Session Naming Conventions for Running Many AI Coding Agents

The moment you go from one AI coding agent to four, a new failure mode appears: you approve a change thinking it's for repo A when the terminal was actually sitting in repo B. It's not a skill problem — it's an identity problem. Ten near-identical zsh windows all look the same. A naming convention fixes it for free.

Why identity is the first thing to solve

Broadcast, restore, and clever keybindings all matter, but they come after identity. If you can't tell at a glance which session is which, every other feature is aimed at the wrong target. Naming is the cheapest, highest-leverage habit in a multi-agent setup.

A convention that scales

1. Project first, task second. checkout-api · fix-refund-bug beats session 3. The project is the part that prevents cross-repo mistakes.
2. Encode the folder. If two sessions are in different worktrees of the same repo, put the branch or folder in the name. That's exactly where the dangerous mix-ups happen.
3. Color by project, not by status. Give each repo a stable color so your eye learns it. Status changes constantly; the project doesn't.
4. Keep it short and stable. A name you rename every ten minutes stops being a landmark.

Make the name do work

A good name isn't decoration — it's the thing you read before you type. The point is to make "which session is this?" a look, not a search. That's the reasoning behind how we tile sessions in Agent Tile: every session is a labeled tile with its project and folder visible at once, and each one relaunches with --resume in its original directory when you reopen the app.

For the broader review-accuracy angle, see reviewing AI-generated code across many agents.

Takeaway

Before you optimize typing speed or broadcast, make every session unambiguously identifiable: project-first names, folder encoded, stable colors. It costs nothing and prevents the single most expensive fleet mistake — acting confidently on the wrong repo.