There is a song by AJR called "The Good Part" with a line that goes "it's so hard, can we skip to the good part?" In the song it is about life, he is tired of the slow uncertain middle and just wants to fast forward to the part where things have worked out. Lately that line has been stuck in my head for a different reason, because that is exactly what AI has let all of us do when building product. We have an idea, and a few prompts later there is a working product in front of us, buttons and all, with almost none of the thinking that used to sit in between. We skipped straight to the good part.
So what is the good part, really? It is the moment you have something you can see and touch. There are buttons, there are screens, something runs, something is usable, something exists. That is the part everyone wants, because it is the part you can show people. Everything that comes before it, the figuring out what to build and why, how deep it needs to go, what actually matters, none of that feels like the good part. It is groundwork, and nobody celebrates groundwork.
AI has made the good part almost free. If a working product can be put together in a weekend, then the buttons and the screens and the thing that runs are no longer the rare, valuable part of the process. They are the commodity. The thing that is actually scarce now is everything we were so happy to skip, the groundwork that decides whether the product is worth building at all and how deep it goes. We ran toward the good part and ran straight past the only part that still holds any value.
And we did not skip it because we decided it was not worth doing. We skipped it because we stopped noticing it was there. When building was expensive, the cost forced you to stop and ask whether this was the right thing to build and how far it needed to go. Now that the cost is gone, so is the prompt to think. The question "should we even build this, and how much depth does it actually need?" used to be unavoidable. Today you can go from idea to shipped without it ever once crossing your mind.
Stuck in the middle
"Stuck in the middle" phrase comes from Michael Porter, who argued in 1980 that a company can win through one of three strategies, being the lowest cost, being the most differentiated, or focusing tightly on a niche. A company that commits to none of them is stuck in the middle, with no clear advantage, not the cheapest, not the most distinctive, not the most specialized (Porter's generic strategies). It does a bit of everything and nothing well enough to matter, and that is exactly where a lot of AI-built products are ending up.
To see why AI pushes you here, go back to how things worked before it. Going deep on one thing was hard, and going broad across many things was also hard. Both cost real effort, so you were forced to pick, and more often than not you ended up going deep on the one thing that mattered, because spreading yourself thin was just as expensive and got you nowhere.
Broad got cheap, depth stayed hard
AI did not lower both costs equally. It made going broad almost free while leaving going deep exactly as hard as it always was. Adding another feature used to be a sprint, now it is an afternoon of work. But making one feature genuinely good, handling the edge cases, deciding how far it actually needs to go, that part AI will not do for you. Ideas are cheap, everyone has a long list of surface level features they could add. Execution to real depth is still expensive, and it is expensive in the one currency AI cannot spend for you, which is your own thinking.
Think of every surface level idea as a shiny object. There are always plenty of them around you, a new feature a competitor shipped, a request a single customer made, an integration that sounds impressive, an AI chatbot because everyone has one now. They all look attractive from a distance. Earlier, chasing a shiny object cost you something real. Picking one up meant weeks of engineering, so before you reached for it you had to stop and ask whether it was actually worth the effort. That cost was annoying, but it was also doing quiet work for you. It filtered the shiny objects down to the few that mattered and pushed you to go deep on those instead of grabbing all of them.
AI removed that cost, and with it removed the filter. Now you can walk around and pick up any shiny object you want without breaking a sweat, so you pick up all of them. Each one is cheap on its own, and none of them feels like a mistake in the moment. But add them together and you have a product that does ten things at a surface level and not one of them well. You did not decide to end up in the middle. You drifted there, one cheap shiny object at a time, because nothing was stopping you anymore.
This is the part I want to make obvious, because it is easy to miss. Going broad feels like progress. Every feature you add looks like movement, the product is getting bigger, there is always something new to show. But it is movement sideways, not down. You are accumulating surface area, not depth, and surface area is exactly the thing AI has made worthless because anyone can generate it in an afternoon.
Who went deep, and who drifted
Here is the point, and Porter made it decades ago. You win by committing to one thing and doing it better than anyone, and you lose by trying to do a bit of everything.
Look at who won by going deep. OpenCode, an open source coding agent, could have spent its time chasing feature parity with Claude Code. Instead it went the other way and rebuilt its entire terminal interface on its own in-house framework, OpenTUI, betting that a genuinely better experience in the terminal mattered more than matching everyone feature for feature. Superhuman picked one narrow group of power email users and served them so deliberately that it capped onboarding at about a hundred new users a week, each given a personal thirty minute session, and built one of the highest retention rates in software. Linear did the same for issue tracking, winning on speed and craft against tools with ten times the feature count. None of them won by doing more. They won by going deeper than anyone was willing to.
Now look at who drifted into the middle, because tech has just as many of those. Evernote started as a fast, focused notes app and then tried to become everything productivity, adding business card scanning, chat, presentation mode, at one point even selling socks and backpacks in its own store. Their CEO admitted the core problem himself, that people loved Evernote but only used about five percent of it, and it was a different five percent for everyone. The app got slow and bloated and users left. Yahoo did the same thing at a larger scale, trying to be search and email and media and a social network all at once, cycling through seven CEOs in fifteen years and never settling on what it actually was, until a company once worth over a hundred billion dollars quietly faded out. Google, meanwhile, just kept going deep on search.
