The Curation of Scarcity
Infinite possibilities really means infinite.
Picture the PM of 2026. They wake up to review three features their AI dreamt up, executed on, and A/B tested overnight. The ideas are fine. Competent. The kind of features that would’ve passed any prioritization meeting. That’s the problem.
This is supposed to be the pitch for why we need better thinking tools. But read it again. We made building infinitely cheap and immediately discovered that building was never the hard part.
There’s a certain genre of tech analysis that identifies a real problem, then proposes tooling as the solution. The argument goes: we have tools for making (IDEs, Figma, spreadsheets) but not tools for thinking. Therefore we need tools for thinking. QED.
What if the bottleneck isn’t tooling? What if the bottleneck is that nobody in the organization is incentivized to say no?
The mantra of the AI-native enterprise is that every feature which can be built will be built. This is presented as opportunity. It’s actually entropy.
Feature bloat already kills most software. Products don’t die from insufficient features. They die from too many. The discipline of resource constraints used to enforce a crude kind of editorial judgment: you couldn’t build everything, so someone had to pick. Now you can build everything. Nobody has to pick. Everyone will build, and nothing will get cut.
Scarcity was doing editorial work for free. Now it doesn’t… and product surface area is like a ratchet: it only goes one direction.
From feature rich to poor performance
There’s a reason “technical debt” has a more famous cousin that doesn’t get talked about as much: path dependence. The Excel plus-sign-as-equals was a clever growth hack to bag Lotus 1-2-3 users. Thirty years later, Microsoft still can’t remove it. That one’s harmless. Most aren’t. Economists call this the QWERTY problem. Physicists call it hysteresis, where the current state depends on history, not just present conditions. Engineers have a better name: load-bearing jank.
You know load-bearing jank. It’s the spreadsheet everyone’s scared to look at the wrong way. The one with seventeen tabs, a few circular references that somehow resolve, and a macro that Kevin from accounting wrote in 2019 before he left. Nobody understands it. Nobody dares touch it. But it runs payroll, so it stays.
Every organization has load-bearing jank. The feature that looks vestigial but props up some workflow nobody documented. The API endpoint that should’ve been deprecated three years ago but turns out sixteen enterprise clients built their entire integration around it. Chesterton’s fence, but for software.
AI creates an interesting fork in the road here. Path one: we build infinite jank at unprecedented speed, and none of it ever gets unbuilt. The ratchet accelerates. Path two: we perform some kind of jank abstraction. Let AI agents interface with our load-bearing garbage so humans never have to touch it again.
I genuinely don’t know if jank abstraction is ostrich-mode or actually smart. The case for head-in-sand: you’re just adding a layer of abstraction over the mess. The mess is still there. Now you have two things to maintain. The case for actually-smart: humans are terrible at interfacing with jank. We forget context, we fat-finger the formula, we resent every second of it. AI doesn’t care. AI can hold infinite context. AI will cheerfully talk to your cursed spreadsheet at 3am without complaining.
Maybe jank isn’t a problem if nobody has to touch it directly. Maybe the future is AI as jank translator. A diplomatic layer between humans and the accumulated cruft of every decision anyone ever made. The same people promising AI will help us build faster might end up using it primarily to interface with the garbage we already built.
The dream of “tools for thinking” assumes the problem is cognitive. That PMs and founders need better interfaces for exploration. But the PMs I know aren’t short on ideas. They’re drowning in them. The backlog isn’t empty, it’s infinite. What they need isn’t more thinking tools. What they need is organizational permission to kill things.
You can’t tool your way out of an incentive problem.
Service functions (legal, finance, HR) haven’t embraced software because their job security depends on process complexity. Product teams ship bland features because nobody gets fired for shipping something measurable, and everyone gets fired for shipping nothing while “thinking.”
“Every team should be a software team” sounds great until you realize what it means in practice: every team now has infinite building capacity and zero editorial authority. Marketing can ship features. Legal can ship features. The PM who used to be the bottleneck is now just another builder in a company full of builders.
What’s the value of a curator when everyone can create?
There’s an old product management joke: the hardest part of the job is saying no. AI just made that infinitely harder. We automated the easy part of product development and amplified the hard part.
We didn’t need infinite builders. We needed one good editor. The market is about to discover what happens when everyone can make and nobody has to decide.
Unless someone already decided years ago.
The buyer-builder doesn’t need tools for thinking because they’ve already done the thinking. Not in a workshop. Not in a brainstorm. In the field, for a decade, watching what actually matters and what’s just noise.
The PM drowning in infinite backlog possibilities has no earned authority to kill anything. Every feature is equally plausible because they’re all equally abstract. The buyer-builder has scar tissue. They know which features are load-bearing and which are jank because they’ve been the user cursing at the jank for years.
Domain expertise isn’t just knowing what to build. It’s knowing what not to build, and having the conviction to defend that no.
A team discovering an industry in real-time has no basis for saying no. Every feature looks equally plausible because they’re all equally new. The buyer-builder has already filtered. The thinking happened in the field, not the backlog.
Infinite builders with zero editorial authority is a coordination problem. One builder with twenty years of editorial authority is a company.
The market is about to discover the difference.