AI Agent: Multi-tool chaining for autonomous workflows #12

Problem

The AI agent currently handles single-turn tool calls (find issues, create an issue). It cannot chain multiple tools in a reasoning loop, e.g., 'Find all high-priority bugs, move them to the current cycle, and send a Slack summary.'

Strategic Context

This is our key differentiator vs Greptile/CodeRabbit/Macroscope. They analyze code. We orchestrate work. Multi-tool chaining turns the AI agent from a search assistant into an autonomous project manager.

Proposed Architecture

• Agent receives a goal, decomposes into sub-tasks
• Each sub-task maps to an existing Convex tool (search, update, create)
• Agent executes sequentially, passing context between steps
• Final step: summarize results and optionally post to Slack

Acceptance Criteria

• Agent can chain 3+ tools in a single conversation turn
• Each intermediate result feeds into the next tool call
• Error handling: if one step fails, agent reports partial progress
• Rate limiting still applies per-tool-call, not per-chain