13:12] results. A lot of people are getting fantastic results. Um, but it's kind of
[13:16] fantastic results. Um, but it's kind of a three to five maybe five to 10 year
[13:18] a three to five maybe five to 10 year kind of problem to figure out what it is
[13:22] kind of problem to figure out what it is we're doing. again
[13:24] we're doing. again uh it would be like the kinds the the
[13:27] uh it would be like the kinds the the parts of agile that sort of filtered
[13:29] parts of agile that sort of filtered through all the consultancies and
[13:32] through all the consultancies and processes and all that stuff to just
[13:34] processes and all that stuff to just become accepted practice in the world
[13:37] become accepted practice in the world that took 10 15 years uh people to
[13:40] that took 10 15 years uh people to figure out okay what's test-driven
[13:41] figure out okay what's test-driven development really for an architectural
[13:43] development really for an architectural paradigm like microservices we just
[13:45] paradigm like microservices we just sucked at those for three to five years
[13:47] sucked at those for three to five years I'm amazed the paradigm even survived
[13:50] I'm amazed the paradigm even survived the initial implementations before
[13:52] the initial implementations before people really figured out how to do
[13:53] people really figured out how to do that. So,
[13:53] that. So, >> oh, I I have a theory. Two things. Uh,
[13:56] >> oh, I I have a theory. Two things. Uh, if I if I can just drop a little seed in
[13:59] if I if I can just drop a little seed in there.
[13:59] there. >> We're here to riff.
[14:00] >> We're here to riff. >> I don't I don't think most people have
[14:02] >> I don't I don't think most people have ever figured out microservices. I don't
[14:04] ever figured out microservices. I don't think teams have ever really figured out
[14:06] think teams have ever really figured out agile. I think the interesting thing is
[14:08] agile. I think the interesting thing is is that we've always had humans in the
[14:10] is that we've always had humans in the loop doing what the humans always do.
[14:13] loop doing what the humans always do. And um and where I think the entire
[14:16] And um and where I think the entire calculus really changes is now we're
[14:19] calculus really changes is now we're taking humans out of the loop. And I
[14:22] taking humans out of the loop. And I don't think we can really afford to not
[14:25] don't think we can really afford to not understand these broader ideas.
[14:30] understand these broader ideas. >> Yeah. But no, I I I've been an
[14:32] >> Yeah. But no, I I I've been an architecture consultant
[14:34] architecture consultant >> for 15 years and uh at least
[14:38] >> for 15 years and uh at least >> and um I've never seen anybody like
[14:41] >> and um I've never seen anybody like personally not yeah we can case study
[14:44] personally not yeah we can case study this company you know Netflix or
[14:46] this company you know Netflix or somebody like that but personally I've
[14:47] somebody like that but personally I've never seen anybody doing microservices
[14:50] never seen anybody doing microservices well actually succeeding with
[14:54] well actually succeeding with microservices they are surviving with
[14:56] microservices they are surviving with whatever they've built. Fair enough.
[14:58] whatever they've built. Fair enough. Fair enough. And and I I threw a bunch
[15:00] Fair enough. And and I I threw a bunch of things in there that are absolute um
[15:04] of things in there that are absolute um uh catnip for various debates. And I
[15:06] uh catnip for various debates. And I don't I don't mean to
[15:07] don't I don't mean to >> if you're listening and you think I just
[15:09] >> if you're listening and you think I just said something deeply foolish, just just
[15:11] said something deeply foolish, just just give me a minute here.
[15:13] give me a minute here. >> Um
[15:13] >> Um >> let him cook.
[15:14] >> let him cook. >> Let me let me cook. Um
[15:18] >> Let me let me cook. Um so much has been upended in the way we
[15:19] so much has been upended in the way we do what we do the way we create systems
[15:22] do what we do the way we create systems out of software that uh we are currently
[15:25] out of software that uh we are currently casting about feeling around in the dark
[15:27] casting about feeling around in the dark trying to figure out smart ways to do
[15:29] trying to figure out smart ways to do things and just like preAI there wasn't
[15:32] things and just like preAI there wasn't one methodology there wasn't one
[15:33] one methodology there wasn't one architecture there was this robust
[15:35] architecture there was this robust debate all these it's going to it's
[15:37] debate all these it's going to it's going to stay that way okay so uh we're
[15:40] going to stay that way okay so uh we're talking about spectrum development
[15:41] talking about spectrum development particularly openspec here because I
[15:42] particularly openspec here because I think it's interesting because I like it
[15:44] think it's interesting because I like it you have found some some luck with that
[15:46] you have found some some luck with that uh there are going to be other AI
[15:48] uh there are going to be other AI assisted engineering
[15:50] assisted engineering workflows, practices, methodologies
[15:54] workflows, practices, methodologies that will just never have anything to do
[15:56] that will just never have anything to do with a tool like this. And uh big tent,
[16:00] with a tool like this. And uh big tent, lots of room for lots of approaches. Um
[16:03] lots of room for lots of approaches. Um and you know, I wouldn't mind uh finding
[16:05] and you know, I wouldn't mind uh finding somebody who hates OpenSpec in a
[16:07] somebody who hates OpenSpec in a principled way to to talk in the future,
[16:09] principled way to to talk in the future, but you and I love it. So, um, and it's
[16:12] but you and I love it. So, um, and it's less about the tool than the process.
[16:13] less about the tool than the process. But
[16:15] But what I've found, let me let me just talk
[16:19] what I've found, let me let me just talk a little bit about what my experience
[16:20] a little bit about what my experience has been. And I have to be clear that
[16:23] has been. And I have to be clear that this is with pretty hobbyish projects.