Depth first, then breadth
To be fair, none of this means a company can never broaden its scope. Googe/Zoom did eventually expand into phone, events, and a whole suite, and that was the right move for them. The point is about order and timing. You earn the right to go broad by first doing one thing well enough that it becomes your base, the thing that attracts your customers and gives you something solid to expand from. Going wide before you have that base is just being stuck in the middle with extra steps. So this depends on the stage you are at, and at almost every early stage, depth comes first.
There is a simple reason the order has to be this way, and it comes down to resources. You only have so much time, attention, and money, and going broad and going deep at the same time is usually not possible with a small team. Worse, the two pull in opposite directions, and broad is always the easier pull. Adding another feature is comfortable and visible, going deeper on the one that matters is slow and frustrating. So if you do not deliberately protect depth, you will drift broad by default, simply because it is the path of least resistance.
It happens at every scale
One more thing worth seeing clearly is that this trap is not only a company level problem. It repeats at every scale. At the macro level it is a whole company like Yahoo trying to be everything. One level down, it is a large company spinning up a new product that tries to do too much at once and lands in the middle, even though the parent company is perfectly focused elsewhere. Down again, it is a single product piling on features until none of them is really good. And at the smallest level, it is one feature that tries to serve five different use cases and ends up serving none of them well. Wherever you are, from a company down to a single screen, you can be stuck in the middle.
And that is the quiet danger of this whole thing. At no single step are you obviously doing anything wrong. You are just saying yes to one more reasonable looking thing, and then another, and each one comes out average because average is all you had the attention for. The trouble is that average spread across ten things does not add up to one good thing. It adds up to nowhere. You end up busy, productive looking, shipping constantly, and somehow with a product that nobody is excited about.
How to stay out of the middle
If going broad is the default and AI only makes that default stronger, then staying out of the middle cannot be passive. You have to fight the pull on purpose. The single most useful habit I have found is to decide, up front, what you are willing to be bad at.
This sounds backwards, because the instinct is to try to be good at everything, and AI makes that feel possible now. But being good at everything is just another way of being stuck in the middle. So you flip it. You pick the one thing that has to be genuinely excellent, the core that your whole product lives or dies on, and you protect it. Then you look at everything else and you consciously allow it to be average, or you choose not to build it at all. Not because those things do not matter, but because your attention is finite and the core matters more.
The hard part is that this requires saying no to things that genuinely look important. Something will come up that seems like it clearly deserves real effort, and most of the time the honest answer is still to give it the minimum and move on, because spending your depth there means stealing it from the core. A good question to keep asking is simple. If I make this excellent, does it change whether the product wins? If the answer is no, that is a thing you are allowed to be bad at, and you should be bad at it on purpose so you can be great where it counts.
Protect your one line
It also helps to realise that this drift does not only happen at the feature level. Often the more dangerous version is at the level of positioning, of who you actually are. It is worth being able to finish the sentence "we are the company that ___" in one clear line, and then using that line as a filter. If you are the company that makes great video experiences, then the real question on every decision is not "is this a good feature" but "does this make our video experiences better, and is it still true that we are the best at this." Zoom could answer that for years, every bit of obsessive work went into the call just working, and the positioning stayed honest. The moment you can no longer say your one line with a straight face, you are already drifting into the middle, you just have not noticed yet. So you keep checking yourself against it, on purpose, again and again.
The cleanest way to run that check is to walk back through Porter's own framework and answer three questions honestly. First, who are we exactly, in one line. Second, what is our edge, are we winning on cost or on differentiation, and is it across the whole market or a narrow segment we have chosen. Third, is the work we are doing right now actually sharpening that edge, or is it just adding surface area somewhere else. If you are the differentiated player, the only move that compounds is more differentiation on the thing you are already known for, going deeper there until the gap between you and everyone else is something they cannot casually copy. Everything that does not push that edge is, at best, a distraction, and at worst the first step into the middle.
And this is exactly where you have to be most careful now, because the whole point of this piece is that AI quietly makes it easier to fail this check. It makes every distraction cheap to build, every shiny object easy to grab, every sideways move feel like progress. So do not let it. Let AI make the execution faster, but do not let it make the thinking optional, and do not let how easy something has become be the reason you decide to build it.
So, can we skip to the good part?
The good part, the visible product everyone wants to fast forward to, is exactly the part AI has made cheap and ordinary. Anyone can get there now. The part we were all so eager to skip, the slow uncertain middle where you decide what matters and go deep on it, turns out to be the only part that was ever worth anything. We have been racing to skip the one thing that was actually the good part.
None of this is black and white. Sometimes skipping to the good part is exactly the right move. If you are just trying an idea, throwing together a quick prototype to see if something is even worth doing, then getting to a working thing fast is the whole point, and stopping to think deeply would be the wrong instinct. I am not handing you a rule that says always go slow and always go deep. The point is to know which mode you are in and choose it on purpose, instead of skipping by default just because AI made skipping free.
And some things you could not skip to the good part on even if you wanted to. The genuinely hard, deep technical work has no shortcut. If you are building something that takes real thought just to get working, you are already doing the deep part, and this whole piece does not apply to you. The trap I am talking about is only the other kind of work, the things that have become so cheap to build that you build them without thinking, not the things that still demand it.
So no, let's not skip to the good part. Let's go back and do the part we kept skipping, the hard, unglamorous, deeply thought through groundwork, and do one thing well enough that it is actually worth showing. That is still the only thing AI cannot do for you, which makes it the only thing that still matters.