AI agent build project template
Build and ship an AI agent for a client: use cases, data and tools, prompts and evals, guardrails, rollout and cost tracking.
A project for building an AI agent or assistant for a client, from the first use case to a monitored rollout. It puts evaluation and guardrails before launch, and keeps model costs visible so you can bill them.
Log the agent's runs in eBRBe's agent-cost ledger during the build and after launch, and bill them with your markup next to the hours.
What's in it
Use cases and success measures
Day 1–5- Workshop: tasks the agent should handle4h
- Success measures and test setChecklist: Collect 30+ real examples · Define what a good answer is · Agree acceptance thresholds6h
- Data access and privacy reviewChecklist: What data the agent may read · Where it is processed · Retention and logging3h
Tools and data
Day 6–15- Connect data sources10h
- Build tool functions14h
- Retrieval and context8h
Prompts and evaluation
Day 13–23- System prompt and instructions6h
- Evaluation harness against the test set8h
- Iterate to the acceptance thresholds12h
- Model choice and cost per runChecklist: Compare models on the test set · Record cost per run · Agree the markup with the client3h
Guardrails
Day 19–27- Refusals, limits and human handoff6h
- Prompt-injection and data-leak tests5h
- Guardrail review with client1h
Rollout and monitoring
Day 28–36- Pilot with a small group4h
- Monitoring, logging and cost tracking4h
- Launch2h
- Handover, runbook and training4h