[16:26] this is with pretty hobbyish projects. Uh, my work as a developer relations
[16:30] Uh, my work as a developer relations leader. Um, I I do a lot of Zoom calls
[16:33] leader. Um, I I do a lot of Zoom calls and Slack messages and occasionally
[16:35] and Slack messages and occasionally travel places and give talks. Um, but
[16:38] travel places and give talks. Um, but I'm managing people a lot. I'm not
[16:39] I'm managing people a lot. I'm not writing code at work. Um I'm, you know,
[16:43] writing code at work. Um I'm, you know, leading teams that that that write code,
[16:45] leading teams that that that write code, but I'm not. And so this is hobby stuff.
[16:47] but I'm not. And so this is hobby stuff. And so, you know, take take with a grain
[16:48] And so, you know, take take with a grain of salt.
[16:49] of salt. >> Um
[16:51] >> Um I find you you mentioned it keeps the
[16:54] I find you you mentioned it keeps the agent on track. And I'd like to dig into
[16:56] agent on track. And I'd like to dig into a little bit about how, but I find it
[16:58] a little bit about how, but I find it keeps me on track is the thing. I am
[17:02] keeps me on track is the thing. I am working right now in this month of
[17:04] working right now in this month of September on
[17:06] September on um something to replace Adobe Lightroom
[17:10] um something to replace Adobe Lightroom for myself. Um I don't know how broadly
[17:12] for myself. Um I don't know how broadly useful it'll be. There's a lot of really
[17:14] useful it'll be. There's a lot of really hard problems in there, but uh I'm just
[17:16] hard problems in there, but uh I'm just sort of tired of their BS and I think I
[17:19] sort of tired of their BS and I think I could I could do better uh now that
[17:21] could I could do better uh now that we've got AI
[17:23] we've got AI >> and
[17:25] >> and I kind of need to be kept on track about
[17:27] I kind of need to be kept on track about what I'm working on when it's when it's
[17:29] what I'm working on when it's when it's as easy as prompting a new feature. Uh I
[17:33] as easy as prompting a new feature. Uh I don't know that that a random walk
[17:35] don't know that that a random walk through the feature space of what a
[17:38] through the feature space of what a Lightroom replacement might be is all
[17:39] Lightroom replacement might be is all that good of an idea. And so I I started
[17:44] that good of an idea. And so I I started with a conversation with Claude saying
[17:45] with a conversation with Claude saying what I wanted to do and you know I want
[17:48] what I wanted to do and you know I want storage to work like this. I want import
[17:50] storage to work like this. I want import to work like this. I want to support
[17:51] to work like this. I want to support these formats. Um I want you know face
[17:55] these formats. Um I want you know face recognition and and uh semantic search
[17:58] recognition and and uh semantic search of images and you know stuff that why do
[18:01] of images and you know stuff that why do I not have these in Apple photos or any
[18:03] I not have these in Apple photos or any of these things? It's it's it's it's
[18:05] of these things? It's it's it's it's crazy. I would just like these things.
[18:07] crazy. I would just like these things. And um we had this this long
[18:10] And um we had this this long conversation and it gave me a markdown
[18:13] conversation and it gave me a markdown file of
[18:15] file of some
[18:17] some proposals and explorations. I'm using
[18:19] proposals and explorations. I'm using particular verbs related to OpenSpec.
[18:21] particular verbs related to OpenSpec. Maybe we can get into them. Um and now
[18:23] Maybe we can get into them. Um and now I'm I'm just sort of walking through
[18:25] I'm I'm just sort of walking through that and and implementing there. Not
[18:28] that and and implementing there. Not just saying, "Oh, hey, that's cool, but
[18:31] just saying, "Oh, hey, that's cool, but let's do this." You know, there there
[18:32] let's do this." You know, there there would be a little bit more of a
[18:34] would be a little bit more of a meandering in my own
[18:37] meandering in my own focused deprived mind through the
[18:40] focused deprived mind through the process because I can just
[18:42] process because I can just >> prompt stuff. So I feel like it keeps me
[18:44] >> prompt stuff. So I feel like it keeps me on track as much as as anything.
[18:46] on track as much as as anything. >> Now that's really interesting because my
[18:50] >> Now that's really interesting because my one of one of my observations was it
[18:53] one of one of my observations was it allows me to go off track.
[18:56] allows me to go off track. >> Can I can I can I uh yang your yin for a
[18:59] >> Can I can I can I uh yang your yin for a moment?
[19:00] moment? >> I I may know what you mean, but I I want
[19:01] >> I I may know what you mean, but I I want to I want to hear.
[19:04] to I want to hear. So, uh, I use OpenSpec in a slightly
[19:08] So, uh, I use OpenSpec in a slightly less hobbyish context. I built and
[19:11] less hobbyish context. I built and maintain a SAS,
[19:12] maintain a SAS, >> which is good that there's a
[19:14] >> which is good that there's a >> a responsible professional in the
[19:16] >> a responsible professional in the conversation here, at least one.
[19:18] conversation here, at least one. >> I have I have customers, I have SLAs's
[19:21] >> I have I have customers, I have SLAs's um, you know, and paying customers, the
[19:23] um, you know, and paying customers, the rest of it. So,
[19:24] rest of it. So, >> you know, I have to be somewhat careful.
[19:27] >> you know, I have to be somewhat careful. Now, uh,
[19:30] Now, uh, where I have been is I'm I'm a
[19:35] where I have been is I'm I'm a classically trained software engineer,
[19:37] classically trained software engineer, right? Well, I'm not classically
[19:38] right? Well, I'm not classically trained, but whatever
[19:40] trained, but whatever whatever that whatever you want that to
[19:42] whatever that whatever you want that to mean. Let's let's I'm I'm old school.
[19:44] mean. Let's let's I'm I'm old school. How about that? I've been around for a
[19:45] How about that? I've been around for a long time. I didn't start coding in the
[19:48] long time. I didn't start coding in the 21st century. And so, I just have a way
[19:51] 21st century. And so, I just have a way of working. And it's
[19:55] of working. And it's it's sort of a chorus of things, right?
