A lot of disciplines need to come together to create a great digital product. This is true at the agent level as much as it is at the human level. The agent version of this often involves jumping between different environments, tooling, models, skills, etc. It's a process that is chaotic, fatiguing, and never ending. But it does produce the best results... so we have endured this pain...until now.
Workbenches is our attempt to fix that. It's an open standard that packages everything behind a certain kind of outcome: model, harness config, skills, MCP, reference files, tooling, and runtime. You might have one workbench tuned for game physics and a completely different one for art direction, and switch between them without rebuilding anything.
The video shows how I use it day to day. I plan in my own harness (Claude Code with Opus), decompose the plan to respective workbenches, they run in the cloud in parallel (close my laptop), and each one comes back on its own branch. I do a light amount of work with CC to assemble them, and it goes up as one PR.
You can also pull down other people's workbenches. A few ones are up now (three.js, frontend, Svelte, Godot), and the goal is that the developers and artists you admire can publish how they get their results, so your product can end up using a blend of their workbenches and your own.
Happy to throw some tokens your way for you to kick the tires, and if you've built product with AI and have a method you think gets better results than everyone else's, I want to talk. We're setting up paid workbenches where creators earn from every run.
The most frustrating part of working with agents daily: they don't get what you're really after. They spend hours on something that "meets the spec" but cuts every corner.
Skills and CLAUDE.md are a good step in the right direction, but don't solve this. What's missing is taste.
Taste is how YOU would do it. What looks good to you, what you've seen in your years of trying and breaking and fixing, and the scar tissue you've built in an area of expertise. It's all sitting in your head and your agent's memories waiting to be used
1
31
The biggest issue I face when working with agents is context. When I start a claude thread to do work, I either need to have a saved .md file, or ask claude to dig around the repo to refamiliarize itself with what we're working on
We created pompeii.ai to solve this
2
324
We're launching something new soon. Before I do, I want to share the worldview that shaped it.
Every engineering team using agents is having the same conversation right now.
"We are shipping more than ever... but at what cost?"
Features getting prompted into existence because the friction to build them has disappeared. Tech debt accelerating because it's not possible to thoughtfully engage with dozens of PRs a day. Cleanup being optimistically deferred to the doorstep of some future model release.
And beneath the tech debt: a type of education debt. When engineers write code by hand, they absorbed system understanding as a side effect. That side effect compounded with every update they pushed. That's also disappearing.
So, how are teams fighting this?
Calling for more "discipline".
"Read code more thoroughly." "Stay sharp by manually coding things in your free time." "Leave the code better than you found it."
I love it. It's true. And it will fail miserably.
It's good discipline to audit every transitive dependency an npm package introduces. It's good discipline to understand every line of a Stack Overflow answer before pasting it in.
Convenience and leverage have beaten discipline every time. And we expect discipline to suddenly win against the most convenient, highest leverage tool ever created?
Self-imposed friction is just simply not the answer for teams to successfully adopt AI.
But undisciplined use of code gen agents = slop.
The problem is in the infrastructure that agentic work gets done. And in the feedback loops within that infrastructure. This work runs through tools designed for humans exclusively at the wheel : terminals, ticket systems, chat apps with bolted on AI.
Agents amplify output without amplifying good product judgement.
The AI systems that exist currently don't reward high quality, thoughtful work enough. Discipline has intangible outcomes in these systems. Especially in team environments.
How you work, your domain expertise, your taste...it doesn't get captured. It doesn't compound.
It should. When working with agents, disciplined work should be SEEN. The choices you wrestled with along the way should be SEEN. The standards you set should be SEEN. And at the end of the day you should rest easy knowing the effort you put into your work is absorbed and redistributed across your entire org.
The answer isn't just more discipline. It's architecture that absorbs and rewards discipline.
Agents are a powerful new vehicle that need a thoughtfully designed new steering wheel.
When your team's product discussions, taste, values, standards live in the same surface as execution... that's a better steering wheel.
When the system your team works in, passively absorbs useful context and applies it automatically to agent work...that's a better steering wheel.
When feedback loops reward thoughtful, high discipline work...that's a better steering wheel.
And we're calling this steering wheel : Pompeii.
I predict that writing code by hand will have a nostalgic resurgence the same way that buying vinyl records did
29
Introducing Nero: open source agent that's real fun to use
* autonomy mode where nero has his own projects & priorities
* use any model through openrouter
* claude code inspired native MCP / skills support
* access nero from any device on your LAN, allowing him to create custom HUD & move between any of them
* locally served web UI
No Claw, no MEMORY.md file, and no need for a mac mini (though you can certainly feel free!)
The part of nero I find coolest rn is the connected displays. With nero running on my local network, I can open https://nero.local/display/[NAME] and nero immediately has a new device available with the name you set
This allows nero to literally move himself from one device to another, and put custom HUD elements across any of them. Coolest feature of any agent I've used by far
29
Introducing Nero : local, proactive, agent... that does things a bit different than 🦞
- persistent node graph memory
- local ui platform to manage and visualize everything
- native claude code support
- call it. text it. live in it.
github.com/pompeii-labs/nero…






