Who actually passed your AI training assessment?

Ruben Hassid reports that he used Codex to complete five introductory AI courses and collect the credentials without answering the questions himself. I think this raises a practical question for a company paying to train its employees: who demonstrated the knowledge that the completion report is supposed to represent?

A course-completion target is easy to put into a management report. The record can show that the required activities were completed. If those activities can be delegated to an agent, the report may leave the participant's understanding unexamined. That changes how much confidence I would put in it when deciding whether someone is ready for a responsibility.

The credential needs a closer look before drawing that conclusion. Google's credential overview distinguishes certificates, skill badges and role-based certifications. Its Cloud Digital Leader standard exam uses multiple-choice and multiple-select questions and is taken with online or onsite proctoring. The assessment conditions matter. Hassid's account of introductory course completion cannot establish that the same thing applies to a supervised certification exam.

I would want to know what the person had to demonstrate, how their contribution was established, and what the result allows me to infer. A badge attached to an account identifies the recorded achievement. The assessment has to provide the evidence about the person's knowledge. Calling everything an AI certification makes those different claims sound interchangeable.

Someone can also learn a great deal from a course while earning a modest completion badge. The lessons may introduce useful concepts, offer a place to practise and help them make sense of their work. Those benefits depend on what the participant does with the material. I would keep them in view while examining how the programme assesses learning.

When buying training, I would begin with the responsibility we want people to carry afterwards. Perhaps employees should be able to use AI to prepare a project plan and review it before making a commitment to a customer. That is a more useful specification than asking for a course that ends with a certificate we can put on LinkedIn.

The supplier should be able to explain how the participant will demonstrate that ability. I would ask to see the assessment task and what the assessor looks for. A finished plan could be coherent because the participant understood the assignment and checked it carefully, or because the tool produced something convincing. The assessment needs to help us tell which contribution the person made.

In a hypothetical exercise, I would give the participant a project brief and let them use AI to prepare a delivery plan. Then we could discuss why particular tasks are included, which assumption affects the deadline and what remains unresolved. They could show the brief, the work with the agent and the revisions they chose to make.

Changing a condition would make that discussion more useful. Suppose the customer needs the first delivery earlier. What can move, and what would the team have to agree before promising it? The participant can work through the change with AI while the assessor watches how they interpret the answer and decide what to keep. That gives us evidence about their judgement during the work.

For an assessment of AI-assisted work, using the tool belongs in the exercise. I want the person's contribution to be visible: how they explain the purpose, recognise a weak recommendation and take responsibility for the result. Someone who can do that well may be very useful even if they delegate much of the first draft.

A company can then report course completion alongside the responsibilities people have demonstrated they can carry. There may still be tasks they need help with. Describing those limits gives a manager something useful when allocating work and deciding what further training would help. A badge count leaves that decision much less informed.

Before buying an AI training programme, I would ask the supplier to show how its assessment connects the credential to the participant's own contribution. After the programme, I would want to be able to name the work that person is ready to take responsibility for. That is the result I would use when deciding whether the training has done its job.