Learning Rust after years of TypeScript

Solo buildersExperienced web developer learning a new language

We started a small CLI in Rust to learn it. The agent and we kept falling into the same borrow-checker traps: Rc where threads were involved, and cloning strings to quiet errors. Each new session repeated the lesson.

How do I stop my AI agent and me from making the same Rust borrow checker mistakes?

With ArcRoom, the agent saves a failed approach with remember_failure and why, and you save lessons with remember, such as 'shared state across threads: Arc<Mutex<T>>, not Rc'. Recall hands the matching lesson to a prompt about the same problem, and arcroom rule --idle-days drops lessons you no longer hit. It is not a tutor: being handed the answer can skip the learning.

Also asked as: My agent keeps reaching for Rc in threaded Rust code. How do I make the lesson stick? · How do I keep a list of language lessons my AI agent will see? · Can an AI agent remember compile errors it already learned from?

The work

When an approach fails to compile for a reason worth knowing, the agent saves it with remember_failure and the why.

ArcRoom keeps
  • Saved decision
  • Recall
  • Expiry rule
The next agent gets

Tried before and failed (Oct 14): Rc<RefCell<Config>> shared with a spawned thread.

Step by step

  1. When an approach fails to compile for a reason worth knowing, the agent saves it with remember_failure and the why.
  2. Lessons we want kept go in with remember: 'shared state across threads: Arc<Mutex<T>>, not Rc'.
  3. Recall hands the matching lesson to a prompt about the same problem.
  4. After a few weeks, arcroom rule --idle-days drops lessons we no longer hit.

What the agent sees

Tried before and failed (Oct 14): Rc<RefCell<Config>> shared with a spawned thread.
Why: Rc is not Send. Use Arc<Mutex<Config>>.

Compared with today's tools

A personal notes file or Claude Code auto memory

Keep a lessons.md and let Claude Code's own memory pick up the rest.

Setup
10 min
Ongoing
write each lesson down by hand, ~3 min

ArcRoom

Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.

Setup
2 min (npx arcroom init)
Ongoing
None

Notes you write yourself teach you more; ArcRoom's edge is that the lessons follow you into Codex or Cursor too.

Where it stops

ArcRoom is not a tutor. If the point is to learn, being handed the answer can skip the learning.

Try it in your project

Works with the agents your team already uses, and any agent over MCP or the HTTP API.

More in solo builders