An essay for leaders of growth-oriented companies.
You have been ignoring the AI conversation for three years, and your instincts have been right.
The hype cycles, the conference keynotes, the LinkedIn posts about transformation by Tuesday — none of it deserved your attention, and you were correct to refuse to give it. Most of what has been sold under the banner of artificial intelligence since 2023 has been marketing dressed as capability, demos dressed as deployment, and promises dressed as proof. The operators who waited for the noise to pass were making the same call their predecessors made in 2016, and 2018, and 2020, and they were right every time.
But something has changed in the last twelve months, and the change has been quiet enough that the operators who got the previous calls correct may, for the first time, be getting this one wrong.
There are companies in your industry right now that have stopped talking about AI and started operating with it. Their month-end close takes three days instead of twelve. Their client onboarding happens in hours instead of weeks. Their senior people spend Tuesday morning on advisory work that generates revenue instead of on data entry that generates nothing. Their partners serve forty percent more clients without working longer hours. There will be no press release. No celebration post. They are simply becoming better — structurally, permanently, compoundingly better — and the gap between them and the companies that have not moved is widening every month.
This is the separation that matters. Not the separation between companies that have heard of AI and companies that have not. Every managing partner in the country has heard of AI. The separation is between the companies that have installed AI as operational infrastructure and the companies that are still thinking about it. That separation is already underway. The question is whether you can see it from where you are sitting.
You have heard arguments like this before. You have heard the technology is here, the window is closing, you must act now speech in 2016 about chatbots, in 2018 about robotic process automation, in 2020 about digital transformation. Every cycle carried the same urgency, and every cycle, patient operators were rewarded for waiting. The technology was immature. The vendors were ahead of the reality. The correct response was patience.
So why should this time be different?
Because for the first time in the history of this conversation, the technology can do the work. Not the work of summarizing the work. Not the work of suggesting how the work might be done. The actual work. The difference between 2016 and now is not one of degree. It is one of kind. The chatbots of 2016 could simulate a conversation. The systems being deployed today can execute a workflow, end to end, with audit trails, with error rates lower than the human processes they replaced. That is not an incremental improvement.
The shift is not gradual. It is categorical. Three reference points:
First, Salesforce announced in 2025 that it had replaced approximately 4,000 customer-support agents with agentic AI systems performing the full ticket lifecycle.
Second, Cognition's Devin, deployed at Goldman Sachs and Infosys, enables a single senior engineer to perform work that previously required a five-person team.
Third, in February 2026, Block (formerly Square) cut nearly half of its 10,000-person workforce, with CEO Jack Dorsey stating publicly that AI had made many of those roles unnecessary, and predicting that “within the next year, the majority of companies will reach the same conclusion.”
These are not pilot programs. They are deployed systems performing measurable work at production scale, with audit trails and error rates lower than the human processes they replaced.
And the companies that have recognized this shift are not waiting for permission. They are moving.
But here is where the conversation usually goes wrong, and where this essay departs from every other piece of writing about AI you have read this year.
Most of what has been written about AI in the last twelve months frames the question as cost reduction. The story is straightforward: deploy AI, replace some of your workforce, capture the savings, improve your margins. Block laid off five thousand people. Salesforce replaced four thousand support agents. Goldman's engineering teams are running at a fraction of their previous headcount. The press has filed these stories under “AI is finally working,” and most leadership teams reading them have absorbed the same lesson: the play is to lay people off.
That lesson is wrong. Not soft-wrong, in the sense of being unsophisticated or short-sighted. Wrong wrong, in the sense of being economically self-destructive at the level of the entire industry, and morally bankrupt at the level of the company.
When firms in the same industry compete to automate their workforces, they enter what the paper's authors describe as an arms race. Each firm sees the savings from its own layoffs. Each firm's CFO sees the margin improvement. What no firm sees, until it has happened to all of them at once, is that they have collectively dismantled the consumer base that pays for their services.
A formal economic paper modeling what happens when many firms in the same market simultaneously automate their workforces. The authors construct a task-based competitive model in which each firm decides what fraction of its workers to replace with AI. Automated tasks save costs. But laid-off workers are also consumers, and their lost spending reduces revenue across every firm in the market.
The paper proves, as a formal equilibrium result, that each firm's profit-maximizing level of automation exceeds what would be collectively optimal — because the firm captures the full cost saving but bears only a fraction (1/N) of the resulting demand destruction.
The over-automation wedge is given by the formula αNE − αCO = ℓ(1 − 1/N)/k, where ℓ is the demand loss per displaced worker, N is the number of firms competing in the sector, and k is the integration friction. The wedge is strictly positive under any competitive market structure with N ≥ 2. It vanishes only under monopoly.
The paper further proves that this is a dominant-strategy equilibrium: no individual firm has an incentive to restrain its automation rate, even knowing that collective restraint would raise everyone's profits.
The math is unforgiving. The paper proves this is not a temporary disequilibrium that markets will sort out. It is a structural property of competitive automation. The arms race intensifies as AI improves. More capable AI does not solve the problem; it widens the distortion. The authors call this the More fluent AI agents, more capable models, more reliable tooling — each of these makes the layoff trap deeper, not shallower. The firms that have decided their AI strategy is “lay people off and capture the savings” are not preparing for the future. They are accelerating into a wall their competitors are also accelerating toward, and arriving slightly faster does not change what happens at impact.
