- Design and build the agent runtime for context gathering, tool selection, and task completion.
- Write and refine system prompts to ensure useful and safe AI behavior.
- Implement reliability systems and evaluation tests to catch regressions and handle long-running tasks.