[19:58] it's sort of a chorus of things, right? So there's there's the ideas, the goal,
[20:01] So there's there's the ideas, the goal, the code, the architecture, the design,
[20:04] the code, the architecture, the design, the value proposition. Like I'm trying
[20:05] the value proposition. Like I'm trying to weave all these things together in
[20:08] to weave all these things together in the way that I am operating the machine.
[20:11] the way that I am operating the machine. So what comes out is something valuable.
[20:13] So what comes out is something valuable. And I'm usually thinking about all of
[20:16] And I'm usually thinking about all of these things as I go. Um, and usually
[20:20] these things as I go. Um, and usually when I prompt an agent, it will uh think
[20:24] when I prompt an agent, it will uh think about the code. None of the other
[20:27] about the code. None of the other things. It won't think about anything
[20:29] things. It won't think about anything else unless you you talk to it. So, I
[20:31] else unless you you talk to it. So, I I'd been very hands-on with the way that
[20:34] I'd been very hands-on with the way that I had been prompting agents. It's no
[20:36] I had been prompting agents. It's no nope, do this. I'd go and watch, check,
[20:40] nope, do this. I'd go and watch, check, etc. Um, so I certainly wasn't I'm I'm
[20:43] etc. Um, so I certainly wasn't I'm I'm not the kind of person that gets
[20:47] not the kind of person that gets comfortable with like completely
[20:48] comfortable with like completely unattended agent loops and I'm not
[20:51] unattended agent loops and I'm not trying to build dark software factories
[20:53] trying to build dark software factories because I've been outside of coding in
[20:56] because I've been outside of coding in the rest of the business long enough to
[20:58] the rest of the business long enough to realize that there be dragons and I'm
[21:01] realize that there be dragons and I'm not prepared to slay them. So I'd been
[21:04] not prepared to slay them. So I'd been very interactive and I have continued to
[21:06] very interactive and I have continued to be very interactive
[21:09] be very interactive but when I started using openspec
[21:11] but when I started using openspec instead of just hey fable let's idate
[21:14] instead of just hey fable let's idate and then build a markdown file that I
[21:16] and then build a markdown file that I could hand over to opus to go and build
[21:17] could hand over to opus to go and build things and I started following the
[21:20] things and I started following the openspec approach the open spec explorer
[21:22] openspec approach the open spec explorer we had this great conversation and I
[21:25] we had this great conversation and I really feel like I'm stepping into the
[21:26] really feel like I'm stepping into the role of the architect again I can be the
[21:29] role of the architect again I can be the architect I can talk about the
[21:30] architect I can talk about the architectural constraints what must be
[21:32] architectural constraints what must be true the the constraint that we work
[21:34] true the the constraint that we work within the why behind a lot of these
[21:37] within the why behind a lot of these decisions and of course being a language
[21:40] decisions and of course being a language model it can just change the context of
[21:43] model it can just change the context of everything else so it understands oh
[21:45] everything else so it understands oh okay yep I understand what you're
[21:46] okay yep I understand what you're talking about semantic data or modular
[21:49] talking about semantic data or modular monolith or the rest of it and why and
[21:51] monolith or the rest of it and why and what the constraints are all of those
[21:52] what the constraints are all of those things have been emerging beautifully
[21:55] things have been emerging beautifully and what what comes out the other end is
[21:58] and what what comes out the other end is is the change and so that's this that's
[22:02] is the change and so that's this that's the
[22:06] This didn't feel that different to what I'd already been doing. And I didn't
[22:07] I'd already been doing. And I didn't realize how rigorous that approach had
[22:10] realize how rigorous that approach had been because it had baked in as many
[22:14] been because it had baked in as many surprises as possible into that
[22:16] surprises as possible into that specification. So now most of what's
[22:19] specification. So now most of what's happening uh moves confidently forward.
[22:23] happening uh moves confidently forward. Well, the agent is always confident. I'm
[22:25] Well, the agent is always confident. I'm confident. And that's the harder one.
[22:29] confident. And that's the harder one. But um every now and again it's going
[22:31] But um every now and again it's going and it says, "Oh, I just noticed um
[22:35] and it says, "Oh, I just noticed um there's a there's a there's a gotcha
[22:38] there's a there's a there's a gotcha here."
[22:40] here." And uh you know, and in fact, what had
[22:42] And uh you know, and in fact, what had happened was
[22:45] happened was I was just trying to do some some basic
[22:47] I was just trying to do some some basic stuff at first. I was just bumping some
[22:50] stuff at first. I was just bumping some dependencies
[22:53] dependencies and that turned into a threeday
[22:56] and that turned into a threeday yak shaving extravaganza
[23:00] yak shaving extravaganza >> as it always does
[23:03] >> as it always does >> because
[23:04] >> because of I'm going to hold my tongue because I
[23:08] of I'm going to hold my tongue because I have so many unkind things to say about
[23:09] have so many unkind things to say about Oracle right now.
[23:11] Oracle right now. >> Oh boy.
[23:11] >> Oh boy. >> Um but it's mostly Oracle's fault.
[23:14] >> Um but it's mostly Oracle's fault. >> Hey, we're not we're not sponsored by
[23:15] >> Hey, we're not we're not sponsored by Oracle here on this occasional podcast.
[23:17] Oracle here on this occasional podcast. So, you know, you say anything you want,
[23:21] So, you know, you say anything you want, >> but it it it just shipped a bunch of
[23:23] >> but it it it just shipped a bunch of buggy drivers and then and then did
[23:26] buggy drivers and then and then did nothing for 6 months and it just turned
[23:28] nothing for 6 months and it just turned into a big ordeal. But what my
[23:31] into a big ordeal. But what my experience was cuz I had this direction
[23:34] experience was cuz I had this direction >> and we had where we were trying to go
[23:37] >> and we had where we were trying to go and we ran into roadblocks. Now, the way
[23:41] and we ran into roadblocks. Now, the way I think about it is
[23:44] I think about it is it's like Google Maps or Apple Maps or
[23:47] it's like Google Maps or Apple Maps or TomTom or whatever turnbyturn directions
[23:49] TomTom or whatever turnbyturn directions that that you use.
