Test the team before you launch the agent. A readiness assessment, where the people who will work alongside an AI agent are tested on how it decides before it goes live, costs a few hours. A failed go-live costs weeks of rework, a backlog of bad decisions, lost trust and sometimes a regulator's attention. The comparison is not close, yet most organisations skip the assessment.
What does a failed go-live actually look like?
Rarely a dramatic crash. More often a slow slide. The agent launches, held items pile up because reviewers do not understand why actions are being held, and people start approving in bulk to clear the queue. Refusals are overridden because customers complain. A certificate lapses, or a run that clears no depth drops the agent to depth 1 the same day, drafts flood the queue, and the team assumes the system is broken.
Within weeks, leaders hear that the agent "doesn't work". The project is paused, re-scoped or quietly shelved. The record tells a different story. Every verdict and every human decision is filed in an append-only record, and it usually shows an agent behaving as designed, surrounded by people who were never taught what that design meant.
What does a readiness assessment measure?
Understanding of the concepts that decide whether the team can oversee an agent, tested before launch:
- Depth. Do people know what read, draft, execute and money-moving actions mean, and that depth is earned, not configured?
- Verdicts. Can they tell a held action from a refusal and a withheld action, and do they know what to do with each?
- Provenance. Can they say whether a fact is sealed, signed, declared or external, and why an external claim cannot authorise a payment?
- Certificates. Do they know certificates lapse and what a drop to depth 1 looks like?
- The record. Do they understand that their decisions are filed and will be read?
The output is not a pass or fail for individuals. It is a picture of the team: which concepts are understood, where the gaps are, and which agent and capability would be the safest to start with. Go-live readiness covers the wider set of conditions, of which people are one.
Why do organisations skip it?
Because it feels like a delay. The agent is ready, the business case is approved, and someone suggests testing the team first. Nobody wants to be the person who slowed the launch. And because training is often scheduled for the week before go-live, there is no time to act on what an assessment would find.
There is also a quieter reason: fear of the result. An assessment might show that the reviewers do not understand depth, or that the team lead cannot explain a refusal. That is uncomfortable to report upward. But it is far less uncomfortable before launch than after.
Some organisations also assume readiness can be judged by asking. Managers are asked whether their teams are ready, and they say yes, because nobody wants to admit otherwise and because they believe it. A test set in the team's own industry replaces that belief with evidence, concept by concept.
What is the trade-off?
Time and attention, mostly. An assessment takes a few hours of each person's time, plus someone to read the results and act on them. If the gaps are large, launch may move by a few weeks while training catches up. That is a real cost, and in some businesses a few weeks matters.
Set against it: a go-live that fails costs the same few weeks at least, plus the decisions made badly in the meantime, plus the credibility of the next attempt. Teams that have lived through a failed launch become sceptical, and the second go-live has to overcome that. A readiness assessment also tells you where to start, which itself saves time: launching first with the capability the team understands best gets evidence flowing sooner.
How does it fit into the go-live plan?
Early enough to act on. Run the assessment when the agent is chosen, not when it is ready. Use the results to shape training, then schedule that training around the go-live, as described in how AI agent training should be scheduled around a go-live. Re-check the weakest concepts before launch, and include anyone who joined the team since the first assessment.
Sector guides show how this works in practice: how to run a readiness assessment in a contact centre and in a utility. After launch, a retrospective closes the loop; the training retrospective template helps.
What should you do differently?
Make a readiness assessment a condition of go-live, like testing the agent itself. Run it as soon as the agent is chosen. Read the results by concept, not by person, and fix the gaps before launch. Choose the first capability based on what the team understands best. And report the results upward without softening them: a known gap with a plan is a much better position than an unknown one discovered by customers. The wider approach is in training teams for AI agents.
Where the Academy fits
Pryme Intelligence Academy offers a free readiness assessment for teams of up to 25 people. The team takes the Foundations test set in its industry, and the report shows scores by concept, the gaps and the first agent to start with. The Foundations track then fills those gaps in about four hours.
Questions people ask
Who should take the readiness assessment?
Everyone who will review, approve or oversee the agent's work, plus their managers. Leaders benefit too, because they make the depth and limit decisions.
What if the assessment shows the team is not ready?
Train the weakest concepts first, start with the simplest capability at depth 2, and re-assess. A delayed launch with a plan is better than a launch that stalls.
Is a readiness assessment the same as testing the agent?
No. Testing the agent shows what it can do on your cases. A readiness assessment shows whether your people can oversee it. You need both.