This is not an exaggeration. By the McKinsey Global Institute digitization index, construction ranks second-to-last in the US and last in Europe across all economic sectors. The data is a decade old, but what matters more is the digitization trajectory by sector, because the trajectory itself barely changes over time.

Labor productivity in construction grew just 0.4% per year from 2000 to 2022, compared to 2% for the global economy overall and 3% for manufacturing. In a sample of countries, fewer than a quarter of construction firms managed to match even the broader economy's productivity growth over the past decade. If construction productivity caught up to the economy's average, it would add $1.6 trillion annually and roughly 2% to global GDP.

Think about that number. $1.5 trillion sitting on the ground, and no one can pick it up. This happens because of structural reasons that make construction fundamentally different from the rest of the economy. Every project is unique. You cannot scale the construction of a bridge the way you scale a SaaS product, because the participants are fragmented. The architect, the structural engineer, the MEP engineer, the general contractor, dozens of subcontractors, the client, each working in their own lane with different standards, different file formats, different ideas of what a finished product looks like. Building codes, fire codes, environmental requirements, seismic standards, all of it varies from jurisdiction to jurisdiction.

Below is a McKinsey map of how deeply technology has penetrated construction. It shows that key digital tools already exist and have proven their value, but deployment is fragmented and has not reached scale. The top layer shows solutions that have become standard practice: digital documents and basic project management. Just below are more advanced tools like Building Information Modeling, widely used but still not covering the entire industry. Further down sit technologies with obvious productivity potential (AI, 3D printing, IoT) whose adoption remains selective. And at the very bottom sit robotics, marketplaces, AR/VR, still barely scaled.

Construction companies have historically spent less than 1% of revenue on IT. That is three times less than automotive or aerospace. And finally, the cost of an error in construction can be a collapsed bridge, a building fire, or a flooded tunnel. The feedback loop in coding might take minutes or hours. In construction it takes years. When Claude Code writes code, you can run it, test it, roll it back. But if an engineer designs a load-bearing structure, you cannot roll it back after the building has been commissioned. These factors create the paradox I will explain below.

Autodesk

I have previously divided software into two types. The first covers deterministic systems like CRM, ERP, and accounting, where absolute accuracy, rule compliance, client data integration, and proven business processes are required. The second covers probabilistic systems, pattern recognition, content generation, and analytics widgets. If AI reproduces those at 90% quality for 10% of the cost, the business collapses.

Autodesk fits neither category. It is software whose output exists not in digital space but in the real world. A building, a bridge, a turbine, a city block. And that connection to physical reality creates the paradox that protects Autodesk from AI-driven devaluation. AI disrupts SaaS because it can reproduce its output. Export the tables to CSV, upload to any AI service, get charts, analysis, and forecasts in minutes. The features people paid tens of dollars a month for become worthless. But AI cannot reproduce a building or generate a turbine in a chat window, because the physical world does not accept probabilistic answers.

Read more

That model is now breaking down, because the nature of the user has fundamentally changed. An AI agent isn't a person and doesn't occupy a seat. It can do the work of ten analysts simultaneously without triggering a single additional license under the old logic. IDC projects that by 2028, the pure seat-based model will be obsolete and 70% of vendors will be forced to rethink their value proposition as AI agents displace manual labor. Forbes ran a headline in March 2026: "AI Is Killing The Seat License. Venture Capital Isn’t Ready For What’s Next."

We've watched a sweeping correction across software companies against this backdrop, regardless of their financial health or their capacity to integrate into the new technological axis. So in this piece I want to try to answer a few questions.

• Which segment of the software industry is most insulated from this devaluation?

• Why does the noise around Claude Mythos represent an opportunity and not a risk?

• And finally, how do you get there first?

In “Equilibrium” I laid out the Caballero model and the three stable states of the IT ecosystem, arguing that we’re currently inside Region II, the AI deployment zone where returns on capital hold at a plateau. In “SaaS Armageddon” I broke down how that same AI is undermining the seat-based model of enterprise software, shifting value from human hours to machine time. But in both pieces I deliberately left one sector untouched. One that behaves in a fundamentally different way from the rest of software.

Cybersecurity.

The Sector That Feeds on Chaos

Read more