The same path on run 1,000
The steps are code you have reviewed, so the same input takes the same route. Only the judgment steps call a model.
Describe the work you want an agent to take off your plate. Whenever builds it: code does the steps you already know, a model handles the ones that need judgment, and it acts through any of 1,400+ tools. Your team sees every step before it runs.
The same steps for support, sales, operations or reporting.
What starts it, which tools it may use, what good looks like, and when a person decides.
The map shows every step and which of them ask a model. Test it on real inputs.
A schedule, a webhook from another app, or each message in Slack or Telegram.
Real briefs for the same kind of job. Pick one and Whenever writes it against your accounts; the code is yours to change from there.
The questions that come after the demo.
The steps are code you have reviewed, so the same input takes the same route. Only the judgment steps call a model.
Billed per run, however many tools it uses. Model tokens are spent only in the steps that ask a model.
The agent is a file you can read, export and hand to a developer, not a configuration inside one vendor.
Take over repeatable work that needs a little judgment: sorting support requests, qualifying leads, drafting follow-ups, summarising reports, and acting in the tools where that work lives.
AI building is free within weekly usage limits. A published agent is billed per run, and model tokens draw on the separate runtime AI allowance each plan includes before they are billed on top.
No. Describe the job in plain words and test it on real inputs. The code is there for a developer to review or export when you want one to.
Sources reviewed September 30, 2026.
Describe it, test it on real inputs, and put it on a trigger. AI building is free within weekly usage limits.