AI Makes Code Cheap. Judgment Becomes the Moat.

There's a popular anxiety making the rounds right now: “If AI makes software cheap enough, especially with vibe coding and agent-assisted development, founders will not be able to build defensible companies.”

I get it. If anyone can prompt their way to an app, the natural follow-up is: what's left to build, and what exactly needs funding?

If you build anywhere content ships, you already know the lens I care about here. No one does cartwheels because you wrote code. They do cartwheels when it shipped, it worked, it shipped on time, and it did not break everything downstream. That means dealing with assets, rights, security, deadlines, collaborators, and the chaotic reality of production and distribution.

That's why I don't think “apps get cheap” is the end of the story. It's the start of a shift.

Six Claims That Matter

1) AI makes building cheap. It does not make winning cheap.

Lowering the cost of implementation is real. Copilot and its successors reduce friction and increase throughput. The cost to ship code is dropping.

But winning was never primarily about typing. Winning is distribution, trust, workflow ownership, integration reality, and human judgment about what is worth solving. In content businesses, the durable advantage is not “we built it.” It's “we shipped it, it holds up, and it delivers.”

When implementation gets cheap, value doesn't disappear. It just moves.

2) If “apps go to zero,” value moves to the habitat that makes autonomy sustainable.

One of the most important points is also the easiest to miss: agents writing code is the easy part. The hard part is everything around it. The habitat is the governed environment where autonomy can run near real assets without creating existential risk.

For content workflows, the equivalent habitat is the governed runtime where autonomy can safely touch both planes of the business: the creative plane (files, metadata, captions, thumbnails, transcripts, derivatives) and the control plane (identities, permissions, accounts, release rules, rights, monetization).

3) The hard part is not building it. It is making it dependable.

AI can generate plausible software. The catch is that someone has to define what “good” means. Evals are the rules of the road: what passes, what fails, and when a human steps in. The moat is doing that continuously, not once as a great demo.

4) The product shifts from “an app” to “a problem owned.” Pricing shifts with it.

“Apps go to zero” is largely a feature pricing argument. If apps become abundant, feature differentiation and seat-based pricing likely get squeezed. But content businesses have always bought outcomes. Does it grow engagement? Do people return? Does it reduce churn?

Founders: don't ship “a tool.” Ship a system that takes responsibility for a measurable chunk of work and moves a measurable business outcome.

5) Operators become builders.

The bigger shift is who gets to build in the first place. Operators and creators don't have to wait for permission to prototype. We can build the first version close to the work, then iterate fast enough to discover what holds up under deadline pressure.

The job changes too. Operators stop being ticket writers and become system directors. We define the workflow, the guardrails, the quality bar, and the handoffs.

6) Security and governance become gating functions.

As soon as agents can act, change code, deploy, and touch data, the blast radius goes up. In content businesses, leaks, IP exposure, unlicensed assets entering workflows, untraceable generated media, and rights violations are existential risks. Governance becomes a core part of the product, because someone still has to be accountable.

Build the Prerequisites That Unlock What Comes Next

If you're building in the prompt-to-prototype era, “we can build it” is no longer a moat. Moats that still count: run it safely, make it dependable, win the workflow, earn trust at scale, own the commercial outcome.

The real story is not about founders building the same old products faster. It's that they can now prototype new categories of experiences cheaply enough to exist, then make them real, safe, and profitable inside the constraints that matter.

So let's not ask “what can I build better, cheaper, faster?” The question is: what entirely new categories of experiences become possible to invent that we could not afford to invent before?