AI agents across the team

Play this lesson

Agents at work

Remy hands a small bug fix to an AI agent before lunch. Course 1 built with Claude Code; in a bigger studio, AI agents help every team: code, tests, translations, support and data.

Agents in the cloud

GitHub's Copilot cloud agent "can research a repository, plan changes, and implement them in the background". GitHub's advice: "Review and test the output before using it in production."

Quick check

Remy's agent opens a pull request that fixes the bug. What happens next?

  1. A person reviews and tests it, like any change
  2. It merges by itself
  3. It ships straight to players
Show the answer

A person reviews and tests it, like any change

Agents' work goes through the same review as everyone's.

Where agents help

Pick the job

Which job should stay with a person?

  1. Deciding to ban a player
  2. Sorting bug reports by topic
  3. Drafting release notes
Show the answer

Deciding to ban a player

A ban affects a real player. An agent can gather the facts; a person decides.

Limits on time and reach

GitHub lists its agent's limits: it "can only make changes in the repository specified when you start a task", and each session has "a maximum execution time of 59 minutes". Shard Garden sets limits like these for its own agents.

Quick check

An AI agent should have access to every system, so it can help anywhere.

Show the answer

False

The Trust and safety stage: each agent gets only what its job needs.

Measure it

The studio tracks what agents do: how many of their pull requests merge, how many need fixing, and how much time people save. An agent that doesn't help is switched off.

The youngest players

Many of Shard Garden's players are children. AI features for them need the most care.

Bolt says: Next: AI and children's safety.

Fact-checked 12 October 2026