@shaharkaminsky I'd be interested in how you would rearrange things to be more optimal
English
Damien C. Tanner
17.4K posts
Damien C. Tanner
@dctanner
Building your AI computer in the cloud @usetoyo. Prev founded @pusher, @thoughtbot
London, England Katılım Kasım 2006
263 Takip Edilen5.2K Takipçiler
Love this direction overall. I also connect with the meticulous spec part and throwing away code part
But a product director submitting the PR and then an engineer having to see it through feels like a forced and suboptimal split of responsibility. I don’t really see what it actually gains. Genuinely curious if there’s a real upside I’m missing
English
@dctanner @inworld_ai @ctatedev this is cool. i haven't gotten around to trying inworld, have only used elevenlabs.
how do you find inworld?
English
I'm living in the future now with this demo-video skill.
My claude/codex will work for an hour, then it interrupts me opening up QuickTime with demo.mp4 it's recorded with its own voice!
Here's the demo-video skill: gist.github.com/dctanner/ee7fe…
Thanks @inworld_ai and @ctatedev
English
@dctanner @inworld_ai @ctatedev @Remotion Here's the mp4 it produced for voiceview.app it even mocked up the screenshots you see on the right (it's all HTML, no actual images in that case)
English
@brynary Could you explain the use cases more? Is the goal similar to OpenAI’s Symphony?
English
Today I'm thrilled to open source what I've been working on...
Meet Fabro (github.com/fabro-sh/fabro), the dark software factory for small teams of expert engineers.
Fabro gets you out of the REPL (read-eval-prompt-loop) by layering deterministic workflow graphs over agent sessions. It's batteries included with cloud sandboxes, quality sign offs, multi-model ensembles, and Git checkpoints.
It's MIT licensed so you can fork and own your AI coding toolchain. Single Rust binary with zero deps.
I'd love to hear what you think!
English
This week has our BIGGEST launch in a while...
UK and EU coverage is coming to the Fiscal Data Feed.
This means the financials for 🇬🇧 🇪🇺 will have:
⚡ Instant data deliverability
📊 As-reported and Standardized
🔍 Click-thru audit number to filing
📈 20 years of history
English
@humungasaurus @ryancarson I also wanted to try docker-in-docker as an even cleaner approach.
English
@humungasaurus @ryancarson I had this and just told it to fix it. It made a dev.sh script that does dynamic port discovery and updates all the env vars in my monorepo. It also duplicates my develop branch docker Postgres so all worktrees start with a db I know has good example data in it.
English
Your homework for today ...
Spend 8 hours completely optimizing your code factory and all of your agentic harnessing.
It's tedious and frustrating, but it's worth it.
I spent all of yesterday cleaning out automations, getting agent browser testing refined, culling skills and helper scripts.
Really fine-tuned my code factory, and wow, it is worth it.
English
Wish him luck on the night shift: sh ./ralph_codex.sh simplify-code-plan.md 100
English
I'd love a meta-refactor agent swam thing, which similar to Claude Code's /simplify, goes through every file in your repo, ensures full test coverage, then refactors to simplify code and reduce LOC.
English
@maxthelion NPCs in your repo is a great way to think about it
English
@dctanner I’ve been thinking about this today. Proactive agents. I was thinking of experimenting with behaviour trees from game ai (npcs). Looks around, checks for code smell, attacks, goes back to checking. Could be awesome.
English
Damien C. Tanner retweetledi
AI Engineer London has quickly become one of my favourite meetups 🙌
Great talks by @badlogicgames @dougli_ and @zeyu1337
Plus a surprise AMA with @steipete.
Epic.
English
@mitsuhiko The person who fixes this and all the fallback code will be a hero.
I’d love linters. But maybe the more flexible solution is to send every edit to a fast LLM like flash with a list of the rules.
English
I feel like I need to start fighting this shit with more linters. They are just not listening.
English
We’ve stuck with a single thread Chief of staff in @usetoyo - it coordinates subagents for every task whilst maintaining context throughout the day.
This is how we work with other humans. It’s the most natural interaction for us.
Dan Shipper 📧@danshipper
this is what my codex chat history looks like. we absolutely need a new UX for coding agent guis a few thoughts: 1. my usual workflow is i have one main chat where im doing all my work for the day and then separate windows for each individual issue / feature. once those are in good shape and reviewed, i send them to my main chat to qa on staging and deploy 2. the "main chat" paradigm is nice because i want one place to have a good idea of "here's everything i need to accomplish today, what's been done, and what hasn't"—but right now it's at the same level of prominence in the UX / priority as every other chat which isn't great. 3. the main chat "orchestrator" should also be able to easily know what's going on in other threads. i need to be able to reference other chats easily and also have each sub thread have some idea of what the others are doing by default (should be able to isolate if i want though) 4. it's pretty wild to be doing your n-th codex run of the day, type what you want, and then have the model be like "okay, let's see what's going on here, ahhh there's a repo, and it has X features..." as if it has no idea what i've been doing that day. seems like a huge waste of time and tokens, we need something better by default.
English
@thsottiaux Love the direction of the Codex App. But rendering (ie showing it after clicking it in sidebar) of threads that are long or have a large diff, are sloooow.
English
Working at OpenAI is fun because questioning everything and taking risks is part of the culture. Within Codex, the team asks itself how we could make it an order of magnitude better every few months and then sets most things aside to go and do it across the entire stack. Some examples were the Codex App and our first deployment of Cerebras inference with WebSockets. We are now well under way on the next bet and it’s making even our best engineers nervous as it’s at the edge of what’s possible today.
English
My current agentic workflow is about 5x faster, better quality, I understand the system better, and I’m having fun again.
My previous workflows have left me exhausted, overwhelmed, and feeling out of touch with the systems I was building. They also degraded quality too much.
This is way better.
I’m not ready to describe in detail. It’s still evolving a bit. But I’ll give you a high level here.
I call this the Night Shift workflow.
English
@dctanner Imagining these agents working in parallel and solving hard problems. Inventions will come with 100x speed.
English
Chatbot architectures kept AI locked in a tiny box
Claude Code was the first real time we all let AI loose in a computer
A simple todo list let it work for hours
Running the test suite let it check its work
Give it browser control and it can get the ui perfect
OpenClaw then gave agents real freedom
We’re discovering a ton of capability overhang in the models
Like a human, it just needs a goal and a computer to get stuff done
2026 is going to see us applying this to all knowledge work
Anything a human can do on a computer, will be done by an agent in a for loops
Yes they’ll be slower than us at first
But you can run as many as you want in parallel
Yes they’ll make mistakes
But they’re only going to get better
Every knowledge work job should expect to be automated, soon
Code was just the first
English