Observation: Not all AI is the same job

When asked about what AI they use, most leaders say the same thing: "I use ChatGPT." Or Claude. Or Gemini. The name changes but the pattern doesn't.

They've found one tool, they use it for everything. One mode of AI for three different kinds of work.

Every knowledge task moves through three phases. First you have to get the information in: meeting notes, documents, emails, research, things people said in hallways. Then you have to do something with it: analyse, compare, restructure, synthesise. Then you have to shape the output: edit, polish, sense-check, add your judgement before it goes to the audience.

Most people use a chat interface for all three. They paste something in, ask it to summarise, then ask it to rewrite, then ask it to check their logic. One tool. One mode. Everything in the same conversation.

It works. But it's like using a paring knife to peel garlic, debone a fish, and slice bread. You can. But there are other better suited knives for different jobs.

The leaders I work with who get real leverage from AI have stopped thinking about which tool to use and started thinking about which phase of the work they're in.

When they're capturing information, pulling together inputs from scattered sources, they use tools that run quietly in the background. Meeting recorders, document scrapers, automated summaries of things that would otherwise disappear. These tools don't need them present. They just need to be switched on.

When they're manipulating information (the most important stage for a leader) researching, comparing, analysing, they use specialist tools that are good at one job. A research tool for research. An analysis tool for analysis. Not a general-purpose chatbot asked to do everything.

When they're refining the output, polishing a board paper, pressure-testing a recommendation, checking whether the logic holds, that's where the conversational AI earns its keep. The back-and-forth matters here because this phase is where your judgement meets the machine's speed.

The pattern I see in organisations that have actually moved the needle with AI is simple: they matched the tool to the phase. The ones that are stuck have one tool doing all three jobs, and they can't work out why the results are mediocre.

This isn't about knowing more tools. It's about seeing the work clearly enough to know which phase you're in and choosing accordingly.

The next time you open a chat window and paste something in, ask yourself: am I capturing, manipulating, or refining right now? If the answer isn't refining, you're probably in the wrong place.