There is a pattern running through the noise right now that most founders are going to miss until it bites them. Software is getting cheaper. Interfaces are getting friendlier. Anyone with a terminal and a bash script can call an LLM API with no runtime dependencies. Meanwhile, the person leading OpenAI's Codex is out telling anyone who will listen that taste matters more than ever precisely because software is getting cheaper to build. AI is being used to walk people through personalized, emotionally engaging experiences that feel real. Cameras in your neighborhood are recording far more than license plates, building behavioral profiles whether you consented or not. And in a piece that sounds academic until it doesn't, a journal just quietly erased two papers by Max Planck, one of the most important physicists in history, replacing them with blank pages and empty PDFs.
That last one is not a footnote. It is the thread. When the record becomes malleable, when tools get cheap, when engagement gets personalized and surveillance gets ambient, the question stops being "what can we build?" and starts being "who decides what's real, what's worth keeping, and what gets quietly deleted?"
That question lands directly on your desk.
The cheap-tools trap
We have watched this play out across dozens of engagements. A founder gets access to a new tool, a new API, a new model. The tool is genuinely useful. It removes friction, cuts hours out of a workflow, makes something that was expensive now cheap. The founder, rightfully, feels like they got an upgrade.
Then they make the mistake every single time: they assume the tool carries the judgment.
It does not. The bash wrapper that talks to your LLM will say whatever you ask it to say. The AI historical tour guide is engaging and personable, which makes it compelling whether it's accurate or not. The cameras track more data than any human analyst could ever review, but the profile they build is only as useful as the question someone thought to ask about it. The journal that deleted Planck's papers isn't wrong because it used a bad tool. It's wrong because someone with authority over the archive made a judgment call, and it was a bad one.
Cheap tools democratize production. They do not democratize taste. They do not democratize the judgment about what to build, what to say, what to keep, what to cut. And in a market where everyone has access to the same tools at roughly the same price, that judgment is the only thing that separates you.
What taste actually means when you're running a real business
Taste is not aesthetic preference. It's not having a good eye for fonts or knowing what color to make the button. Those things matter, but they're downstream of something harder.
Taste, in the sense that actually determines whether a business survives, is the capacity to make good calls with incomplete information about what is worth doing in the first place. It means knowing which customer problem deserves a six-week build and which one deserves a one-sentence email. It means knowing when the AI-generated copy is good enough and when it's going to make your clients feel like they're being handled. It means knowing when the cheap tool is a leverage play and when it's a shortcut that will show.
We see founders lose this around the $2M mark more than anywhere else. Not because they get lazy, but because they get busy. The operational chaos that got them to $2M becomes the enemy of the judgment that got them there. They start optimizing for speed and cheapness because they have to, and somewhere in that process they stop asking the harder question: is this actually good?
The surveillance problem is your problem too
This deserves a harder look than most founders give it. The Flock camera story is uncomfortable not because cameras are new, but because the behavioral profiling is ambient and invisible. Residents didn't opt into a surveillance system. They opted into "community safety." The profile got built anyway, from data they didn't know was being collected, for purposes nobody clearly defined.
If you are building any product that touches user behavior, even a scheduling tool, even a fitness tracker, even a farm management app, you are making choices about what data you collect and what you do with it. Most founders make those choices by default, not by design. They take whatever the SDK tracks because it's already there. They store whatever the database schema allows because nobody said not to.
That is a taste failure with a legal tail attached to it. What we tell clients is simple: decide what you actually need, collect only that, and be able to explain your choices to your most skeptical customer in plain language. If you can't, you've already made a bad call and you probably made it by accident.
The archive problem is your problem too
The Planck retraction is disturbing for a reason that has nothing to do with physics. When an institution can erase foundational work and replace it with blank pages, the question stops being about the papers themselves and starts being about what other things can be quietly revised without anyone noticing.
We have seen this happen to businesses. Not through malice, but through negligence. The contract that got signed and filed but never versioned. The original product vision that lived only in a Slack thread that got archived. The decision about why that feature was built the way it was, known only to the engineer who left in 2022. When the institutional memory degrades, the organization starts making the same mistakes twice, except the second time nobody knows it's a mistake because nobody can find the record of the first time.
Good operators build their own archives. Not for compliance and not for lawyers, but because judgment compounds when it's legible. The decisions you documented are the ones you can learn from. The ones that lived only in someone's head are gone the moment that person walks out the door.
What this moment actually requires
Here is what we see clearly: the next few years will produce a lot of very cheap, very capable, very mediocre products. The tools are democratized. The models are good. The barrier to shipping something that technically works is lower than it has ever been.
The barrier to shipping something worth using is higher than it looks. Because taste is harder to fake when everyone has the same tools. When the baseline moves up, what reads as good also moves up. The founders who will own their category in three years are the ones who are using cheap tools with expensive judgment. They know what they're building and why. They know what to keep and what to cut. They know what their customers actually need versus what's easy to give them. They build archives of their own thinking so their judgment compounds instead of evaporating.
That's not a technology problem. It's not a budget problem. It's not even a hiring problem, though we can help with all three. It is a clarity problem. And the only cure for a clarity problem is to stop optimizing for speed long enough to ask whether you're building the right thing.
The tools will keep getting cheaper. Taste won't.