AI at work: a subscription is not a skill
The same tool can produce very different results. Anthropic’s research raises a business question: how much do we invest in AI, and in the people using it?
A company buys access to AI, sends employees a link and waits for productivity to rise. One person gets a useful draft, another gets polished text with little substance, and a third gives up after the first poor answer. The subscription is identical. The results clearly are not.
Using AI is also a workplace skill. Anthropic’s March 2026 report, Learning curves, found higher conversation success among more experienced Claude users. The researchers caution that this association does not establish causation: early adopters may differ from other users, while learning through use could explain some of the gap. These are observations from one platform, not a measurement of every employee or AI tool. Source: Anthropic.
A good collaborator needs a good brief
Consider a fictional small company improving answers to customer questions. Asking AI to “write a customer reply” leaves it guessing about the product, tone and company promises. Providing the issue, verified product information and an example of a good reply offers a much stronger starting point.
There is no need to hunt for a magic sentence. Explaining the task, context and success criteria matters more. Do we need a short response, a comparison, or a summary separating confirmed facts from open questions? If we cannot define the goal ourselves, the tool will struggle too.
Measure time until the task is finished
A commercially useful result survives review. A draft created in one minute offers little benefit if correcting inaccurate information takes another half hour. Measure the entire process: preparing instructions, generating a draft, checking it and making final edits.
An initial internal trial could focus on a recurring task with a clear output, such as summarising public product descriptions. Compare the usual workflow with the AI-assisted version and assess quality too: missing information, invented claims and required corrections. This is an editorial suggestion for measurement, not a finding from Anthropic’s research.
Useful guidance comes from actual work
When someone finds an effective process, document it so a colleague can repeat it. A short input example, the expected output and a checklist of items to verify may be more useful than a lengthy presentation about AI’s future.
The business opportunity lies in turning an individual trick into shared knowledge. A process dependent on one enthusiast is fragile. A team that understands why it works can improve and adapt it.
A practical next step: choose one routine task, assign someone to review the output and compare time and quality after several attempts. Then consider expanding. AI can accelerate a workflow, but the company still needs to know what a job done well looks like.
Video: The prompting playbook
Video in English.
This article was created with AI assistance.
AI translated
