Not gradually. Not in some theoretical future. Now.

The floor is falling out of the code labor market in real time, and the people most exposed are the ones who never noticed they were renting their value from a skill that was about to become a commodity.

Here is the uncomfortable reframe… cheap code does not kill software. It kills the middlemen of software. It kills the generic builder, the undifferentiated agency, the feature factory with no moat, no audience, and no data. It kills the business model of “someone needs to write this, and I am the one who can.”

That model is done.

Klarna already lived this story. They pushed hard on AI-led cost cutting, then reversed course, not because the technology failed, but because raw automation alone is not a business. The savings were real. The growth and product quality and customer trust they sacrificed to get there were realer.¹

McKinsey is projecting meaningful compression² across traditional tech services from agentic AI, and simultaneously pointing at where the winners are moving… data, cloud, product engineering, higher-order integration.

Not cheaper labor. Better leverage.

So what survives? What actually holds value when writing software is no longer the hard part?

The Seam

Position yourself where code is cheap, but consequences are expensive.

That is the seam. Everything interesting happens there.

When software gets easier to produce, the scarce assets shift. Attention becomes scarce. Trust becomes scarce. Embedded reach becomes scarce. Judgment about what to build and why, that gets more valuable, not less. The people who can own a niche, own a workflow, own a data surface that others return to, they compound. The people who can only produce output do not.

Six positions hold up well. Here is how I think about each.

1. Own Distribution

If everyone can spin up an app, the scarce asset is not the app. It is the audience. The channel. The embedded workflow people already live inside.

AI-native business models are being framed less as cost stories and more as revenue and business-model stories precisely because this is true. Distribution is the durable moat. Own a niche audience. Own a brand people trust when outputs actually matter. Own the workflow that sits between a person and their daily consequence. That is worth something that does not cheapen at the rate of tokens.

2. Own Proprietary Data or Feedback Loops

Cheap code does not erase the value of exclusive data. It does not erase clean telemetry, domain-specific labels, or the compounding edge of a product that improves from real customer behavior.

If everyone can generate an app, the one with better data and tighter iteration loops wins. Full stop. Build products that collect structured usage data. Build systems that improve from what users actually do. Turn messy workflows into reusable signals. Create internal datasets others cannot clone. The generation layer commoditizes. The data layer does not.

3. Become the Verifier, Not Just the Generator

Generation gets cheap. Verification gets expensive.

Enterprises can tolerate cheaper code. They cannot tolerate silent failure at scale. Security, QA, compliance, observability, auditability, model output verification, these rise in value as generation volume explodes. McKinsey’s work on agentic AI in services specifically flags value shifting toward higher-order integration and risk-bearing work. That is the other side of the automation trade.

If you want to sit in a durable place: AI evaluation, regression detection, compliance pipelines, security review, migration validation. The world is about to produce a lot of code it cannot fully trust. Someone has to check it. That someone is not a commodity.

4. Integrate Messy Reality

The farther software gets from the screen and into money, health, logistics, operations, and regulated flows, the less “zero-cost software” applies in any meaningful sense.

Connecting old systems to new agents, bridging regulated industries to modern UX, wiring operations to automation, controlling physical systems with software, this stays hard. It stays expensive. It stays sticky. Industrial and autonomy stacks are drawing serious attention right now for this reason. AI matters most where it can control expensive real processes, not where it generates another dashboard.

Legacy integration is not glamorous. It is also not going away.

5. Become the Product Person With Technical Leverage

Pure coding labor gets squeezed first. People who can define the right problem, shape the workflow, and direct AI systems toward commercial outcomes get more valuable.

Klarna’s own pivot, from pure cost-cutting back toward growth and product quality, is evidence that raw automation alone is not the endgame. Someone still has to decide what to build, why it matters, and whether it is working. That person needs to get closer to revenue. Closer to customer pain. Closer to decision rights. Use AI to multiply throughput. Do not let it define your worth.

6. Build in Hard Verticals

Horizontal SaaS gets pressured first. It is easiest to clone. Vertical software, tied to domain nuance, compliance, and embedded workflows, tends to hold.

Banking, healthcare, industrials, logistics, energy. These are slower to replace because the switching costs and liability costs are real. McKinsey’s work on banking and services reinforces that inertia is a real economic force, but once it cracks, the advantage goes to whoever understands both the domain and the AI layer. That combination is not easy to fake or replicate cheaply. It is built from years of operating in the vertical, not just knowing how to code.

The Practical Positioning

Bad place to be: generic app developer, undifferentiated agency, surface-level SaaS with weak switching costs, feature factory with no owned audience or data.

Good place to be: domain expert with AI leverage, operator with proprietary workflow access, product builder with distribution, trust, and compliance layer, migration and integration specialist, owner of recurring high-quality data.

The delta between those two columns is not technical skill. It is whether your value is embedded in something that compounds, or whether your value is rented from a capability that just got commoditized.

The Blunt Version

Move up the stack from code to consequences.

Code will cheapen. Judgment, trust, distribution, and embeddedness will not cheapen nearly as fast.

For anyone technical… use AI aggressively yourself. Stop selling hours of coding. Attach yourself to revenue, risk reduction, or unique data. Build assets that compound: audience, workflow, dataset, brand, infrastructure. Avoid businesses where the customer mainly values “someone to write the software.”

That business model is not dying slowly. It is getting disrupted from below by something that works 24 hours a day for the cost of an API call.

The people who win this transition are not the ones who resist that. They are the ones who positioned themselves somewhere that could not be reached by it.

Do not compete on software creation.

Compete on software ownership, verification, and embedded advantage.

That is the play.