Turn your team's work into a build-ready agent spec

We interview the people who run your process and pair what they tell us with your system context. You get a detailed agent specification your team can build and deploy.

Start a pilotSee a sample spec

We turn automation ideas into specs. A 10-day pilot gives you a ready-to-build agent spec.

Interview the team

Watch each task being done. Employees share their screen while an AI interviewer asks follow-up questions. Behind it, AI agents probe for deeper meanings and check answers for contradictions.Employees share screens. AI models probe for deeper meanings.

Each 30-minute interview runs in the browser. 30 to 40 employees take part.30 to 40 employees each take part.

Cross-check each step

Compare accounts across people. Many people describe the same step, and we compare their accounts to find where the written SOP and daily work differ.We find where written SOPs and daily work split.

Every account is cross referenced. Each step is checked against what several people say.Confirmed against what several people say.

Bottlenecks and hidden rules. The comparison surfaces SOP drift, bottlenecks, and rules nobody wrote down.We find SOP drift, slow spots, and unwritten practice.

Map the data

We map your data. With your IT team, we map lineage, owners, and a business glossary.With IT, we map lineage, owners, and a glossary.

Rows stay closed. We read metadata about the tables, and we never open the rows.We read table details, never the data.

We map owners and lineage. You see which system holds each figure, and who owns it.You see which system holds each figure.

Specify each agent

Every rule has a source. The spec covers every agent the workflow needs, with the steps, data, and identities for each one.The spec lists steps, data, and identities per agent.

Cited to interviews. Each rule in the spec cites the interview where people described it.Rules point to where people described them.

A named approver. Each agent gets a scoped service identity and a named approver.Each agent has a scoped identity and a sign-off.

Every rule traces back to an interview

The spec shows where practice drifts from the SOP. High-risk actions go to a named approver. The context layer is a living record that grows with each new process, tool, rule or hire, so the next spec starts from it.

Accounts payableAgent specsInvoice matching ShareExport

Invoice matching agent

Finds the PO for each new invoice, checks number, amount and quantity, and sends exceptions to the controller with the evidence.

CoversSteps 3 to 5
Time today31 hours a week
Systems
ApproverTRTheo R.

Rules from the interviews

  1. Match on PO number, amount and quantity. Allow 2% on the amount.

    NKNora K. 18:05+6

  2. Send any invoice over $25,000 to the controller before it posts.

    TRTheo R. 09:41+11

  3. If the PO changed after the invoice date, match against the latest version.

    SMSam M. 12:20+3

Identity and access

Service identitysvc-ap-match, its own account
ReadsSAP MM purchase orders, vendor master, the AP inbox
WritesMatch status in the PO tracker
NeverPosts payments or changes vendor bank details

Human approval

TRTheo R., ControllerPostings over $25,000 and any vendor bank change wait for Theo.GrantReviewReverse
SourcesRules cite the interviews they came from
IdentityEach agent has a scoped service identity
ApprovalA named human approves high-risk actions

You get a build-ready agent spec

  • Hear from the people who do each step.
  • Spot unwritten rules and where the SOP differs.
  • Agree on each term's meaning, and its source.
  • Your team builds and deploys from the spec.
  • Every high-risk action has a named approver.

Book a Pilot

Tell us your function and team size.

Thanks. We have your request and will be in touch.