# How to Run 10 Side Projects in Parallel with AI Agents

> A practical system for managing many repos with AI coding agents: one screen for status, a morning routine, session hygiene and clean git everywhere.

- URL: https://shotmatic.app/blog/run-10-projects-in-parallel
- Category: Multi-Project
- Published: 2026-09-27
- Author: Tu Nguyen

## Key takeaways

- With agents doing the typing, one person can move 5–10 projects forward — the limit becomes how much status you can hold in your head.
- Move that status out of your head: one line per task, per repo, on one screen, built from files in each repo.
- Start the day with a fixed routine: look at stuck work, push yesterday's leftovers, then resume sessions one by one.
- Let agents run in parallel, but review and release one project at a time.

An AI agent can keep five projects moving while you sleep on none of them. What it can't do is remember,
for you, that the game is stuck on an ad SDK, the book is waiting on a cover, and the app repo has fourteen
uncommitted files from Thursday.

This post is about running many projects in parallel without that turning into chaos. It is part 4 of
*The ShotMatic Method*; it builds on [state files](https://shotmatic.app/blog/memory-lives-in-files) and
[phases](https://shotmatic.app/blog/phases-not-todo-lists).

## The new constraint is status, not speed

With one project, status lives in your head for free. With ten, you pay for it every time you switch:

- *Which repo was doing what?*
- *Which one has been quietly stuck since Tuesday?*
- *Which Claude session had this task?*
- *Did I push that?*

Each question costs a minute or two. Asked for ten repos, every morning, that is where your first half hour
goes. The fix is not a better memory — it is **not needing one**.

## One screen, one line per task

Build (or use) a single view where every active task from every repo appears as one line:

```text
GAME   Dino Buddies    T11 ads boost          Phase 8/14   ● stuck
BOOK   Dog Training    write chapter 7        Phase 5/9
APP    Tip Split       store listing          Phase 4/6
APP    Piston          waiting for API key    Phase 3/6    ● stuck
```

Three properties make this view work:

1. **It is generated from the repos**, not typed into a separate board. If the agent updates the state
   file, the line updates. Nothing to keep in sync by hand.
2. **Stuck work sorts first.** A blocker or a week of silence pushes the line to the top.
3. **It fits on one screen.** If you have to scroll to see all your projects, you will stop looking at the
   bottom half.

## A 5-minute morning routine

Run the same routine every day, in the same order:

1. **Look at what is stuck.** Unblock it, or explicitly park it in the backlog.
2. **Clean up git.** Push yesterday's leftovers in every dirty repo so nothing lives only on this laptop.
3. **Pick at most three projects** to actively move today.
4. **Resume their sessions** one by one, each starting with *read the state file*.
5. **Leave the rest waiting** — they are not forgotten, they are on the screen.

The point is that the routine is mechanical. You make decisions in step 1 and step 3; the rest is clicking.

## Session hygiene across many repos

Parallel agents are where context gets messy. A few habits keep it clean:

- **One task, one session.** Don't let one conversation wander across projects.
- **Write the session id next to the task**, or you will spend minutes hunting through recent sessions.
- **Don't trust a session you can't find** — start a new one with the state file instead. Nothing important
  should be only in the chat.
- **On a second machine**, bring the sessions with you or accept starting fresh; the state files make
  either safe.

## Git across ten repos

Uncommitted work is the most common way multi-project setups lose work. Keep two separate habits:

- **Marking a task done commits only the status change** — a small, precise commit you can trust.
- **"Just push it" is a separate, deliberate action**: review the list of files, keep a sensible message,
  push. Doing this for every dirty repo once a day is enough.

Never let a tool rebase or force-push for you across many repos. When git fails, read the real error.

## Parallel agents, serial reviews

Agents can run in parallel. **Your review should not.** Reviewing two diffs from two projects at once is how
wrong changes slip through. Let the agents finish, then review and release one project at a time — the
[small-version rhythm](https://shotmatic.app/blog/ship-small-versions) makes each review short.

## Where ShotMatic fits

This is the job ShotMatic was built for: up to ten boards on one screen, one line per task with its phase
and next step, stuck work floating to the top, a Master list of every task from every repo, a GIT strip that
shows which repos are dirty, and ✦ to resume the exact Claude session for any line. It reads the files your
agents already write, runs locally, and needs no account.

## FAQ

**Is it realistic to work on 10 projects at once?**

Only if most of them are waiting at any moment — on an agent, a review, a store, or you. The goal is not to type in ten places at once but to never lose track of where each project stands.

**How do I know which Claude session belongs to which project?**

Record the session id next to the task, or name sessions after tasks. ShotMatic stores the mapping locally so the resume button always opens the right conversation.

**What is the biggest risk when running many repos?**

Silent stalls and uncommitted work. A project that has not moved for a week, or a repo with changes that were never pushed, is easy to miss when you have ten of them.