Questions and answers.

It takes you from the notes of a client meeting to a decision about AI. You get the workflow as a step map, with every task labeled Automate, AI-assisted, Agent, or Keep human, each with a reason tied to your notes. It recommends the one agent worth building first, or says plainly that none fits. For that agent, it writes a design document: goal, trigger, systems, what it may do alone and what needs approval, guardrails, and test cases. Everything exports to a PDF for the client.

Consultants who bring AI into client work, whether the focus is operations, change, technology, or learning. It is built for the conversation with the client: what to ask, what was said, and what to recommend.

No. Type notes during the meeting, paste your notes or a transcript afterward, or describe the workflow in your own words. Meeting mode suggests what to ask next while you type. If you already record with another tool, paste the transcript in.

Before the analysis runs, the Clarifier lists what your notes settle and what they do not. Then it asks one question at a time, each from your last answer, until the facts that decide the recommendation are settled. You pick a suggested answer or type your own, and say how you know it: the client said it, the client was unsure, or it is your assumption. You can analyze at any point. Anything still open becomes a question for the client.

AgentStepper says so, and recommends the cheaper option: which steps to automate, and where AI can help a person. An honest "no agent here" is a useful answer for a client.

One agent is easier to prove. Each agent needs its own guardrails, approvals, and tests, and agents that hand work to each other add new ways to fail. The next agent comes after the first one runs reliably. The other steps keep their own labels in the step map.

Yes. Edit any step, change its label, or mark it as wrong, then regenerate the design document. Your edits always win. After a follow-up meeting, AgentStepper suggests changes from the new notes, and nothing changes until you accept them.

No. It decides where an agent belongs and designs it. The design document is what a technical team builds from.

Those tools document how the work runs. AgentStepper starts from what you learned and answers the next question: where AI belongs, which agent to build first, and how to keep people in charge of it. You can paste notes and transcripts from any tool you already use.

It is built not to. Systems, numbers, and people come from your notes, or they are marked as assumptions and listed for the client to confirm. Every step and design item points to the words it came from. Checks run in code after every answer, and anything that fails them is removed.

No. AgentStepper uses Claude, made by Anthropic, under commercial terms that do not allow Anthropic to train its models on your content. Personal names are replaced before your text is sent, and AgentStepper's own logs never contain your text. Security and data has the details.

Only you. No other AgentStepper user can see your work, and sign-up is by invitation. Your client sees what you choose to share, usually the PDF.

Yes. Deleting an analysis removes all of it. A copy can remain in the database's daily backups for up to 7 days, and then it is gone.

AgentStepper is in early access, by invitation. Pricing for consultants and firms will be set with the first users. Ask about access.

Write to chuck@perpetuoso.com, or use Ask about access on the Pricing page. I read every request myself.