Conversation flow

A typical session with Andy moves through three phases, ask, plan, hand off, all in a single Google Chat thread. You don't have to follow this exact arc; you can stop after any phase.

Phase 1: Ask

Mention Andy with a question or request:

@MeetAndy how does the password reset flow work in this codebase?

Andy reads the relevant code in the Space's repositories and replies in a thread with an explanation. You can keep asking follow-up questions in the same thread.

For pure questions, this is the whole flow. Andy answered, you read the answer, you're done.

Phase 2: Plan

When you ask Andy to do something, like change code, add a feature, or fix a bug, Andy doesn't start coding right away. Instead, it produces a plan for you to review.

@MeetAndy add a dark mode toggle to the settings page

Andy replies in the thread with a structured plan message that looks roughly like this:

🤖 AI Plan Ready for Review

Task: Add dark mode toggle to settings page
Complexity: 🟢 Low

[summary of the planned changes]

[ 📄 View detailed plan ]

[ ✅ Approve ]   [ ✏️ Make changes ]   [ ✓ Mark as done ]

Three buttons:

Approve: Andy starts implementing the plan (Phase 3)
Make changes: you reply in the thread with what you want changed; Andy updates the plan
Mark as done: you decide not to implement; the conversation ends here

You can also reply in the thread without clicking any button, for example, "actually, also include a system-default option", and Andy will refine the plan.

You don't have to use the buttons to close a task either. Just tell Andy in the thread, for example "mark this as done" or "close this task", and Andy recognizes the request, closes the task, and posts the usual feedback survey.

Phase 3: Hand off implementation

When you click Approve, Andy hands the plan to the Coding Agent, which actually writes the code. Andy posts a follow-up:

Working on it…

[ ⛔ Stop ]

Andy then updates the thread as the Coding Agent works, typically a few short status messages over several minutes. You can stop the run at any time with the Stop button.

When the Coding Agent is done, Andy posts:

Implementation complete

Task: Add dark mode toggle to settings page
Branch: [View branch]

[ 📄 View implementation details ]

[ 🚀 Create pull request ]   [ ✏️ Make changes ]   [ 🗑️ Discard and exit ]

Three options:

Create pull request (or Create merge request for GitLab): Andy opens a PR/MR with the work
Make changes: reply in the thread with revisions; Andy iterates on the implementation
Discard and exit: throw away the branch; nothing is opened

See Reviewing Andy's work for what to do once the PR is open.

Stopping at any point

You don't have to go all the way to a PR. Common patterns:

Just ask: get an explanation, learn the codebase, no plan needed
Just plan: get Andy to think through an approach, then implement it yourself or assign to a teammate
Plan and implement: let Andy ship the change end-to-end

Tips

Be specific in the first mention. "Add dark mode" gets a generic plan; "Add dark mode to the settings page using our existing theme tokens, defaulting to system preference" gets a much better one.
Reply in the thread to refine. Even after approving a plan, you can reply mid-implementation with "actually, also do X", and Andy will adjust.
Use memory to teach Andy facts about your project so you don't have to repeat them every time.