AI got introduced under the premise “Point it at a problem, get a solution” and more money got poured into that idea than ever before for any technology in history. Now we’re reaching a point where the cards are being shown, and that promise is untrue. You cannot just point AI at a problem and get a solution. So far we’ve identified three areas where you still have to do the work to make AI an effective tool. These areas take up more time and mental energy than AI has capacity for, and involve a level of thinking that AI is not programmed to do.
1. Pre-Planning: AI will jump forward with whatever it’s programmed to do. It cannot and will not come up with a novel way of attacking a problem. You give it something you want done, it does it the same way it always does, and the proof is in the pudding. You see every vibe coded app, website, piece of content, look and sound the same. Because it’s going to do what it’s going to do when you give it a prompt. Pre-Planning involves putting together all the pieces and optimizing your path forward for your specific use-case. Often this means pulling in inspiration from far outside the realm you are specifically working in, and reaching for answers from areas of the world your product or service has nothing to do with. Like if you’re building software and you take inspiration from a door handle or car dial. These translations back to the real world, and to things your customers are familiar with, make the difference between a usable product and slop. And it takes your knowledge of both the world and your customer to make it happen.
2. Active guidance: As the AI builds, it’s getting better at asking for steering. The other day I answered 30 questions regarding a one-page build which I’d already laid the framework for prior to starting the build. This is good, because instead of the AI once again running off and building you something generic, it’s building you something specific by asking you for the direction it needs. AI is getting better at this, and it means more human involvement in the AI. Hence the magic bullet facade being gone. But this is where potentially AI becomes even more valuable alongside human effort, because as you steer the ship it can run the engine. You can be the rudder and it can be the power unit driving the whole thing forward. In the past, entrepreneurship was exceptionally difficult because you were always flip-flopping between thinking and doing. The maker/manager conundrum keeps entrepreneurs from fully optimizing because you are constantly context switching throughout the day. With AI, you can stay fully in manager mode, thinking about the problems, thinking about the business, thinking about the strategy, thinking about the customer, instead of switching back and forth to building. It used to be that we’d carve off one feature and then wait 3 months to get it built. Now we carve it off, it runs in the background for a few hours, and then we steer it, fully live in a few days at most. Everything that we needed to do can now get done, instead of waiting endlessly for builds to happen. This will unlock a whole new era of growth, albeit with a whole new era of energy drain because my goodness does it involve a lot of thinking, planning, and decision making, all in real time.
3. A framework for the AI output to live in. Vibe coding is backfiring because you build quickly, but not stably. Productivity slows down over time because all code gets thrown into a framework you don’t have control over, and your pace of output slows down with every subsequent push. A framework for your code to live in, siloed out by parts of the product, is the only way to make sure your development speed doesn’t slow down. Our team has done a really excellent job of building a system so that we can grow our product and feature set over time without losing speed. This required a multi-year planning process. Which perhaps could be faster now, but there are a lot of structural decisions to make to ensure it all works well and doesn’t hit issues as you scale. This is also becoming a more fluid process now, where you have to constantly adapt your system.
The next era of AI should be the one that really drives growth. But it will look flatter than the first era when all the revenue accrued to the labs. This phase will be companies implementing AI using these principles, and more, to make real gains in their growth.
