- A correct output does not prove its author has acquired the skill.
- Alternate demonstration, supported practice and independent explanation.
- Assess learning through new, role-specific situations.
What the Anthropic experiment invites us to examine
Research published by Anthropic on 29 January 2026 examines the effect of AI assistance on learning programming skills. Within this experimental setting, the way people interact with the tool matters for learning. The task is bounded; conclusions should not automatically be extended to every profession or to performance over several months.
For corporate training, this raises a useful question: did the participant merely obtain an answer, or can they explain why it is appropriate? We suggest organising training around that distinction. It helps develop AI usage that strengthens judgement.
Define an observable skill
“Knowing how to use AI” is too vague to support an assessment. Describe an action: compare two proposals against explicit criteria, produce a faithful summary of a dossier, or verify that a code change addresses the requirement. The skill includes verification, not only writing a prompt.
Then define what makes the result acceptable. For a summary, this might include fidelity to sources, separation of facts from assumptions and coverage of essential information. For analysis, add the ability to explain criteria and identify missing data.
Design a four-part workshop
Start with a narrated demonstration. The trainer explains the context provided, choices made and checks performed. They also show an inadequate result and how to correct it. This makes visible the reasoning that can disappear behind a well-written answer.
Follow with guided practice, then an exercise using a different case. Finish with a debrief in which participants explain their work and what they checked. In some courses, part of the exercise can be completed without an assistant to observe learning. The choice depends on the target skill, not a uniform rule for everyone.
Adapt to roles and starting levels
A leader, communications specialist and developer have different tasks and verification responsibilities. Prepare examples close to their daily work. Beginners need guidance on limitations and appropriate context. More experienced users can work on evaluation, reproducibility and process integration.
Training material should remain reusable. A role-specific guide can include an example of context, approved sources, a checklist and what to do when uncertain. Avoid prompt catalogues without explanation: they become difficult to adapt when the task or tool changes.
Check learning after the workshop
Immediate satisfaction tells you about the training experience. It does not establish whether working practices have changed. Plan a review of a real task after a period of use, with a sample of outputs and a discussion of difficulties.
The manager and trainer can then distinguish a skills gap, a data issue or an unsuitable process. The next step may be another exercise, corrected learning material or a tool adjustment. The programme becomes an observable progression rather than reducing competence to the number of requests sent to an assistant.
| Skill | Exercise | Verification criterion |
|---|---|---|
| Summarise | Prepare a note from approved documents | Fidelity, omissions and identifiable sources |
| Analyse | Compare options using a matrix | Explicit criteria and visible assumptions |
| Verify | Identify weaknesses in an answer | Explanation of necessary corrections |
| Transfer | Handle a new case | Adapted reasoning, not simple repetition |
Sources & methodology
This insight combines cited publications with editorial analysis. Illustrative examples are not client results. Vendor features and terms may change.