Mavera DocsDelegating work: make it run without you
Programs, skills, autopilot, and trackers — the difference between using a tool and having a staff.
Delegate a bounded research program
Set the objective, cadence, hard cap, launch, and first-run inspection.
Choose Programs from the sidebar.
Good afternoon
Choose a surface from the sidebar to begin.
- 1Open ProgramsChoose Programs from the sidebar.
- 2Start a programCreate a long-running research program.
- 3Set the objectiveWrite the standing question the program should keep answering.
- 4Choose cadenceSet when the work should run.
- 5Set a hard capChoose a bounded credit limit for each run.
- 6Launch locallyStart the simulated program.
- 7Inspect the first runOpen run history, evidence, and the next scheduled pass.
The real unlock isn't asking questions — it's handing over jobs.
The delegation ladder
Level 1 — a question → chat. Seconds to minutes.
Level 2 — a task → "Run a pricing study with [audience] on [product]." One approval, minutes.
Level 3 — a chain → the Workflow builder: describe the goal, review the designed steps and cost, run the whole chain.
Level 4 — a project → a Program: "Build me a GTM blueprint for [product]." It plans, executes, reads its own findings, re-plans — for hours, with cited sources — and accumulates a structured report. Pause it, step it, steer it from the program page.
Level 5 — a standing order → Skills + Autopilot + Trackers (below).
Skills: playbooks you write once
Skills page → create a skill = a named procedure in plain language:
Name: Monthly message check When to use: First week of each month Playbook: Take our current homepage headline and the top competitor's. Run a 100-person A/B with the [core customers] audience. Compare against last month's result. Flag any shift over 5 points.
Click Run any time — the playbook lands in Mave's chat and executes. Your best researcher's process, bottled, runnable by the newest hire.
Autopilot: overnight research
Give Autopilot a budget of credits per night and standing questions. It runs while you sleep and the digest is waiting in the morning. Set it conservatively (a few hundred credits) and raise it when you trust the output.
Trackers: alerts when numbers move
Trackers watch a metric across runs — message performance, sentiment, competitor mentions — and alert you when it moves. The observer's answer to "tell me when something changes so I don't have to keep checking."
Delegate the verification too
End any delegated job with a verification clause in the prompt: "Before you present results, check them against the earlier study in this conversation and flag anything inconsistent." Fable 5 actually does this — self-checking is a real capability of the new reasoning generation, not theater.