Don't Build Anything Yet

I know what you're thinking. You've read the last few newsletters. Everyone should code. Agents are replacing departments. Ship things fast. Move.

Hold on.

This week I sat in a strategy session with my team where we spent over an hour trying to define what our AI transformation roadmap looks like. And the single most important thing I said in that meeting was: don't build anything yet.

I know that sounds contradictory. I just told you everyone should be shipping code. But here's the problem I'm watching play out in real time, not just at my company but everywhere: teams are building things just to build them. They're automating random processes. They're creating one-off tools. Everyone has their own custom thing. And suddenly you've got 50 versions of the same solution flying around with zero coordination.

That's not a takeoff loop. That's chaos.

Here's what I think actually needs to happen before you build anything with AI. Three things. In order.

One: Get your data together.

AI is only as good as the context you give it. If your data lives in six different systems and nobody has a single source of truth, your AI outputs will be garbage. My team built a Notion database out of sheer desperation. They pulled data from Jira, Salesforce, NetSuite, Kayako, and about four other places into one spot. Was it an AI project? Not really. It was a data hygiene project. But it's the foundation everything else sits on.

You can't build agents on top of broken data. Full stop.

Two: Build your brain lifts.

A brain lift is your collected knowledge, instructions, and points of view on a topic. It's what turns generic AI output into output that sounds like you, thinks like you, and makes decisions like you would.

I'll be honest. I hated brain lifts for two years. We were told to build them internally, and I resisted. They felt like homework with no clear payoff. Then I started actually using them. And the difference between AI output without a brain lift and AI output with one is night and day. Without it, you get generic, average, safe answers. With it, you get your spiky point of view reflected back at you, sharpened and extended.

The reason brain lifts matter so much is that common knowledge produces common results. If you ask AI to build a customer retention strategy and you give it no context, it'll tell you to chase new customers, spend 30% of revenue on sales and marketing, and build the sexiest new features. That's what every business book says. But our playbook says the opposite: stop chasing new customers, keep the ones you have, only build what they're actually asking for. That's a spiky point of view. And unless you put it in a brain lift, AI will never give you that answer.

Three: Document your current state of affairs.

Before you automate anything, you need to know what you're actually doing today. What's manual? What's a time suck? What are the processes where humans are clacking away at keyboards inputting data that should be flowing automatically?

I think this is one afternoon of work. You sit your team down, map out every process, and be honest about which ones are still running on spreadsheets, Post-it notes, or just "somebody remembers to do it." That inventory becomes your roadmap.

Only after you have all three, the data, the brain lifts, and the current state, do you start building. You prioritize by biggest waste of time, highest feasibility with current tools, and lowest level of effort. You stack rank them. And then you chip away, one thing at a time, instead of trying to boil the ocean in a sprint.

My team wanted to jump straight to building agents. I had to pump the brakes. Not because building isn't important. It is. But building without context, without data, without a clear understanding of what you're solving for, that's just expensive tinkering.

Get the foundation right. Then build like crazy.