[23:51] that that you use. >> TomTom. Okay. Boomer.
[23:53] >> TomTom. Okay. Boomer. >> Yeah.
[23:56] >> Yeah. >> I just want to be inclusive.
[23:58] >> I just want to be inclusive. >> Please go on.
[23:58] >> Please go on. >> Don't shame me for that. [laughter]
[24:01] >> Don't shame me for that. [laughter] >> There are dozens of them. Dozens.
[24:03] >> There are dozens of them. Dozens. >> Dozens. So many. Yeah. But whatever
[24:05] >> Dozens. So many. Yeah. But whatever you're using for your turn byturn
[24:07] you're using for your turn byturn directions, GPS navigation, sometimes
[24:10] directions, GPS navigation, sometimes you will hit a roadblock that isn't in
[24:12] you will hit a roadblock that isn't in the map that wasn't in the plan or
[24:15] the map that wasn't in the plan or sometimes you just make a mistake and
[24:17] sometimes you just make a mistake and miss an exit. And what does your GPS do?
[24:20] miss an exit. And what does your GPS do? It just recalculates
[24:22] It just recalculates >> and it gets you back on track. You you
[24:24] >> and it gets you back on track. You you you go around the obstacle, gets you
[24:25] you go around the obstacle, gets you back on track. So for me the value that
[24:30] back on track. So for me the value that I observed early on was the ability to
[24:35] I observed early on was the ability to absorb meandering. So like my strategic
[24:38] absorb meandering. So like my strategic brain can plan and propose and explore
[24:42] brain can plan and propose and explore and again throwing out more openspec
[24:44] and again throwing out more openspec terms but once we're applying the change
[24:48] terms but once we're applying the change if something comes up
[24:51] if something comes up it doesn't end up getting us lost like
[24:54] it doesn't end up getting us lost like um
[24:54] um >> okay I I I wanna I want to reflect that
[24:59] >> okay I I I wanna I want to reflect that and and relate to that. Okay.
[25:01] and and relate to that. Okay. >> It also occurs to me I I just want to
[25:02] >> It also occurs to me I I just want to take a few minutes now that we're a half
[25:04] take a few minutes now that we're a half hour into this conversation. Let me
[25:06] hour into this conversation. Let me explain what this silly technology is
[25:08] explain what this silly technology is that we're talking about. And this is in
[25:10] that we're talking about. And this is in no way sponsored. Uh I'm not even a
[25:12] no way sponsored. Uh I'm not even a contributor. I don't know the folks who
[25:14] contributor. I don't know the folks who write it. I'm just a guy who discovered
[25:16] write it. I'm just a guy who discovered this thing and and said this thing looks
[25:17] this thing and and said this thing looks useful. Uh so if it sounds like I'm
[25:21] useful. Uh so if it sounds like I'm shilling for it, we're just talking
[25:22] shilling for it, we're just talking about a technology we like. That's all
[25:23] about a technology we like. That's all we're doing. So it is fundamentally a
[25:26] we're doing. So it is fundamentally a command line tool.
[25:27] command line tool. >> Um and a set of skills, very particular
[25:29] >> Um and a set of skills, very particular set of skills. Um
[25:32] set of skills. Um and when you you know say on a Mac
[25:35] and when you you know say on a Mac you'll brew install openspec and uh uh
[25:39] you'll brew install openspec and uh uh you have a new project you'll you'll
[25:41] you have a new project you'll you'll install the skills into the dotcloud or
[25:44] install the skills into the dotcloud or agent or you know wherever they go in
[25:46] agent or you know wherever they go in your particular coding harness. Um and
[25:48] your particular coding harness. Um and it's got support for a couple dozen of
[25:50] it's got support for a couple dozen of those. Um then there's this workflow.
[25:53] those. Um then there's this workflow. Okay. So you got those skills, you got
[25:54] Okay. So you got those skills, you got that command line tool. Um and there's
[25:57] that command line tool. Um and there's this workflow. So if you know what you
[25:58] this workflow. So if you know what you want to do, you propose a change and
[26:01] want to do, you propose a change and that's the that's the propose opx
[26:04] that's the that's the propose opx propose command. And that can be a
[26:07] propose command. And that can be a sentence, that can be uh you know
[26:09] sentence, that can be uh you know paragraphs and bullet points, whatever
[26:12] paragraphs and bullet points, whatever your definition is of the thing that you
[26:14] your definition is of the thing that you want to get done. And that's called a
[26:16] want to get done. And that's called a change. So and I I know you know all
[26:18] change. So and I I know you know all this just in case the audience doesn't.
[26:19] this just in case the audience doesn't. So you you uh propose a change and there
[26:23] So you you uh propose a change and there might be a little back and forth on that
[26:25] might be a little back and forth on that proposal. There might be some things
[26:26] proposal. There might be some things that are unclear and the agent might
[26:27] that are unclear and the agent might say, "Well, you know, what do you mean
[26:29] say, "Well, you know, what do you mean about this?" Or, "Uh, it's not clear
[26:30] about this?" Or, "Uh, it's not clear what library to use for this. Do you
[26:32] what library to use for this. Do you want to spike this?" There can be
[26:34] want to spike this?" There can be productive discussion there, right?
[26:36] productive discussion there, right? >> If you don't quite understand your
[26:38] >> If you don't quite understand your change well enough to jump in there, you
[26:40] change well enough to jump in there, you can explore, which again, this is like
[26:44] can explore, which again, this is like uh, you know, a feature that's been in
[26:46] uh, you know, a feature that's been in most coding harnesses all along. But the
[26:48] most coding harnesses all along. But the explore command is intentionally like,
[26:51] explore command is intentionally like, "Hey, man, I don't quite know what I'm
[26:52] "Hey, man, I don't quite know what I'm doing. I want to think about this
[26:55] doing. I want to think about this um you know how is raw image decode
[26:58] um you know how is raw image decode going to work in my Lightroom
[27:00] going to work in my Lightroom replacement. There's a lot to talk
[27:02] replacement. There's a lot to talk through
[27:03] through >> and usually that process you're
[27:06] >> and usually that process you're interacting with the the model and the
[27:07] interacting with the the model and the skill
[27:09] skill >> um and whatever existing specs and
[27:10] >> um and whatever existing specs and existing code bases are in there.
