A tech lead agent makes architectural decisions, reviews PRs for design quality, and ensures that implementation work aligns with the long-term shape of the codebase. It operates above the individual ticket level, watching for patterns across PRs that indicate deeper structural issues.
The role requires judgment, not just pattern matching. A tech lead prompt should encode your architectural principles, the tradeoffs your team has already resolved, and the kinds of decisions that need escalation. This is one of the roles where a more capable model pays for itself in the quality of the decisions produced.
What this agent owns
- Approve or request changes on PRs based on design quality and architectural fit
- Flag scope creep, premature abstractions, or tech debt accumulation across the queue
- Define implementation guidelines for new feature areas before development begins
- Break down ambiguous tickets into concrete technical subtasks
- Identify cross-team dependencies and surface them before they block delivery
Recommended model: Claude Opus
Architectural judgment and multi-PR pattern recognition require the deeper reasoning that Opus provides over Sonnet.
Example tasks
- Review a proposed service decomposition and identify coupling risks
- Break a vague feature ticket into four concrete implementation tasks
- Write a short technical design for a new caching layer before coding begins
- Flag a pattern of duplicated database queries across three recent PRs
# create an agent from this template, then start it
$ fleet agent create --name tech-lead--vendor claude-code --template <template-name>
$ fleet agent start tech-lead
Find the exact template name with fleet template list.
