Whitepaper
Acme Agents
An agent is only as useful as the work it can finish without supervision. This paper opens up the engineering behind Acme agents: how a planner breaks a request into steps, how each step picks a tool or a saved workflow, and how memory carries what was learned from one run to the next. It also covers how we grade agents with automated reviewers before release, and why the teams with the highest usage start by automating the five or six requests their colleagues send most.
Download this whitepaper to learn:
- Workflow matching: How each run starts by finding the closest saved workflow – one shipped with Acme or one your own team wrote last month.
- Common asks: The handful of request shapes that cover most daily work – status checks (“which renewals close this month?”), recaps (“sum up yesterday’s standup”) and small errands (“book a room for Thursday”).
- Automated review: How reviewer models score every release candidate against real test sets, so quality is checked before anyone sees a change.