[27:12] existing code bases are in there. Usually that's iterative. It's asking
[27:13] Usually that's iterative. It's asking you questions and it's freaking
[27:15] you questions and it's freaking challenging. It it often when I'm
[27:18] challenging. It it often when I'm exploring I'm like oh yeah I didn't
[27:20] exploring I'm like oh yeah I didn't think of that. That's a really good
[27:21] think of that. That's a really good question. I I I I it's it's hard work to
[27:24] question. I I I I it's it's hard work to go through to do that architecture uh
[27:27] go through to do that architecture uh work. So if you explore you explore for
[27:30] work. So if you explore you explore for a little while and then that becomes a
[27:31] a little while and then that becomes a proposal. If you kind of understand
[27:32] proposal. If you kind of understand things well enough, you propose. Now,
[27:34] things well enough, you propose. Now, you've got a thing called a change in
[27:37] you've got a thing called a change in flight. And a change is um a proposal,
[27:42] flight. And a change is um a proposal, which is like a description of kind of
[27:43] which is like a description of kind of what you're doing, a design document,
[27:45] what you're doing, a design document, which is just a more detailed
[27:46] which is just a more detailed implementation,
[27:47] implementation, >> a task list, and not to be forgotten, a
[27:51] >> a task list, and not to be forgotten, a set of actual specs. You know, when blah
[27:54] set of actual specs. You know, when blah blah blah, then shall this shall blah,
[27:56] blah blah, then shall this shall blah, you know, that sort of formal spec
[27:57] you know, that sort of formal spec language. That's all in markdown and on
[28:00] language. That's all in markdown and on a perch change basis any
[28:04] a perch change basis any >> but it does have shades of BDD that
[28:08] >> but it does have shades of BDD that become
[28:09] become not like an executable spec per se but
[28:14] not like an executable spec per se but like you can see it from there.
[28:17] like you can see it from there. >> You can you can
[28:20] >> You can you can get I'll get there. There's
[28:22] get I'll get there. There's >> the specs don't just go into a folder to
[28:24] >> the specs don't just go into a folder to die and never be heard from again. um
[28:28] die and never be heard from again. um when you know development of something
[28:30] when you know development of something complex, new changes are going to
[28:33] complex, new changes are going to interact with old specs and and and
[28:35] interact with old specs and and and we'll get there. So um once you got a
[28:38] we'll get there. So um once you got a proposal and you like it, you read the
[28:40] proposal and you like it, you read the design document, it makes sense, then
[28:41] design document, it makes sense, then you apply it and apply goes and writes
[28:45] you apply it and apply goes and writes all the code. Um and typically some of
[28:48] all the code. Um and typically some of the tests are whatever automatic
[28:50] the tests are whatever automatic verification the agent can do. Some of
[28:52] verification the agent can do. Some of them might be manual verification. Uh
[28:54] them might be manual verification. Uh there's a new version of OpenSpec that
[28:56] there's a new version of OpenSpec that has a verify step um a verify command
[28:59] has a verify step um a verify command that you can get a different model to go
[29:01] that you can get a different model to go in and and and check things out.
[29:03] in and and and check things out. Actually, I'll confess I haven't done
[29:05] Actually, I'll confess I haven't done that yet. But um when you're done,
[29:07] that yet. But um when you're done, everything works. Then you archive the
[29:09] everything works. Then you archive the change and it is no longer an active
[29:11] change and it is no longer an active change. But the the design document, the
[29:14] change. But the the design document, the proposal, the task list and the specs
[29:17] proposal, the task list and the specs stay in the open spec directory as as a
[29:21] stay in the open spec directory as as a as a official part of the repository of
[29:23] as a official part of the repository of this project.
[29:24] this project. >> And this this is coming back around to
[29:27] >> And this this is coming back around to something you said how it keeps the
[29:28] something you said how it keeps the agent on track. At least you've found
[29:29] agent on track. At least you've found that it keeps the agent on track. um
[29:33] that it keeps the agent on track. um all the modern harnesses and the the the
[29:36] all the modern harnesses and the the the contemporary models contemporary as of
[29:38] contemporary models contemporary as of this recording do a really good job
[29:40] this recording do a really good job saying okay what's out there in the
[29:41] saying okay what's out there in the codebase what am I doing what do I need
[29:43] codebase what am I doing what do I need to look at and you know maybe if there's
[29:45] to look at and you know maybe if there's a little markdown file of whether it's a
[29:47] a little markdown file of whether it's a agent MD or claw MD or some other sort
[29:49] agent MD or claw MD or some other sort of advice you know they'll find those
[29:51] of advice you know they'll find those things but now explicitly there's this
[29:55] things but now explicitly there's this long list of specs and the specs are
[29:58] long list of specs and the specs are merged into the master spec
[30:01] merged into the master spec for the application. And this isn't
[30:06] for the application. And this isn't part of the problem with writing
[30:08] part of the problem with writing functional specs in the bad old days was
[30:10] functional specs in the bad old days was it was a focus on the software, the
[30:13] it was a focus on the software, the thing you were building and not the
[30:15] thing you were building and not the people who used it and what they were
[30:16] people who used it and what they were doing.
[30:17] doing. >> Um, and it was very easy to just get
[30:20] >> Um, and it was very easy to just get absorbed and off the off into weird
[30:22] absorbed and off the off into weird territory and and not thinking about the
[30:25] territory and and not thinking about the people who use your thing. Um,
[30:27] people who use your thing. Um, >> there's there's actually another
[30:28] >> there's there's actually another dimension that's just worth mentioning.
