Running Zed agents in parallel

[vicpara]

Body

What are you proposing?

I want to run multiple Zed agents in parallel, switch between them to see their status. One Zed agent might take 10m-20m+ and I cannot just fiddle waiting for it to complete. LLM uptime and load may get Zed to take even longer than usual to complete the task while navigating API errors.

Why does this matter?

It's a waste of time waiting for 1 Zed agent to complete the work.

What problem does this solve?

Time, productivity and efficiency.

What becomes easier or possible?
I can stay in flow and get more work done instead of waiting for Zed to complete one task.

Are there any examples or context?

No, you got this.

Possible approach

Use branches and work trees to keep Zed agents not clashing. I DON'T WANT TO USE CLAUDE CODE OR CODEX for this. Otherwise I'd be doing that.

[Nyrok]

Worktrees are the right approach for filesystem isolation. The harder part of parallelizing agents is scoping each one correctly.

Without explicit constraints per agent, even isolated branches can produce contradictory decisions. One agent refactors a shared utility, another builds on the old version. When you merge, the conflict is conceptual not just textual.

A tight role + objective + constraints block per agent (what it should do, what it should NOT touch) reduces that problem a lot. I've been building flompt for exactly this, a visual prompt builder that decomposes prompts into 12 semantic blocks and compiles to Claude-optimized XML. Worth structuring per-agent instructions before scaling up parallelism. Open-source: github.com/Nyrok/flompt

[vicpara]

This spammy non informative message is useless and an embarrassing way to advertise your company. it doesn't belong here.