What to Build When Anyone Can Build Anything

The price of “being able to build” has dropped

With the rise of AI coding tools, apps and utilities that used to demand expertise and time now take shape in a fraction of it. There’s a feeling that “anyone can build anything.” But it’s worth pausing here. If anyone can build it, then building alone no longer sets you apart.

Basic economics: the price of something whose supply suddenly surges goes down. If AI has exploded the supply of “building software,” then the value of “being able to build” falls structurally. So where does the value go? The answer is surprisingly simple — it moves to the opposite side of building: choosing, judging, being trusted, distributing, accumulating.

This piece reasons backward from that shift toward what is impressive — and advantageous — to build.

What gets cheap, and what stays scarce

First, sort out what AI makes cheap and what it can’t.

Gets cheap (commoditized)	Stays scarce (value rises)
Writing code	Deciding what to build (problem selection)
Building a demo / MVP	Not breaking in production (reliability, ops)
“Plausible-looking” mass output	Taste (the eye to select and finish)
General knowledge / information	Proprietary data, field presence, relationships
One-off output	Assets that compound over time

The law is simple: the cheaper generation gets, the more value shifts to the opposite of generation. Start from this asymmetry and “what’s advantageous to build” comes into focus.

Four sources of being hard to copy

Restated in a sentence, “advantageous to build” means “hard to copy even when it can be built.” In an era where anyone can build anything, advantage is nearly equal to hard-to-copy-ness. And its sources reduce to roughly four.

① Physical and regulatory reality (atoms)

Bits (information) get cheap; atoms (the physical, the real) do not.

• Things involving the physical world — hardware, robotics, manufacturing, energy, logistics
• Domains walled by regulation, licensing, safety — healthcare, finance, infrastructure, law
• Things that require messy real-world integration — fitting into existing operations, equipment, and people

Even if anyone can build the software, the part that touches reality is itself the barrier. A demo is a night’s work; embedding into a real operation takes time and credibility.

② Proprietary data and trust

Models converge, so differentiation moves to a proprietary data flywheel.

• Real-time / private / hard-won data that others can’t obtain
• Trust and relationships inside a specific industry (AI can’t be the trusted insider)
• A mechanism that accumulates data, improves accuracy, and raises switching costs the more it’s used

If everyone can use the same model, the contest moves to “what can you feed that model?”

③ Taste (the eye, editing, curation)

When generation is free, the bottleneck becomes selecting, refining, and the last 10%.

• The editorial skill to pick the good from a flood of generated output and polish it
• A consistent authorial voice and world — where “who made it” carries meaning
• Brand and trust — the authority of “if this person / this place says so”

In a world where mass production is free, the opposite of mass production — selection and finish — becomes scarce.

④ The compounding of time

Assets that can’t be copied overnight and can only be built by continuing.

• A corpus (data, works, records) accumulated over years
• A community or readership you’ve grown
• A reputation built up over time
• “Correctness” and “unbreakability” polished through years of dogfooding

AI hands speed to everyone. Precisely because of that, the accumulation of time itself — which speed can’t shortcut — becomes the difference.

What’s advantageous to build — categories

Lining up things that ride these four sources:

• Things that embed into the physical world — hardware / robotics / manufacturing / energy / logistics, regulated industries
• Things that spin a proprietary data flywheel — products that accumulate and strengthen with use
• Editing, curation, authorial voice — placing value on the opposite of mass output
• Reliability and operations that don’t break in production — see below
• Network effects, standards, communities — two-sided markets, protocols, standards (assets you can’t build alone)
• Genuinely new capability that pushes the frontier — new models / algorithms / science / language implementations themselves

“Doesn’t break in production” rises in value in the AI era

Reliability deserves a note. AI drastically lowers the bar for a demo, but not the bar for production. If anything, the more the world fills with “code that runs but can’t be trusted,” the more valuable it becomes to verify it and bring it to a trustworthy state. Testing and verification grow more important in the AI era precisely because of this asymmetry (a theme I also covered in The Lineage of Software Testing).

The “anything” in “you can build anything” usually means a recombination of what already exists. Building what doesn’t exist yet — and carrying what you built all the way to unbreakable — remains a different order of difficulty.

A filter for individuals — four questions

When deciding what to build, filter with these and you’ll miss less often.

1. Can AI copy it overnight? — If so, no advantage. Is there accumulation, proprietary data, trust, or field presence?
2. Was “being able to build” really the bottleneck? — If not, it’s more effective to own the true bottleneck (distribution, trust, problem selection).
3. Does it compound over time? — Is it an asset that grows stronger the more it’s used and continued?
4. Does it ride an unfair advantage only you have? — deep expertise, proprietary data, a unique perspective or voice.

Summary

In a sentence: build not “what can be built” but “what can’t be copied even when it can be built.”

Building one more app that anyone can build with AI rarely produces value. If you build, always layer on top of it one of: proprietary data / community / trust / connection to reality / depth you can only accumulate by continuing. The sources of hard-to-copy-ness reduce, most of the time, to (a) physical & regulatory reality, (b) proprietary data and trust, (c) taste, or (d) the compounding of time.

Ironically, the era of “anyone can build anything” is also the era in which everything other than building matters more and more. More than what you build, it’s what you accumulate around it that sets you apart.