Agentic commerce
Agents now research, shortlist and increasingly transact with vendors for people. Each step is a trust decision, and at speed an agent needs what human buyers improvised: a signal about the other side it can check first.
The gap
The trust artifacts built for humans do not survive an agent's scrutiny:
- A questionnaire is self-reported and stale the moment it is filed.
- A row of customer logos is a claim, not evidence.
- A PDF report is a snapshot no machine can re-verify.
Unable to tell real posture from a polished page, an agent trusts too easily or hands the decision back to a human.
The signal an agent can act on
A vendor-trust signal that works for an agent has four properties:
- Witnessed. Read from live evidence, not self-reported: the agent weighs an observation, not a claim.
- Signed. Cryptographically verifiable, so the agent can confirm it independently instead of trusting a page.
- Queryable. Available in one call over a standard interface, so checking a vendor is a step in the agent's loop, not a scraping project.
- Honest about absence. With no data, the signal says so, and the agent treats the gap as a gap.
Trooth publishes that signal: a public Trust Profile per company, carrying a signed witness statement for each reading Trooth takes, re-checked on a schedule, queryable over the Model Context Protocol, honest about unknowns. Only the witness statement is signed, never the company's declarations. Agents can check it before acting and keep its provenance.
What this is not
A trust signal is not a verdict. Trooth signs what it witnessed; it does not warrant fitness for a purpose or replace the buyer's judgment.
The agent or its person decides if the witnessed state suffices, on dated evidence.
See how an agent uses it and the Trooth Network.