[30:30] dimension that's just worth mentioning. Yes.
[30:30] Yes. >> Uh those specs were about building
[30:32] >> Uh those specs were about building software and we don't build software and
[30:35] software and we don't build software and we have not built software for a very
[30:37] we have not built software for a very long time.
[30:39] long time. >> Developers change software. Software
[30:41] >> Developers change software. Software engineering is change cycles that are
[30:43] engineering is change cycles that are nested and that's a very important
[30:46] nested and that's a very important detail
[30:47] detail >> because when you change software there's
[30:50] >> because when you change software there's important context that doesn't exist in
[30:52] important context that doesn't exist in the code.
[30:54] the code. >> It exists in a list of previous
[30:56] >> It exists in a list of previous decisions and commitments. And now, and
[31:00] decisions and commitments. And now, and I'm I'm just seeing this in my my
[31:02] I'm I'm just seeing this in my my current project, my Lightroom
[31:03] current project, my Lightroom replacement, I have this list of, I
[31:06] replacement, I have this list of, I don't know, 50 60 proposals that are the
[31:09] don't know, 50 60 proposals that are the whole vision for the thing, explorations
[31:11] whole vision for the thing, explorations and prop changes, 50 or 60 changes that
[31:14] and prop changes, 50 or 60 changes that are going to be an initial version of
[31:16] are going to be an initial version of this thing. And occasionally it'll say,
[31:17] this thing. And occasionally it'll say, well, okay, but this contradicts a spec
[31:20] well, okay, but this contradicts a spec that you've got from back here, and
[31:21] that you've got from back here, and that'll have to change. What do you
[31:23] that'll have to change. What do you think about that?
[31:25] think about that? For me,
[31:30] what I'm most excited about when it comes to openspec, like and again, I I I
[31:33] comes to openspec, like and again, I I I fully admit that there are, like you
[31:35] fully admit that there are, like you said, there are other approaches. Um,
[31:37] said, there are other approaches. Um, none of them are really right. None of
[31:39] none of them are really right. None of them are really wrong. It's just a
[31:40] them are really wrong. It's just a different set of trade-offs.
[31:41] different set of trade-offs. >> Mh.
[31:42] >> Mh. >> But for me, I always go back to I think
[31:44] >> But for me, I always go back to I think it was Peter no, Peter Ner, Peter Ner, I
[31:48] it was Peter no, Peter Ner, Peter Ner, I don't know how to pronounce his last
[31:49] don't know how to pronounce his last name. I've only ever read it. Do you
[31:50] name. I've only ever read it. Do you know who I'm talking about?
[31:51] know who I'm talking about? >> Yes.
[31:53] >> Yes. uh he wrote a paper back I think in the
[31:55] uh he wrote a paper back I think in the 80s called uh uh it was like programming
[31:59] 80s called uh uh it was like programming as theory building. So he has this
[32:01] as theory building. So he has this concept of uh what he calls a theory but
[32:05] concept of uh what he calls a theory but um essentially there is there is the
[32:09] um essentially there is there is the program
[32:11] program and the program is more than the code.
[32:14] and the program is more than the code. The code is like a raification of the
[32:17] The code is like a raification of the part of the program that happens in the
[32:19] part of the program that happens in the machine. But there is there is something
[32:22] machine. But there is there is something bigger that that exists outside of the
[32:25] bigger that that exists outside of the machine because the reality is the
[32:26] machine because the reality is the software that we build exists to deliver
[32:29] software that we build exists to deliver value and where it delivers value
[32:32] value and where it delivers value ultimately is outside of the computer.
[32:34] ultimately is outside of the computer. The the problem is outside the solution
[32:35] The the problem is outside the solution is inside. And so by definition just
[32:40] is inside. And so by definition just looking at the code you're only looking
[32:42] looking at the code you're only looking at part of the program part of the part
[32:46] at part of the program part of the part of the solution or part of the problem.
[32:49] of the solution or part of the problem. He says that code is uh is lossy.
[32:52] He says that code is uh is lossy. And what I found is is continuing to use
[32:56] And what I found is is continuing to use open spec
[32:58] open spec um as a as a consistent
[33:01] um as a as a consistent process and and even ritual uh allows me
[33:06] process and and even ritual uh allows me to expand not only the code but the the
[33:09] to expand not only the code but the the program as theory. It's it's the
[33:12] program as theory. It's it's the >> it's the theory building part because
[33:14] >> it's the theory building part because >> what it's maintaining in those markdown
[33:16] >> what it's maintaining in those markdown files and the opinionated structure that
[33:19] files and the opinionated structure that it adopts is basically the kind of
[33:23] it adopts is basically the kind of learning that happens in a team.
[33:26] learning that happens in a team. Like I remember one of my jobs as a
[33:27] Like I remember one of my jobs as a coder I show up and they say, "Hey,
[33:29] coder I show up and they say, "Hey, Michael, go do this thing." And I I'm
[33:31] Michael, go do this thing." And I I'm like, "Oh, okay. Yeah, we need to do
[33:33] like, "Oh, okay. Yeah, we need to do this. I know how to do that." D. And all
[33:36] this. I know how to do that." D. And all of a sudden, why is production down? Oh,
[33:39] of a sudden, why is production down? Oh, you didn't know that there was a
[33:41] you didn't know that there was a loadbearing bug in that particular class
[33:44] loadbearing bug in that particular class and nobody ever touches that. Like
[33:46] and nobody ever touches that. Like that's
[33:47] that's >> no,
[33:48] >> no, >> it just like this is stuff that doesn't
[33:50] >> it just like this is stuff that doesn't fit in comments.
[33:52] fit in comments. >> This is the tribal knowledge.
[33:54] >> This is the tribal knowledge. >> Yeah. And what I really appreciate is it
[33:59] >> Yeah. And what I really appreciate is it gives you a mechanism for capturing the
[34:03] gives you a mechanism for capturing the context behind decisions, not just the
[34:06] context behind decisions, not just the code that resulted. And that is
[34:09] code that resulted. And that is enormously valuable.
