For seven months I’ve been building our own AI agent. It’s taught me something I hadn’t expected.
The hard part isn’t building. The hard part is choosing what not to build. And finishing the things you do build.
Everyone has access to the same tools. That’s why I think the difference between companies over the next few years will be who gets them working day to day. That’s execution.
The hardest part was leaving things out
Recently I built a dashboard with AI that was meant to show operations, finances and growth. The first version had everything in it. It was so cluttered that I didn’t know what to look at. Now it shows eight numbers at the top. The hard part was choosing what to cut.
I’ve called it bad taste before, when you make something because you can and not because it’s needed. I said that about content. It applies to what you build, too.
Start with a problem, not a tool
Any technology moves faster when you have a problem that’s big enough. Our own AI agent started with a problem, not a tool. A tool without a problem becomes a side project. A problem that’s big enough gets solved.
Execution is what runs every day
Execution means putting a decision into practice. Not starting it. Finishing it, so it works every day.
It’s a long way from a demo that works once to something a whole team relies on.
For us, it’s not the cleverest features that have made the difference. It’s the ones we use every day: the morning review of the ad accounts, the check of every creative, and corrections that become rules.
None of it is especially exciting to look at. It’s just finished. And that’s why it works.
Structure is a necessary evil
I’m not naturally structured. Outside work I always think I have more time than I do, and I forget things.
Even so, I care enormously about structure at work. Because structure is what makes you prioritise the right things and know why you’re doing what you’re doing. You can’t get around it if you want to go far.
The funny thing is that AI can help with that part too. These days the agent remembers things for me. What I did last week, and why I chose one solution over another.
So AI isn’t just what makes it easy to build more. Used properly, it’s also what helps you finish.
Say no to whatever takes your focus
I’ve always said you should say no to the clients who pull your focus away from the core work. The same goes for AI projects.
This is how I try to do it myself:
- One problem at a time. Find the problem that costs the most, and start there.
- Finish it before the next one starts. Even when the next one looks more fun. A half-built system isn’t worth half as much. It’s worth nothing until someone uses it.
- Measure whether it gets used. Not whether it’s been built.
- Let a person approve it. AI does the work. A person makes the decision.
- Get 1% better every day. One thing at a time makes the big problems small.
What we can’t do without
None of the things we use every day were the most exciting to build. They were the ones we finished.
I think the winners of the next few years will be the ones who’ve got one system running so well that they can’t do without it.
If you’d like to hear how we’ve done it at Amplify, I talk about it in my AI talk.
