Narrative Reckoning
Why the market is punishing CEOs who can't tell the right AI story
Every CEO I talk to says they have an AI strategy.
Most of them don’t.
That gap shows up in the market. Many companies are getting punished. A few getting rewarded. It isn’t about who is “using AI.” It’s about whether leadership can explain where value comes from now. It is worth repeating that markets don’t pay for value. They pay for the story about value. The market is already doing its own sorting, narrative haves on one side, narrative have-nots on the other, and most leadership teams are on the wrong side without knowing why.
Old mental model still running
For thirty years, enterprise software sold a simple promise. We give you a tool, you do the work. Databases, CRMs, ERPs, analytics dashboards. Software enabled. Humans solved. Every SaaS company on earth was built on this mental model. Per-seat pricing reflected it perfectly. Each seat was a human doing the work.
That mental model is now wrong.
Aaron Levie put it cleanly. “In an era where the software, because of AI, is solving the problem for the customer and not just enabling a solution, the ceiling gets blown up.”
This is the shift from enabling to solving. And it’s the reason the market is repricing everything.
When a CEO gets on an earnings call and says “we’re AI-first” or talks about “summarization” use cases, the market hears the old mental model. Enabling. Tools. Features bolted onto existing workflows. The enterprise equivalent of “we have a website” in 1999.
When Alex Karp talks about Palantir, he talks about an organizational architecture where AI doesn’t assist humans but operates alongside them. Making decisions, executing workflows, surfacing only the exceptions that need human judgment. That’s solving. When Ravi Kumar at Cognizant says “you make the agentic capital permanent and unleash people to be the variable component,” that’s solving. Both are telling the market: the software does the work now. Humans supervise.
CEOs are still telling the market: we added AI to help your people work faster. That’s a story about a better stapler. Market is paying for stories about replacing the department that uses staplers.
Abundance changed the physics
When intelligence is abundant, the companies that win aren’t the ones with the most muscle. They’re the ones with the best muscle memory, the ones that convert raw token power into specific, defensible outcomes. Value is moving up the stack. Processing commoditizes. The harness, orchestration, judgment, domain-specific knowledge, institutional memory, is where the margin lives.
I call this the choke point distinction. Basmati rice is a commodity. Real estate is also a commodity. But the biryani maker doesn’t need to own rice fields. The restaurant owner must own the building. Both are commodities. One is a choke point. The question every CEO should be asking: is our AI capability rice, or is it real estate? Is it substitutable, or does it compound?
Most software companies are selling rice and pricing it like real estate. The market can definitely notice that.
If you listen to earnings calls, three questions keep coming up. The stock price reflects whether leadership can answer them.
One: What is your Agentic strategy, not even your AI strategy ?
Most agentic projects are failing today. They fail because agents don’t understand how enterprises actually work. The tribal knowledge, the operating norms, the workflows and processes, the culture and judgment calls, the regulatory constraints. Ravi Kumar of Cognizant calls this context engineering. I prefer a different frame. Harness engineering.
The distinction matters. The model is muscle. Raw power. The harness is muscle memory, the encoded patterns, reflexes, and institutional knowledge that determine whether that power creates value or creates chaos. A bodybuilder has muscle. A surgeon has muscle memory. Both are strong. Only one you’d trust with a scalpel. Every enterprise needs agents with muscle memory, not just muscle. The harness, the memory, permissions, integrations, context, judgment rules, is where that muscle memory lives.
CEOs who can articulate their harness engineering strategy, how their company uniquely encodes institutional knowledge into agents, get rewarded. Those who say “we’re building agents” without explaining how those agents develop muscle memory are selling raw muscle in a world drowning in it.
Two: What is your model for value creation when intelligence is a commodity?
If your competitive advantage was providing intelligent capabilities, analytics, recommendations, automation, you have a problem. RAG frameworks, orchestration layers, vector databases, things that were custom-built eighteen months ago are now features of foundation models. The model layer is eating the application layer. When the model maker ships the interface, every SaaS company that thought they’d own the layer above just lost that bet.
The survivors own something the model layer can’t eat. Proprietary data that compounds. Workflow integration that takes years to replicate. Regulatory expertise that requires institutional knowledge. The trust of a specific customer segment built over decades.
Three: How do you build defensibility when capability commoditizes?
Intelligence capability is commodity now. A few big players will own the model layer. Everybody else accesses it like a utility. The question investors are really asking is whether your position gets stronger over time or whether someone rebuilds it next quarter with a better model.
A thin configuration layer on top of a foundation model is still rice. A harness that accumulates context, workflow control, and institutional memory with every customer interaction is real estate. It compounds.
Nobody’s redesigning work
Here’s the biggest blind spot. The future of work is humans and agents together. And this is not recognized by enterprises yet.
Everyone frames AI as either replacing jobs or augmenting workers. Both frames are wrong because they start from the current org chart and ask where AI fits. The right question: if you were building this company from zero today, with agents as a given, what does the org look like?
The manager of 2027 doesn’t manage twelve people. She manages three people and forty agents. What does performance management look like when your direct reports hallucinate? What does supervision mean when your workforce operates at machine speed? Who teaches the agents the tribal knowledge, the operating norms, the judgment calls that determine whether they create value or create liability?
That’s a new job category. It doesn’t exist yet. These aren’t employees in the traditional sense. If agents are agentic capital, permanent silicon workforce, then the humans engineering their harnesses are capital allocators. They’re encoding decades of tribal knowledge, operating norms, and judgment calls into systems that execute at machine speed.
Compensation models don’t work anymore. If one person supervising forty agents produces ten times the output of a traditional team of twelve, the old pay band makes no sense. You don’t train people to do the work. You train them to teach agents, supervise agents, and handle exceptions agents can’t.
The CEO who presents this vision on an earnings call, a specific, credible plan for how their organization becomes a human-agent enterprise, will be perceived differently.
Most boards were built for the old era. They think in seats, modules, annual contract value, and incremental feature releases. They ask “what’s our AI strategy?” when they should ask “what’s our human-agent operating model?”
The deficit shows up in three places. CXOs are running pilots and calling it progress. The stories are lame. “We’re summarizing documents with AI.” “We’re AI-first.” “Our copilot increases productivity by 20%.” Old world with AI sprinkled on top. Pricing hasn’t evolved either. If one AI agent does the work that used to need ten software seats, companies will ask why they’re paying for ten.
This is starting to show up
February market shakeup wasn’t the reckoning. It was the preview.
Palantir gets a 242x P/E because Karp tells a story about the operating system for human-agent institutions. Cognizant gets re-rated because Ravi tells a story about agentic capital converting infrastructure spend into business results. Both have proof points. Both are specific. Both give investors a mental model for how value compounds.
Most everyone else is adding AI features and calling it strategy. Running pilots and calling it progress. Telling stories about the old world with new technology.
Markets don’t pay for value. They pay for the story about value. I said that last time. The corollary is harsher. If you can’t tell a credible story about where value lives when intelligence is abundant, the market will tell a story about you.
If you don’t tell the market where your value is moving, it will decide for you