Can't wait to take @AmpCode Spaces for a ride. My workflow has been organically moving towards this way of working anyway. Constantly working in the weeds feels unproductive. Agents operate at that level so much more effectively. https://t.co/FpcY7FA9et
@Voxyz_ai Visuals are nice touch particularly treemap. Adding to my regular cleanup routine. Mostly use cloud agents these days but occasional privacy sensitive local work means disk bloat
@rockatanescu I agree but caveated with production systems required to evolve still need explicit architectural context ADRs and ingrained decoupling and observability. This reduces maintenance drift, bugs/token spend in future cycles.
@maxedapps More nuanced IMO. AI dramatically lowers implementation cost, but not total cost of ownership.
Even as a dev, in most cases I'd rather pay for mature software than own its reliability and maintenance. Any time not focused on the core business is expensive.
@ryancarson Clear illustration of why observability is so critical. It's always mattered but the stakes are so much higher for labs. Fundamentals are still important.
@sqs Absolutely. I've found architecting/optimizing for parallel agents and applying the principles of trunk-based development works well for multi-agent builds. Look forward to folding orbs into this process.
@davis7 Fable doesn't necessarily have to write the code just be strategic advisor, architect and judge. Even then human in the loop still required but less than Sol.
When benchmarking AI we often romanticise human dev systems which frequently produced mediocre output, verification was often nonexistent or weak. Applying AI to these workflows has in many cases, improved quality. Perfection is not the baseline, it's the legacy systems we've came from.
Still evolving the process but trying to reach the point where the following is consistent:
Use parallelism as a barometer.
If agents can't work independently, the problem is coupling in the architecture not the agents. Conflicts and worktree surgery are symptoms.
One slice, one context and agent.
Every agent task maps to a vertical slice inside a single bounded context covering domain logic, API and UI, tests, docs.
Cross-context edits are a smell.
When a task needs to touch files in several contexts, either the boundary is wrong or the task isn't parallelizable.
Slice vertically.
Cut work along small targeted features, not along controllers, services, and repositories.
Cross-context communication via contracts. Events, interfaces, and APIs, nothing implicit. Each agent is confined to a slice and dependency contracts not implementation.
Don't obsess over DRY.
Repeated low-level code is cheaper than the coupling that comes from sharing it.
Verify the footprint.
A judge agent enforces the principles and rejects work that strays outside the declared slice.
The more you code with AI, the more chaos and slop you get, the more architecture you need
You can't get great results out of coding agents without this skill
My new video explains why
@OpenAI@OpenAIDevs Seems to be good now. GLM 5.2 in @AmpCode worked well as a fallback . Plan to build this into my workflow not just as redundancy but as a broader multi-harness multi-model setup routing by the shape/complexity of the task.
Codex degraded right now, intermittent stream failures across both Codex Desktop and Pi agent, so doesn’t look harness-specific. Feels like an @OpenAI model/streaming or rate-limit issue. @OpenAIDevs
@MatthewBerman More an architectural issue than agent problem. Decoupled, cohesive design doesn't require the whole codebase in memory. Just the module under change and contracts with adjacent dependencies.