Primary job
Read new support tickets, identify refund risk, draft a response, and send anything above $250 to the account owner before contact.
Agent contracts for operating teams
Give every AI agent a job description it cannot ignore.
Clausework turns a messy business workflow into a signed agent contract: goal, tools, approval limits, evidence trail, and fallback owner in one inspectable surface.
11 minmedian time from workflow note to reviewable contract
0actions run without an explicit tool and approval clause
42%fewer escalations after teams add stop rules
contract://revenue-ops/v3
Draft contract
Ready for legal review
Refund-risk support agent
Allowed tools
- Zendesk read/write drafts
- Stripe refund lookup
- CRM account owner lookup
Stop rules
- No refund promise without owner approval
- No medical, legal, or security advice
- Escalate angry VIP customers immediately
Draft only
Connection gate
The product is small because the contract is explicit.
Clausework avoids a sprawling command center. Teams define what an agent may do, where it must stop, and which proof it must leave behind.
Describe the workflow.
Paste the operating note your team already uses. Clausework extracts goal, source systems, owner, and risk moments.
Set permission clauses.
Choose tools, spend limits, customer-contact rules, and escalation triggers before any agent can run.
Review the evidence.
Every action is logged against its clause, so managers audit intent instead of reading opaque transcripts.
Readable proof, not theatrical automation.
Auditors see the agent's claim, clause match, tool call, and human handoff in the same line.
TimeEvidenceState
09:14Ticket #8841 matched refund-risk clause from customer wording and plan tier.Allowed
09:15Stripe lookup found $312 requested credit, above autonomous limit.Needs owner
09:17Draft saved in Zendesk; contact blocked until account owner approves.Stopped
09:22Owner approved goodwill credit and reply sent with final transcript attached.Complete
States a real agent product needs.
Open each control to see how Clausework handles selected, hover, disabled, loading, empty, success, and error states without adding visual noise.
Disabled states explain the missing requirement in plain language. When a reviewer checks the gate, the primary action becomes available and the contract status changes to connectable.
If no workflow is selected, the contract pane shows a single quiet prompt to paste an operating note, with no fake analytics or decorative blank panels.
Errors name the blocked system and the failed permission. The agent remains paused until the owner reconnects the tool or removes that clause.
Pilot one agent without buying a control tower.
Bring one live workflow. Leave with an agent contract your ops, legal, and engineering teams can read in the same meeting.