Created from a single voice note with Agent Craft

Six companies in, I still make the same mistake at the start of every…

Six companies in, I still make the same mistake at the start of every one. I come in with a clear picture of what I'm building. I'm excited about it. I start talking to people, doing demos, getting early feedback. And then, slowly, the picture changes. Sometimes completely. With AgentCraft I had a version in my head that looked nothing like what we shipped. We started B2C. We pivoted to B2B. The whole time I thought I was wasting months. Looking back, I wasn't. That's just the process. But here's the thing I've learned after doing this six times: the founders who struggle most aren't the ones who pick the wrong idea. They're the ones who stop listening once they've started. They fall in love with the version in their head and stop paying attention to what the market is actually telling them. So I want to share the workflow I use now. Not a philosophy. An actual repeatable pattern. Step one: pick a space that genuinely interests you and has something new in it. New technology, new behaviour, new constraint. It doesn't have to be a fully formed thesis. You just get going. Step two: talk to as many potential customers as you can in the first 30 days. Not to pitch them. To hear them describe their problem in their own words. The language they use will end up in your product. Step three: write down every place where your original assumption got corrected. Not because you need a journal. Because the corrections are your map. If you're not finding corrections, you're not talking to enough people. Step four: be prepared to pivot. Fast. I know that sounds obvious but most founders treat a pivot like a failure rather than a signal. I've come to think of it as the product getting smarter. Step five: don't make a move toward scale until the use case is tight. This is where most of the expensive mistakes live. You hire ahead of clarity and then you're managing headcount against an idea that's still shifting. The COO I spoke to recently put it well without meaning to. She's on the road constantly. She goes to conferences, meets customers, hears things that matter. And by the time she's back at the hotel she's forgotten half of it. The insight that was sitting right there at the conference disappears. She said she'd been meaning to sort her LinkedIn for two years. That's not a discipline problem. That's a workflow problem. Her version of step one was clear: capture the thought where it lives, not six hours later at a desk. Thirty seconds on a voice note between sessions. That's it. The rest can be handled. The same principle applies to every part of building. The insight is usually available. The workflow either captures it or it doesn't. For the founders reading this who are somewhere in the middle of that early iteration phase: the confusion is normal. The feeling that it's not coming together fast enough is normal. The process only looks clean in hindsight. What you can control is how well you're listening, how fast you're correcting, and whether your tools are working for you or adding friction you don't need. If you want to see how we've tried to solve the last part for exec-led content, take a look at agentcraft.ai. Thirty-day free trial. No prompts, no models to manage. Just your thoughts and a workflow that turns them into content. Start there.