Cutting through the noise
The big debates about artificial general intelligence and superintelligence are fascinating, and worth following. But they won’t tell you whether AI can save your team five hours a week this month.
For that, it helps to know what today’s tools are genuinely good at: working with language and documents, quickly, at scale, with a person checking the result.
Where it genuinely helps
- Drafting and rewriting. First drafts of emails, proposals, job adverts and web copy, in your tone of voice.
- Summarising. Long documents, email threads and meeting notes turned into the few points that matter.
- Searching your own information. “Ask our documents” tools that answer questions from your policies, manuals and past work.
- Sorting and routing. Reading incoming enquiries and sending them to the right person, with a suggested reply.
- Extracting data. Pulling details out of invoices, forms and PDFs into the systems that need them.
- Everyday technical help. Spreadsheet formulas, small scripts and explanations of technical jargon.
Where to be careful
- Facts. AI can be confidently wrong. Anything factual, legal or financial needs checking.
- Decisions about people. Hiring, credit or anything similar needs human judgement and accountability.
- Confidential data in consumer tools. Free tools may use what you type to train future models. Use business-grade tools with clear data controls.
- Anything that has to be right every time. Keep a person in the loop, or use traditional automation instead.
What about AI agents?
Agents, AI that can take actions on its own such as booking, updating records or sending emails, are improving quickly. They’re promising for narrow, well-defined jobs.
Start small: low-risk tasks, clear limits, an approval step before anything important happens, and a log of what the agent did.
Starting safely
- Pick one repetitive, language-heavy task and try AI on just that.
- Choose business-grade tools where your data isn’t used for training.
- Write a one-page AI policy: which tools, for what, and what never goes in.
- Keep a person reviewing outputs until you trust the results.
- Measure the time saved, honestly, before rolling it out further.
Next step
If you’d like help finding the right first use, or building something that works with your own documents safely, that’s what our AI and emerging tech work is for. For automation that doesn’t need AI at all, see what to automate first.