[34:12] enormously valuable. >> That makes excellent point.
[34:14] >> That makes excellent point. >> It makes the agent part of the team
[34:16] >> It makes the agent part of the team instead of the bot that you say, "Okay,
[34:18] instead of the bot that you say, "Okay, now make code do this."
[34:21] now make code do this." >> Right. Right. Excellent point. There's a
[34:24] >> Right. Right. Excellent point. There's a learning loop there.
[34:25] learning loop there. >> There there is and and
[34:26] >> There there is and and >> built in
[34:28] >> built in >> uh the reality again I in my experience
[34:30] >> uh the reality again I in my experience I see it interacting with previous
[34:34] I see it interacting with previous changes you know like you got this from
[34:36] changes you know like you got this from this change and you decided in you know
[34:38] this change and you decided in you know 2 what you called 2.2 to using that
[34:40] 2 what you called 2.2 to using that numbering system which doesn't help my
[34:42] numbering system which doesn't help my anti-functional spec uh spiel, but
[34:46] anti-functional spec uh spiel, but anyway, you know, whatever. It's got a
[34:47] anyway, you know, whatever. It's got a name. It's called Fred Back in Spec
[34:49] name. It's called Fred Back in Spec Fred. Uh you you deferred that and it
[34:52] Fred. Uh you you deferred that and it sounds like you want to get back into it
[34:53] sounds like you want to get back into it now, you know. So, it it's it's got that
[34:56] now, you know. So, it it's it's got that history. It's a very good point. It's a
[34:57] history. It's a very good point. It's a very good point. I think of it as
[34:58] very good point. I think of it as guardrails for me, but it's also really
[35:00] guardrails for me, but it's also really a lot of state, a lot of knowledge uh
[35:02] a lot of state, a lot of knowledge uh about the project that the agent has
[35:04] about the project that the agent has access to, the harness has access to and
[35:07] access to, the harness has access to and and we have conversations
[35:10] and we have conversations and and some of these things basically
[35:12] and and some of these things basically become integrated architectural decision
[35:15] become integrated architectural decision decision records. Given this context and
[35:18] decision records. Given this context and and this this problem
[35:21] and this this problem and this decision criteria, we evaluated
[35:24] and this decision criteria, we evaluated these options and landed on that one.
[35:26] these options and landed on that one. the these are here are all the
[35:27] the these are here are all the trade-offs of all the different options
[35:29] trade-offs of all the different options and here's why this set of trade-off
[35:30] and here's why this set of trade-off matters most most for this project like
[35:33] matters most most for this project like that is now context that the next turn
[35:36] that is now context that the next turn with the next bot that's going in there
[35:39] with the next bot that's going in there isn't going to just blindly undo it
[35:41] isn't going to just blindly undo it because it's inconvenient.
[35:43] because it's inconvenient. >> Right. Right. Or it's just an
[35:45] >> Right. Right. Or it's just an inconsequential
[35:46] inconsequential thing not to be considered uh and you
[35:48] thing not to be considered uh and you know we're focused on our current prompt
[35:50] know we're focused on our current prompt and our our current task. Excellent. or
[35:52] and our our current task. Excellent. or it becomes a decision that no longer
[35:54] it becomes a decision that no longer holds, that is now being surfaced and a
[35:58] holds, that is now being surfaced and a human is making that distinction rather
[36:01] human is making that distinction rather than a stochcastic um uh
[36:05] than a stochcastic um uh chatbot. But for me that operational
[36:07] chatbot. But for me that operational learning is is one of the most valuable
[36:09] learning is is one of the most valuable pieces.
[36:11] pieces. Um and it I noticed it at first in a
[36:14] Um and it I noticed it at first in a very small way because
[36:17] very small way because I'm on Windows for better or for worse.
[36:21] I'm on Windows for better or for worse. Um, and I know that puts me in the
[36:24] Um, and I know that puts me in the minority because your first instinct was
[36:26] minority because your first instinct was to send the link to this podcast um,
[36:31] to send the link to this podcast um, studio online to my phone.
[36:34] studio online to my phone. >> I I in fact texted it to you. Yeah. Not
[36:36] >> I I in fact texted it to you. Yeah. Not not helpful.
[36:37] not helpful. >> I did.
[36:38] >> I did. >> Yeah.
[36:38] >> Yeah. >> Uh, I mean that works for you and you
[36:40] >> Uh, I mean that works for you and you know, Slutton does the same thing and a
[36:42] know, Slutton does the same thing and a bunch of, you know, all all of my Mac
[36:44] bunch of, you know, all all of my Mac friends are like, "Wait, you're not on
[36:45] friends are like, "Wait, you're not on an Apple device."
[36:46] an Apple device." >> Probably because I didn't put it in the
[36:47] >> Probably because I didn't put it in the calendar like I said I was going to, but
[36:49] calendar like I said I was going to, but you know, still. That's all right. But
[36:50] you know, still. That's all right. But then I got it moments later before I had
[36:52] then I got it moments later before I had to send you a snotty a snotty response.
[36:55] to send you a snotty a snotty response. But um the uh you know I'm on a I'm on a
[36:59] But um the uh you know I'm on a I'm on a Windows machine. But uh to help me help
[37:02] Windows machine. But uh to help me help me deal with that, I spend a lot of time
[37:05] me deal with that, I spend a lot of time in Zshell in in the Windows subsystem
[37:08] in Zshell in in the Windows subsystem for Linux. And so that's that's where I
[37:11] for Linux. And so that's that's where I try to spend most of my time. And uh
[37:15] try to spend most of my time. And uh that's where I installed OpenSpec. And
[37:17] that's where I installed OpenSpec. And of course, I've got Cloud running
[37:19] of course, I've got Cloud running wherever Cloud's running. And and it
[37:22] wherever Cloud's running. And and it says, "I can't find OpenSpec the first
[37:24] says, "I can't find OpenSpec the first time." And then I said, "Oh, it's in
[37:27] time." And then I said, "Oh, it's in WSL." Oh, okay. And then it just shells
[37:32] WSL." Oh, okay. And then it just shells over to WSL and assumes Bash and it's
[37:34] over to WSL and assumes Bash and it's not in Bash because I don't use Bash. I
[37:36] not in Bash because I don't use Bash. I haven't got anything configured in Bash.
