Most lists of AI tools for business leaders are out of date by the time you read them. Products get bought, renamed, folded into something else, or quietly stop being the best option. A list of names tells you what was true when somebody wrote it.

What holds up longer is the category and the judgement. There are five kinds of tool worth a leader’s attention, and for each one the useful question is whether to buy it or build it. Below is what we’ve tried, what we use now, and which way I’d go in each category.

I’ll name products, because a post that refuses to would be useless. Treat them as examples of a category rather than recommendations, and expect the names to change before the categories do.

1. Research and working through a decision

Reading around something, pulling together what’s known, arguing a decision through before you commit to it.

We’ve used Gemini and ChatGPT for this. We’re on Claude now. The differences between them matter less than people think, and I’d be wary of anyone who tells you otherwise with much confidence, because the ranking changes every few months.

Buy. This is the one category where I’d pay properly for the best you can get and not think too hard about it. It’s the tool your team will touch most often, and the gap between a good one and a cheap one shows up every single day.

2. Writing and content

This is the category with the most products chasing it, and the one where I’ve abandoned the most.

Lex was the one I got real value from, and not for the writing. I’d paste in a draft and it would tell me whether the argument was landing or whether I was just stating things one after another. Something that pushes back on your thinking is a different proposition from something that produces text, and it’s rarer. Pressmaster was decent at writing for a specific audience and giving me pointers on how I was talking to particular people. It also did a lot of other things I didn’t need.

I don’t use either now. The pattern is the same one I hit with most specialist AI tools: they do one thing well and eight things you’ll never touch, and to get at the one thing you have to work the way they work.

Build, mostly. Not in the sense of writing software. I mean setting up a general assistant with your own material, your own voice, and your own standards, so it does the specific job you wanted rather than a generic version of it.

3. Connecting your systems together

Getting information out of one system and into another without a person copying it across. This is where most of the real time goes in a small business, and where most of the errors come from.

We used Zapier, then Make, and we’re predominantly on n8n now. Sometimes the answer isn’t a platform at all. Some of what we run is small scripts that Claude Code wrote and we dropped into place, because a whole automation platform was overkill for the job.

Buy first, then build. Start with something off the shelf, because you don’t yet know what you need. Once you’ve run it for a few months you’ll know exactly which bits matter, and at that point building is cheaper and fits better.

Bring AI to where you already work, rather than moving your work to wherever the AI lives.

4. Reporting and dashboards

Pulling numbers from several places into something you can actually look at on a Monday morning.

We do most of this with Claude Code and Google Scripts. Get this from there, combine it with that, put it in a spreadsheet in this shape. It’s unglamorous and it works.

Build. This is the category where the answer has genuinely changed in the last two years. Building a reporting tool used to mean a developer and a project. Now it’s an afternoon. If you’re evaluating expensive dashboard software, it’s worth finding out what building the thing you actually want would cost before you sign anything.

5. Your own knowledge and documents

An assistant that knows your processes, your policies and what you told a client eighteen months ago, so people stop asking whoever’s free.

Most of the major assistants do a version of this now, and for most businesses that’s enough. The work isn’t in the tool, it’s in having documents worth pointing it at, which is a separate and less exciting problem.

Buy, and read the terms. Check whether your content is used to train the underlying models on the plan you’re actually on, because that usually differs between the consumer and business tiers.

The trap most of these tools set

Every SaaS company on earth is currently adding AI features to look relevant. Some of it’s good. The failure runs the same way every time, and it isn’t a small business problem, it happens at every size.

You buy something because it looks like it solves your problem. It gets you seventy per cent of the way and then stops, because it was built to solve a generic problem that resembles yours rather than yours specifically. Now you either do extra work round the edges, or you change how you operate to suit the software. People usually pick the second one without noticing they’ve picked it, and a year later the business runs the way the tool wants.

One question worth asking before you buy anything

Does this come to where we already work, or does it expect us to move into its house?

The most value we get from the software we pay for now isn’t the AI features the vendor built. It’s that we can reach into those systems from where we’re already working. Increasingly that happens through something called MCP, a standard way for an AI assistant to connect to other tools, but you don’t need to know the term to apply the test. Ask whether it connects to what you already use. Ask whether your people can get at it without opening another tab and learning another interface.

Bring AI to where you work. Don’t follow a product down a journey somebody else laid out for you.

When buying is clearly right

I’m not arguing you should build everything, and there are places where buying wins without much of an argument.

Anything regulated, where the compliance burden is the product and you don’t want to own it. Anything sitting on a proprietary dataset you can’t replicate. Anything behind a paywall, where what you’re buying is access to information rather than software. In those cases the thing you’d be building isn’t the tool, it’s the data underneath it, and you can’t.

Not everyone has to build

One thing I want to be careful about. We’re a technical company, and when I say we build our own, that’s easy for me to say and harder to act on if your team doesn’t look like ours.

The way this actually works in a business is that a few people build, and most people use what’s been built. Not everybody needs to manage connectors or write automations. Some will want to, and those are the people worth investing in. The rest need to be good at using what those people make, and at spotting when something they’re doing by hand every week shouldn’t be. That’s the whole team moving, at different speeds, rather than everyone becoming technical.

Common questions

What AI tools do business leaders actually need?

One good general assistant for research, thinking and writing. Something to connect your systems together. Somewhere your own documents live that people can ask questions of. Reporting and dashboards are usually cheaper to build than to buy now. Everything else is optional and most of it gets abandoned.

Should we buy AI software or build our own?

Buy where the thing you’d be building is really a dataset, a compliance burden or paid-for access to information. Build where the job is specific to how your business runs, because a generic tool will get you most of the way and then make you work its way. Building has got dramatically cheaper, so it’s worth checking the cost before signing a long contract.

Why do specialist AI tools get abandoned?

Most do one thing well and a lot of things you’ll never use, and reaching the useful part means working the way the tool works. If your work is generalist, which most business work is, a general assistant set up properly tends to outlast the specialist one.

What is MCP and does it matter?

A standard way for an AI assistant to connect to other software. It matters because it lets you work in one place and reach into everything else, rather than moving between tools. You don’t need the term to apply the idea: ask whether a product connects to what you already use.

Does everyone in the team need to learn this?

No. A few people will build things and the rest will use them. What everyone needs is enough understanding to spot when something they do by hand every week shouldn’t be done by hand.