Skip to content
Topic

AI at Work

Using AI in a job — the professions where it changes the day, the admin it removes, and the governance questions that come with client and employee data.

Most professional AI advice is either generic productivity tips or vendor case studies. What actually differs between jobs is not which tool you use but which constraints you work under: who reviews the output, what data you are allowed to process, and where being wrong is expensive. This hub is organised around that. It covers the administrative load AI genuinely removes across most professional work, and then the profession-specific parts — the audit trail, the regulated claim, the confidential record — that decide how any of it can be used at all.

The admin layer is the reliable win

Across almost every profession the same categories come back: correspondence, proposals, reports, meeting notes, documentation and the reassembly of the same facts for different audiences. None of it is the skilled part of the job, all of it is frequent, and it is where the hours quietly go. Templating those once and generating per case is the change with the highest hit rate and the least risk, and it is a better place to start than anything more ambitious.

Know what needs a human signature

Every regulated profession has a line, and it is usually clearer than people assume. AI drafts; a qualified person reviews and signs. Clinical notes, financial reports, legal advice, employment decisions and anything with a compliance obligation all follow that pattern. The failure is not using AI for these — it is letting a plausible draft through without the review the record assumes happened.

Settle the data question before you start

Client information, employee records, patient data and commercially sensitive material each carry obligations that most consumer AI tiers do not meet. Find out which tools your organisation has approved for which categories before anything is pasted, prefer stripping identifiers where the task allows it, and remember that a request usually works just as well with names and numbers removed.

Automate the process, not the judgement

The pattern that goes wrong across every profession is automating a response to a signal nobody has interpreted — outreach triggered by a health score, screening decided by a model, a risk flag actioned automatically. Signals are worth surfacing and worth investigating. Turning one into an action without a person in between is how teams end up sending cheerful emails to customers who are already leaving.

Start with the administrative layer, because it is where the hours are and the risk is lowest. Keep the judgement, the signature and the difficult conversations human. And answer the data-governance question first, since it is the one that determines whether any of the rest is available to you at all.

Try it yourself

The free plan includes 100 credits and needs no card.