The ground is moving. Build a business that moves with it.

The pace is the story

Every few months, one of the frontier labs ships a model that can do something the last one couldn't. Sometimes it's reasoning. Sometimes it's writing code, using tools, or working for an hour without supervision. Each release resets what "state of the art" means, and each one lands on every competitor's desk on the same day.

That changes what a business advantage is. If your edge is "we have access to a smart model", you are renting it. So is everyone else. The edge lasts until the next release, and the next release is already being trained.

The wipeout is not theoretical

Markets are already pricing this in, and not gently.

In early 2026, software stocks sold off hard on fears that AI agents could do the work inside enterprise tools and hollow out per-seat subscriptions. Traders started calling it the SaaSpocalypse. By late March, the iShares software ETF had fallen more than 14% in that stretch, its worst since 2008, according to The Motley Fool. Some analysts at J.P. Morgan called the selling "broken logic", pointing out that investors were afraid of two contradictory things at once. Both views can be partly right. What matters is that nobody is sure, and uncertainty is exactly what a moving target looks like.

The damage is easier to see at company level.

Chegg sold homework help. In October 2025 it cut 388 roles, about 45% of its staff. Its shares were down 99% from the November 2024 peak, roughly $14.5 billion of value gone. The company blamed generative AI and the drop in search traffic from Google. A product people used to pay for became something a chatbot did for free.

Salesforce sells the software that others are afraid of losing to AI, and it re-engineered itself first. In September 2025 its CEO said support headcount had fallen from 9,000 to 5,000 after it put its own agents on the help desk. His words: "I need less heads."

Klarna shows the other edge of the blade. In February 2024 it said its AI assistant was doing the work of 700 customer service agents. By May 2025 its CEO said the company was hiring humans again, because cheaper AI support had produced "lower quality". He said Klarna wanted customers to know there would always be a person if they wanted one.

Read those three together. One company lost its market. One re-cut its own cost base. One over-rotated and had to walk it back. The target moves in every direction, and being early or late are both expensive.

Where the real value is

The winners aren't the ones with the flashiest model. They're the ones who can point to a number that moved.

Hiring. Unilever, working with Pymetrics and HireVue, reported cutting its hiring cycle from about four months to about four weeks, and saving more than 50,000 hours of screening and interview time. Treat those as company-reported figures, since they weren't independently verified. The direction is what matters. The win wasn't "we use AI". It was a hiring process that gave candidates an answer in weeks, and gave recruiters their time back.

Cost, through real-time analytics and root-cause analysis. Microsoft Research studied more than 40,000 cloud incidents across over 1,000 services. Even with 2023-era models, more than 70% of on-call engineers rated the AI's suggested root causes and fixes as useful in a live setting (3 out of 5 or better). Models are far stronger now. The business case is easy to see: shorter outages, fewer engineers woken at 3am, fewer repeat incidents. Real-time data plus a model that reads it quickly is a cost line that moves.

Neither example is about a clever prompt. Both are about system design: where the data comes from, how it's cleaned, how the model's output gets checked, who steps in when it's unsure, and how the result lands inside a process people already run. The model is one box in that diagram. The rest of the diagram is what a competitor can't copy by signing up for the same API.

So what for your business

There is no moat in the old sense. Not for a startup, and not for a large company with a famous brand. Anything that's merely "AI-powered" is a feature a competitor can match in a quarter.

Three things still hold up:

• Your data and your context. What you know about your customers, your processes and your edge cases, in a form a system can use.
• Trust earned through results. A track record of the number going the right way, with the failures handled honestly.
• Speed of re-evaluation. How quickly you notice that something better exists and adopt it without breaking what works.

That last one is the real strategy. Your value proposition is a moving target, so it has to move on purpose. Plan to rewrite it as the market shifts. It may settle in two years. It may take ten. Nobody can tell you which, so build the company that survives either answer.

What to do this month

1. Write your value in one sentence a customer would pay for. Use a number: days to hire, cost per ticket, minutes to resolve an incident. If you can't, that's the first problem.
2. Pick one workflow and instrument it before you automate it. You can't prove improvement without a baseline.
3. Test model-swap readiness. Can you move this product to a different model in a week without breaking it? If not, you've built a dependency, not a system.
4. Build a small evaluation set from your own real cases. Fifty good examples will tell you more about a new model than any benchmark.
5. List what you rent and what you own. Models and tools go in one column. Data, workflow knowledge and customer trust go in the other. Invest in the second.
6. Put a quarterly review on the calendar. Measure the value, test the newest model, re-price the offer, ship what you learned. Repeat.

None of this needs a bigger budget. It needs a clear number, a short feedback loop and the discipline to review it on schedule. The companies that win this phase won't be the ones that predicted the future. They'll be the ones that got good at adapting to it.