Comparison
| Fully autonomous agents | Terse | |
|---|---|---|
| Mode | Open-ended exploration | Deterministic workflow with model judgment |
| Best at | Broad research, loose tasks | CRM updates, routing, alerts, reporting |
| Write control | Agent decides | You decide (agent.tools.*) |
| Observability | Varies | Built-in run history, stats, approvals |
| AI role | End-to-end autonomy | Judgment step inside a controlled flow |
The Terse pattern
- Deterministic fetch — pull data from your CRM or enrichment API.
- Model judgment — summarize, score, or recommend.
- Deterministic write — update the CRM, post to Slack.
Choose an agent platform when
- the task is open-ended research or exploration
- the output does not need to land in a specific system every time
- a human supervises the session directly
Choose Terse when
- the workflow must update your CRM correctly every time
- you need reproducible, auditable behavior around writes
- you want AI inside a system you can test, review, and deploy