01
Research agent
Collect sources, extract facts, compare findings and prepare a brief with links for a human to approve.
AI agent development
Turn repeatable work into a system your team can inspect.
An AI agent should have a job, a source of truth, a bounded set of actions and a clear place for a human to review. That is the foundation I build around.
Typical systems
Start with the work that keeps coming back.
02
Content operations
Move from question discovery to brief, draft, internal links, schema checks and publishing review.
03
Internal assistant
Give a team a focused way to query approved documents, create summaries and route the next action.
Autonomy needs a boundary.
The work includes the boring but important parts: prompt and tool contracts, source restrictions, error handling, audit logs, evaluation examples and handover notes.
- Defined inputs and outputs
- Approval gates for consequential actions
- Evaluation set before launch
How an engagement works
- We map the current workflow and identify the repeated decision.
- We choose the data boundary and define what the agent must never invent.
- We prototype with real examples, then test failure cases.
- We document the system and hand over the code, prompts and operating notes.
Good fit: research-heavy teams, founder-led brands, content operations, documentation and internal knowledge work. A fully autonomous system is not promised where a review step is the responsible design.