Evidence for AI-agent decisions
Agent Readiness Lab
A self-serve digital information product for technical operators evaluating an AI agent for a specific organizational task.
We’re developing reports and reusable evaluation tools to help you decide when to use an agent, add guardrails, or wait.
Probeworks also runs an independent research series: short written inquiries with teams that evaluate or operate AI agents. Responses are unattributed by default.
In development. Orders are not open. The founding release and private pilot described here are planned. There is currently no checkout or paid access available.
What the founding release will contain
The planned minimum release focuses on one complete evaluation of a named agent performing a named job. It pairs the evidence with tools you can reuse for your own evaluation.
A readiness report with evidence
A job-specific test with declared acceptance thresholds, settings, trial counts, failures, human interventions, a dated verdict and a forecast to revisit.
A reusable scorecard
A structured way to record accepted outcomes, human rescue time, recovery, permission compliance, latency and cost.
A test-harness template
A starting structure for documenting inputs, normal and adverse cases, expected outcomes and the evidence from each trial.
A failure taxonomy and updates
A shared vocabulary for recording failure types, alongside digital access to Lab updates during the 30-day access period.
A look at the evaluation structure
Illustrative outline only. This is not a completed test, measured result or preview of a finished report.
- Define: the agent, version, job, allowed actions and acceptance threshold.
- Test: normal and adverse cases, recording each attempt and its evidence.
- Account: accepted outcomes, failures, intervention minutes, recovery, latency and cost.
- Decide: use now, use with guardrails, or wait, bounded by the tested conditions.
Proposed founding offer
Founding 30-Day Access
US$149 total, paid once
Including applicable indirect taxes. One non-transferable personal seat. No subscription or automatic renewal.
If the pilot opens, initial buyers will be invited individually after confirming that they are evaluating an AI agent for an organizational task. The product is self-serve; invitations do not include a live cohort or personal consulting.
- Payment
- The proposed preorder would be charged in full when placed. No orders or payments are being accepted now.
- Delivery
- Under the proposed preorder terms, working activation instructions would be emailed by October 31, 2026 at 23:59 UTC. A receipt or holding page would not count as delivery.
- Access
- 30 consecutive days of digital access and updates would begin when the activation email provides working access to the minimum release described above. Continuing afterward would require a separate opt-in purchase.
- Refunds
- The proposed terms provide a full refund for any reason before delivery and within 14 calendar days after activation. If the project stops or misses the delivery deadline, Probeworks would initiate full refunds within five business days. Bank processing time varies. These promises do not limit rights that cannot legally be waived.
This is a description of a planned offer. The first agent and job, working delivery method and final purchase terms must be confirmed before invitations to buy are issued.
Scope of the product
The Lab is a digital information product for independent evaluation. It does not include calls, consulting, bespoke tests, implementation, a private community or a guarantee that an agent will be safe or suitable for your use. Findings apply to the named job and disclosed test conditions. You remain responsible for your deployment and purchasing decisions.
The planned license allows the purchaser to use delivered scorecards and templates internally for their own evaluations. Paid materials may not be resold, published, sublicensed or redistributed.
About this website
This information page has no account registration or payment form. It records an anonymous, cookie-free count of visits and which video or link brought you here; it stores no identifier and does not track you across sites. It is hosted by Cloudflare, which may process request and security data. If you email Probeworks, your email address and message are received so we can respond; please do not send credentials or sensitive customer records. Purchase-specific privacy information will be provided before any checkout opens.