[37:38] haven't got anything configured in Bash. And it says, "I still can't find it." I
[37:39] And it says, "I still can't find it." I said, "It's in I'm use Zshell." Oh,
[37:41] said, "It's in I'm use Zshell." Oh, okay. Yep, there it is. And then that
[37:44] okay. Yep, there it is. And then that was it. That was the end of it. every
[37:45] was it. That was the end of it. every subsequent session knew that. Uh,
[37:48] subsequent session knew that. Uh, another one was it started spinning up
[37:50] another one was it started spinning up all these different work trees and
[37:51] all these different work trees and things like that and uh, putting
[37:54] things like that and uh, putting everything in branches
[37:56] everything in branches and you know that's fine. I'm not I'm
[37:59] and you know that's fine. I'm not I'm not here to yuck anybody's branchy yum,
[38:01] not here to yuck anybody's branchy yum, but
[38:03] but I am a
[38:06] I am a I'm a old school continuous delivery
[38:10] I'm a old school continuous delivery trunkbased development kind of human
[38:12] trunkbased development kind of human being and and I believe if you if you're
[38:14] being and and I believe if you if you're doing continuous delivery right then the
[38:16] doing continuous delivery right then the branches slow you down. Um but
[38:22] branches slow you down. Um but and also I think you know continuous
[38:24] and also I think you know continuous integration is valuable but uh in the
[38:27] integration is valuable but uh in the beginning it just always started to uh
[38:30] beginning it just always started to uh do things in branches. I said well this
[38:32] do things in branches. I said well this isn't very helpful.
[38:33] isn't very helpful. >> I do we do trunkbased development here
[38:35] >> I do we do trunkbased development here and it says oh okay got it and then that
[38:37] and it says oh okay got it and then that goes into your config.yml YAML which is
[38:40] goes into your config.yml YAML which is basically
[38:46] >> another file even talked about right but it's it's
[38:48] it's it's >> yeah things it needs to know
[38:51] >> yeah things it needs to know >> but that is the that is the value of
[38:53] >> but that is the that is the value of persistent context it knows standard
[38:55] persistent context it knows standard operating procedures it knows uh
[38:58] operating procedures it knows uh enormous context about past decisions
[39:02] enormous context about past decisions and where we've run into similar
[39:04] and where we've run into similar crossroads and how we evaluated those
[39:08] crossroads and how we evaluated those And not only did I feel like my agent
[39:10] And not only did I feel like my agent got smarter, but my team got smarter as
[39:14] got smarter, but my team got smarter as well because it was flagging
[39:17] well because it was flagging inconsistencies that we had in our
[39:20] inconsistencies that we had in our approaches that we didn't know about.
[39:24] approaches that we didn't know about. Um, so that was just another side effect
[39:29] Um, so that was just another side effect that you you saw that uh help move you
[39:33] that you you saw that uh help move you forward.
[39:34] forward. >> Yeah. My guest today has been Michael
[39:37] >> Yeah. My guest today has been Michael Karduchi, developer, speaker, teacher,
[39:40] Karduchi, developer, speaker, teacher, software architect, rock on tour,
[39:43] software architect, rock on tour, magician
[39:45] magician >> and knowledge grass, grief monkey, grief
[39:48] >> and knowledge grass, grief monkey, grief monkey.
[39:49] monkey. >> Michael, thanks for being a part of the
[39:50] >> Michael, thanks for being a part of the show.
[39:52] show. >> Tim, it's always a pleasure. Good to see
[39:54] >> Tim, it's always a pleasure. Good to see you.
All frames
Total: 40. Hero frames flagged with star.
frames/frame_0001.jpg(t=00:00)frames/frame_0002.jpg(t=00:39)frames/frame_0003.jpg(t=01:24)frames/frame_0004.jpg(t=02:00)frames/frame_0005.jpg(t=02:08)frames/frame_0006.jpg(t=04:41)frames/frame_0007.jpg(t=05:39)frames/frame_0008.jpg(t=06:39)frames/frame_0009.jpg(t=07:14)frames/frame_0010.jpg(t=08:18)frames/frame_0011.jpg(t=08:44)frames/frame_0012.jpg(t=09:15)frames/frame_0013.jpg(t=10:09)frames/frame_0014.jpg(t=10:21)frames/frame_0015.jpg(t=10:22)frames/frame_0016.jpg(t=12:03)frames/frame_0017.jpg(t=16:14)frames/frame_0018.jpg(t=17:40)frames/frame_0019.jpg(t=18:46)frames/frame_0020.jpg(t=18:58)frames/frame_0021.jpg(t=21:47)frames/frame_0022.jpg(t=22:40)frames/frame_0023.jpg(t=22:58)frames/frame_0024.jpg(t=23:42)frames/frame_0025.jpg(t=23:50)frames/frame_0026.jpg(t=25:28)frames/frame_0027.jpg(t=26:15)frames/frame_0028.jpg(t=28:03)frames/frame_0029.jpg(t=28:17)frames/frame_0030.jpg(t=28:54)frames/frame_0031.jpg(t=29:24)frames/frame_0032.jpg(t=29:38)frames/frame_0033.jpg(t=29:57)frames/frame_0034.jpg(t=31:24)frames/frame_0035.jpg(t=31:51)frames/frame_0036.jpg(t=32:15)frames/frame_0037.jpg(t=32:25)frames/frame_0038.jpg(t=33:10)frames/frame_0039.jpg(t=36:06)frames/frame_0040.jpg(t=36:18)