AI, explained plainly
No jargon, no hype, and no pretending this is simpler than it is. If you run a business and want to understand what people are actually talking about, start here.
The vocabulary
Terms worth knowing
Choosing a model
Pick the tier, not the brand
Every provider ships roughly the same four tiers, and the names change every few weeks. Design around the tier and you can swap the model as the market moves.
| Tier | Relative cost | Best for |
|---|---|---|
| Frontier | £££ | The most capable tier from each provider. Worth it for genuinely hard reasoning, long document analysis and complex agent work where a wrong answer is expensive.Research, analysis, complex multi-step agents |
| Balanced | ££ | The workhorse tier, and the right default for most business tasks. Strong enough for professional writing and nuanced instructions at a fraction of frontier cost.Business writing, summarising, day-to-day agents |
| Fast / small | £ | Optimised for speed and volume. Ideal where you are making thousands of calls and each one is simple — classification, routing, extraction.High-volume chat, routing, data extraction |
| Open weight | Self-host | Models you run on your own infrastructure. Slower and more work to operate, but the data never leaves your estate — which is sometimes the whole point.Regulated sectors, on-premise, privacy-critical |
Capability
Can it actually do the task at an acceptable quality?
Speed
Does a person sit waiting for the answer, or is it a background job?
Cost
Ten requests a day or ten thousand? This dominates at volume.
Privacy
Where is the data processed, and what does your DPA actually say?
Questions we get
Straight answers
Still want to talk to a person?
We write up how we build our own systems in public. Read that first if you want to judge us before you speak to us.