The AI tool market in 2026 is enormous, noisy and mostly irrelevant to a small serious team. The honest answer to 'what should we be using' is: fewer things than you think, chosen carefully, used properly.
One general assistant
ChatGPT or Claude, on a paid business plan with data protection in the contract. Used daily for drafting, summarising, restructuring notes and answering internal questions. One is enough. Running both in parallel is a distraction.
One writing surface that lives where the documents already are
Notion AI, Google Gemini in Docs, or the equivalent inside whichever suite the team already pays for. This is where the AI actually earns its keep, because the drafts appear next to the material they refer to.
One meeting note-taker, used with permission. Fathom, Fireflies, Otter or a native platform equivalent. Configured to only join meetings that have been explicitly opted in, and with recordings stored in the client's own tenancy. Used well, this alone removes a whole category of admin.
One automation layer
Zapier, Make or n8n, quietly connecting the inbox, the CRM, the calendar and the forms so the AI drafts land where they need to be without a human copy-and-paste.
For anything sensitive: a private assistant on the organisation's own infrastructure, with a model contracted not to train on the data, scoped to the organisation's own knowledge. This is the part the industry rarely writes about and the part serious clients rely on most.
What we deliberately do not use
Anything that requires shipping client data to a service without a data processing agreement in place. Consumer image or video tools with unclear rights on the output. Autonomous agents pointed at production systems without a human in the loop. Chatbots on the marketing site that pretend to be human. Any tool that has been paid for and not touched in the last month.
The pattern is simple
Fewer tools, better contracts, longer time spent learning each one properly. The teams that get real leverage from AI in 2026 are not the ones with the largest stack. They are the ones with the shortest.
Continue the conversation
Have a related piece of work?
Bring the process, source material and constraints. We can discuss the right starting point.
Start your brief ↗