Every Company Wants AI. Almost None Are Ready for It.

Every week, someone tells you something new.

Build AI agents. Use this LLM. Automate your workflows. Integrate your CRM. Hire a prompt engineer — wait, no, that’s already obsolete. Now you need an “AI strategy.” And a vendor to go with it.

I’ve talked to dozens of founders and business owners over the past year. Most of them feel like a rabbit in front of a snake. They know something is coming. They know they need to act. But every day the picture changes, and the result is often: they do nothing. Or worse, they buy a tool, nobody uses it, and they conclude that “AI doesn’t work for us.”

Here’s what I’ve learned: the tools are rarely the problem. The foundation is.

The mistake most companies make

When I show entrepreneurs how my team uses AI, the first thing they expect to see is a fancy tool. Instead, I show them a folder of Markdown files.

Not sexy. But it’s the most important thing we’ve built.

Before any AI assistant can help you meaningfully, it needs context. Context about your customers, your projects, your conversations, your decisions. Without that, even the best model is just a very expensive autocomplete. It will hallucinate, give generic answers, and frustrate your team.

The question isn’t “which AI tool should we use?” It’s: “do we have our knowledge in a shape that an AI can actually work with?”

Most companies don’t. Knowledge lives in people’s heads, in scattered emails, in Word documents nobody can find, in Slack threads that disappear into history. When your best salesperson leaves, most of what they knew walks out the door with them.

This isn’t a new problem. But AI makes it urgent in a new way.

What a knowledge foundation actually looks like

I’ve been building what people in the productivity world call a “second brain” — a structured, searchable knowledge base that lives in plain text files. I use a tool called Obsidian, but the tool matters less than the principle: your knowledge should live in a format that’s both human-readable and machine-readable.

Markdown files tick both boxes. You can read them in any text editor. And they happen to be the preferred format for large language models.

The structure I use is simple. There’s an inbox — raw material comes in here, transcripts of meetings, notes, ideas. Then there are structured notes: company notes, people notes, project notes, principles. Everything is linked. A customer note links to meeting notes which link to people who were in the room.

When I get off a call now, I don’t sit down and write a summary from memory. The transcript goes into the inbox, an AI agent reads it against my existing knowledge base, and within minutes I have a structured meeting note in the right place with the right fields filled in. Action items flagged. Context from previous conversations pulled in.

What used to take 30 minutes — and often didn’t happen at all — now takes two.

What this unlocks for teams

We’ve taken the same approach with our sales team. We have a shared knowledge base — same structure, same principles — connected to our CRM. Not replacing the CRM, just connecting to it.

The result: when I hand off a prospect to a colleague, I don’t need an hour-long briefing. They open the repository, read the context, and know what matters. The conversation we had six months ago, the concern the customer raised last time, the decision we were waiting on — it’s all there.

More importantly, when we’re advising clients on AI adoption, this is where we start. Not with tools. With this question: if you had to hand over your most important customer relationship to a new colleague tomorrow, what would they need to know — and where would they find it?

In most companies, the honest answer is: they’d have to interview everyone who’s ever worked on that account.

That’s the gap AI tools can’t fix. That’s what you need to fix first.

The practical starting point

You don’t need to build a second brain overnight. Start with one area where knowledge loss is most painful. For many companies, that’s sales. For others, it’s customer success or project delivery.

Pick one type of note — say, meeting notes with customers. Define what should be in them. Date, attendees, key decisions, next steps, open questions. Make it a habit, or better, make it automatic. Record your calls. Let a tool transcribe them. Feed the transcripts into a structure you own.

After three months, you’ll have something valuable: a searchable record of what your customers have told you, what you’ve promised, what they care about. That’s when AI tools start to get genuinely useful — because now they have something real to work with.

The harder truth

The companies I see getting real value from AI aren’t necessarily the ones with the biggest budgets or the most sophisticated tools. They’re the ones who did the unglamorous work of getting their knowledge out of people’s heads and into a shared, structured, accessible place.

That work is not exciting to announce on LinkedIn. There’s no vendor selling it to you. But it’s the difference between AI that impresses in a demo and AI that actually changes how your business runs.

So before you schedule that next AI tool demo: ask yourself what the AI would have to work with. If the answer is “mostly nothing” — start there.