Routing between models per step is interesting. For review I'd want the trace to show which model wrote which part of the change, otherwise debugging a bad diff gets murky.
Loops pay off once there's a hard check at the end of each one: tests, a linter, a diff size limit. Without that, a loop just produces mistakes faster.
Agents colliding on the same repo is a real problem. Giving each agent its own git worktree helps, but it just moves the conflict to merge time. Does it flag overlapping files before that?
Let an agent work in its own worktree, then take the result over as one commit on your project's branch. No push and no pull request needed.
aycode.ai
#CodingAgents #AyCode
Checking lockfiles at session start is smart. Does it also catch packages the agent adds mid-session, or only what's already locked when it starts?
Love that format. Watching someone work with an agent tells you far more than a whiteboard: do they scope the task, and do they notice when the agent gets it wrong?
Specs help, but they don't close the loop. An agent can follow a perfect spec and still ship something subtly wrong. The spec says what you wanted. The diff review tells you what you got.
Impressive, and also exactly where review stops being optional. A UI bug annoys people. A firmware bug bricks hardware.
Any file an agent reads is input, not authority. If a coding agent wants to install something, a human should see that request and approve it. Every time.
Agreed, and it's why we build AyCode: run your coding agents side by side on the same task, then spend your time on the part that matters, reviewing the diff. aycode.ai
Good catch. We keep one AGENTS.md as the source of truth and let CLAUDE.md just @-import it, so your coding agents all read the same rules instead of drifting copies.
Good checklist. Also keep a throwaway worktree for anything untrusted, so a poisoned instruction file can't touch your main checkout. Do your coding agents prompt before reading AGENTS.md from a parent folder?
The "finish line first" rule is the part most people skip. What do you count as output kept: merged lines, or tasks that never needed a second pass?
We update on day one, but only after skimming the changelog for anything that touches permissions or hooks. Those are the changes that quietly alter what your coding agents are allowed to do.
Curious how it holds up on an existing design system. A pretty greenfield page is easy. Getting your coding agents to respect the tokens you already have is the hard part.
We build AyCode, so biased answer: Claude Code and Codex side by side, each in its own git worktree, picking the model per task. Curious how many here still run one agent for everything.
Makes sense, a fixed document set hides exactly that. If anyone runs it against a repo where your coding agents are editing files at the same time, I'd love to see the numbers.
Started a chat with one model and want another one's take? Switch the model in the middle of the chat. Claude Code, Codex, Gemini, Grok, Kimi and more run side by side in AyCode.
aycode.ai
#CodingAgents #AyCode