Stephen Messer is co-founder of Collective[i] and Intelligence.com, and has been writing about the AI economy on Artificial CommonSense at reloadnyc. This column synthesizes much of Messer’s recent writing, and is related to several others, including “What It Means to Be AI-First,” “The Oldest Trick in Management Just Stopped Working,” “The Weakest Link,” “The Next Computer Is Alive,” and “The Death of Privacy. The Rise of Unbreakable Communications.”
Most companies think they have an AI strategy.
They have licenses. They have pilots. They have a chief AI officer, an oversight committee, a vendor roadmap, and a slide deck that says “responsible innovation” in a reassuring font.
What they do not have is a different company.
Their salespeople still type into CRM systems. Their managers still spend half their weeks gathering information from one team and relaying it to another. Their customers still wait while work moves through the same chains of approval. The old workflows remain. The old hierarchy remains. The old software architecture remains. AI has simply been added to it.
That is not transformation. It is decoration.
I have called this the “AI Shuffle”: the corporate habit of exchanging one technology logo for another while preserving every underlying assumption about how work gets done. It feels like progress because it generates activity. It does not produce an advantage.
The companies that pull away in this transition will start from a much more difficult question: What work should no longer exist?
Not: How can AI make this process 10% faster?
Not: Which chatbot should we license?
Not: How many employees are using the tool?
What can we delete? What decisions can move closer to the customer? What information no longer needs to be collected, reconciled, summarized, and passed up a chain of people before anyone acts?
That is the difference between adding AI to a company and becoming an AI-first company.
Start with subtraction
The conventional corporate response to a new technology is addition. Add a tool. Add a dashboard. Add a project team. Add a layer of governance. Add another system to the stack.
But the first instinct of an AI-first company should be subtraction.
In “The Art of Subtraction,” I argued that companies should question every requirement, remove unnecessary steps, simplify what is left, and only then automate. That sequence matters. Automating a bad process does not make it a good process. It makes the bad process faster, harder to see, and more expensive to unwind.
Take sales forecasting. For decades, companies have asked individual sellers to enter projections into CRM, then asked managers to interpret them, then scheduled calls where leadership negotiates a number that everyone knows is partly theater. The data is late, incomplete, and distorted by incentives. The meeting exists because the system cannot observe the buying process directly.
The AI-era alternative is not a more elegant forecasting meeting. It is a system that analyzes the buyer’s actual behavior, market conditions, timing, relationships, and signals across the commercial process. The goal is not to make the old ritual more efficient. It is to make the ritual unnecessary.
That is why the companies winning with AI are playing a different game. They begin with a specific business constraint and a measurable outcome. They do not measure usage. They measure whether the constraint has moved.
Software is not the only thing at risk
This is why the AI conversation is not really about software.
Yes, traditional software is vulnerable. Much of the enterprise stack was built to organize human data entry: applications that store records, route tasks, generate reports, and help managers reconstruct what happened after the fact. AI agents will increasingly observe activity, maintain context, initiate work, and recommend or execute the next best action.
But software is not going down alone.
The management structures built around it are also being challenged. In “Software Is Not Going Down Alone,” I made the case that AI will pressure the layers created to gather information, translate it across functions, prepare it for meetings, and relay decisions downward.
That does not mean leadership disappears. It means that the leaders who create value will be different.
The people who will matter most are builders: people who understand a real business problem, can use technology to solve it, and are close enough to customers and operations to know whether the solution works. The people who lose relevance will be those whose role depends on preserving friction, controlling access to information, or managing processes no one would design from scratch today.
In “Find Your Builders. Or They’ll Leave and Start Without You,” I argued that too many companies have placed their AI future in the hands of people selected to prevent mistakes rather than create new capabilities. Governance matters. Security matters. But a company that treats every low-risk experiment as if it were a high-stakes autonomous decision will discover that its competitors have learned more while it was still approving a pilot.
The safest move in AI may be the one that makes you irrelevant. Responsible deployment does not require paralyzing every use case. It requires separating the applications that demand rigorous control from the ones where learning must begin now.
The real moat is above the model
The debate over AI is still trapped at the model layer: whose benchmark is best, who has the largest training run, whether a particular frontier company is overvalued.
Those questions matter. They are not the most important ones.
The models will improve. They will also proliferate. Open and closed systems will compete, prices will decline, and capabilities that once seemed exclusive will become available to more companies. The durable advantage will not come from having access to a model everyone else can rent.
It will come from what sits above it.
In “The Only Fight That Matters in AI,” I described that battleground as the orchestration layer: the systems that determine which model handles a task, retain context across work, connect intelligence to proprietary data, and learn from the outcomes of real decisions.
That is where lock-in lives. Not in a prompt. Not in an interface. Not in an employee’s temporary familiarity with a tool.
The moat is the learning system: a company’s ability to connect proprietary context, trusted relationships, operating data, and feedback from the market. This is why, in “Your Buyer Has a Process,” I argued that commercial intelligence must move beyond what a seller enters into a CRM. A buyer’s process unfolds across relationships, timing, incentives, and signals that no single sales rep can fully see.
The same is true of human networks. The old warm introduction was valuable because it compressed trust. But it was also opaque and dependent on gatekeepers. In “The Warm Intro Is Dead,” I explored how verified relationship intelligence can make that trust more visible and usable—if it is built with the right controls and consent.
This is bigger than the firm
AI is often discussed as a workforce issue or a technology-budget issue. It is neither. It is an institutional issue.
The systems that govern housing, infrastructure, energy, capital formation, communications, and privacy were designed in the same pre-AI world as corporate hierarchies: a world where collecting and interpreting information was slow, expensive, and centralized.
That is why permitting matters. In “Time Kills All Deals,” I argued that America’s permitting machinery has become an economic bottleneck. The point is not to automate judgment away. It is to eliminate the administrative drag that turns building a home, opening a business, or investing in infrastructure into an endurance test.
It is also why the AI infrastructure buildout deserves more serious attention than the usual bubble-versus-no-bubble debate. In “The Trillion-Dollar Trade Wall Street Isn’t Seeing,” I argued that data centers, power, and compute capacity are not simply costs attached to a speculative technology cycle. They are strategic options on the next industrial architecture.
And it is why we should resist simplistic narratives. Circular capital flows can create excess, as I wrote in “The Most Expensive Money in the Room.” But it is possible for financing structures to be frothy and for the underlying transition to be real. The important question is what survives if the financial enthusiasm recedes: infrastructure, skills, proprietary intelligence, and operating capabilities—or merely expensive stories.
The choice in front of leaders
Every company now faces the same choice.
It can use AI to preserve yesterday’s institution: the same departments, workflows, data silos, approval chains, and management rituals—just with a more impressive interface.
Or it can use AI to build the company that should have existed all along: one that sees more, learns faster, acts closer to the customer, and spends less time administering work than creating value.
The first path will produce plenty of announcements.
The second will produce a widening gap between companies that appear to be adopting AI and companies that are actually being remade by it.
The window to choose is open now. It will not remain open indefinitely.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
Suggested author disclosure: Stephen Messer is co-founder of Collective[i] and Intelligence.com. The views expressed are his own.
For publication, I would also consider adding a linked endnote module—“Read the related Artificial CommonSense columns”—with the remaining pieces, including “What It Means to Be AI-First,” “The Oldest Trick in Management Just Stopped Working,” “The Weakest Link,” “The Next Computer Is Alive,” and “The Death of Privacy. The Rise of Unbreakable Communications.”
This story was originally featured on Fortune.com