The standard intuition is that firms benefit from cost-reducing technology. This holds when one firm automates in isolation — the firm captures cost savings, and aggregate demand is approximately unaffected. The intuition fails when many firms in a market automate simultaneously.
Consider the symmetric case: every firm in a market of N firms automates the same fraction α of its workforce. Cost savings per firm scale with α. But aggregate demand falls by ℓNα, and each firm's revenue falls by ℓα.
The paper proves that for any market with N ≥ 2 firms and incomplete income replacement (η < 1), the symmetric equilibrium produces lower aggregate profits than the cooperative optimum. This is not a transfer from workers to firms — it is a deadweight loss, harming both.
The paper notes specifically that “even a planner who places zero weight on worker welfare would reduce the automation rate below the equilibrium level.” This is the unusual property of the trap: firms don't just hurt workers; they hurt themselves.
Named after the chess piece in Through the Looking-Glass who must run faster and faster to stay in the same place.
In the context of the paper, the term refers to the result that as AI productivity improves — that is, as the per-task cost saving s grows — the gap between the privately optimal automation rate and the collectively optimal rate widens, not narrows.
Each firm perceives a market-share gain from automating beyond rivals, but at the symmetric equilibrium these gains cancel, leaving only the additional distortion.
The implication: improvements in AI capability do not relieve the pressure on firms to automate aggressively. They intensify it. Every advance in agentic AI makes the trap deeper.
This is the part of the AI conversation that almost no one is having out loud. The press will not file this story. It is too inconvenient for too many of the companies whose AI strategies are now publicly committed.
Now, here is the harder claim — the one that is going to separate the leaders this essay is for from the leaders it is not.
Show me a firm whose AI strategy is to lay people off, and I will show you a leadership team that has stopped trying to grow.
I want to be precise about what I mean by this. I am not saying that responsible automation cannot include workforce changes. It can. I am saying that when a leadership team's first instinct on encountering the most significant technology shift of their professional lives is to ask how do we shrink?, they have revealed something about themselves that goes deeper than strategy. They have revealed a poverty of vision.
A company that is genuinely trying to grow looks at AI and sees something different. It sees an opportunity to take the people it has spent years training, evaluating, and trusting — the people whose institutional knowledge represents enormous unrecoverable investment — and redirect them at the parts of the business that compound. A company that is genuinely trying to grow has more it wants to do than its current capacity allows. Every senior associate redirected from data entry to advisory work is a senior associate who can serve clients the company could not previously serve. Every partner whose calendar gets back the four hours a week that used to disappear into reconciliations is a partner who can take on the new client, develop the new service line, mentor the next generation, or build the new market. The work that AI can now do was never the work that talented people should have been doing. It was the work talented people had been forced to do because no other tool was capable of doing it.
The leadership team that sees this clearly does not respond to AI by laying people off. It responds by finally getting to do the things it has wanted to do for years. The talented staff are redirected toward growth. The growth pays for itself many times over. The company scales. The clients are better served. The talent stays — because finally, they are doing work worthy of their expertise.
The leadership team that sees AI and immediately reaches for the layoff list is telling you, without realizing it, that the company has no growth thesis. There is no new market they want to enter. There is no new service line they want to build. There is no client base they want to deepen relationships with. There is no expansion they want to fund. The only lever they can imagine pulling is spend less. And so when a tool arrives that lets them spend less, that is the only thing they think to do with it.
A company in that condition is going to fail. Not because of AI. Because the leadership had already given up on growth before AI arrived, and AI just made the giving-up cheaper to execute. They will save money for two quarters. They will reveal — to their clients, to their remaining staff, to the talent market — that they are in defensive mode. The best people will leave. The clients will sense the diminished bench. The competitors who chose differently will, eighteen months from now, be operating with deeper expertise, faster delivery, lower errors, and the kind of client relationships that come from having more capacity, not less. The shrinking company will find that its margin improvement was a one-time event and its competitive position has eroded permanently.
The math from the Layoff Trap paper agrees with the moral observation. The companies that win the next decade are the companies that use AI to grow. The companies that lose the next decade are the companies that use AI to shrink. The data, the math, and the moral logic all point at the same answer. There is a right way and a wrong way to automate, and the wrong way is what most of your competitors are about to choose, because their leadership teams have nothing else to reach for.
The Layoff Trap paper introduces a parameter, η (eta), representing the rate at which displaced wage income is recovered through reemployment, transfers, or upskilling into higher-value roles.
When η < 1 — the historical norm for displaced workers, who consistently suffer large and persistent earnings losses (Jacobson, LaLonde, and Sullivan, 1993) — automation destroys aggregate demand and the over-automation trap activates.
When η ≥ 1 — when displaced workers move into roles paying as much or more than the roles they lost — the externality flips: automation can create demand rather than destroy it. The paper notes that “in that regime, displaced workers are already thriving in higher-paying roles.”
The firm that automates by laying people off chooses η ≪ 1 for those workers. The firm that automates by redeploying its people into higher-leverage work chooses η > 1 for those same employees. The economic mechanism and the moral choice are not separate questions. They are the same question, expressed in different vocabularies.
If you are a leader who is reading this and quietly recognizing a different impulse in yourself — an impulse not to shrink, but to finally do the things the company has wanted to do but never had the capacity to do — then this essay is for you, and the rest of it explains what your competitors have already started doing, and what it will take for you to do it correctly.
If you are a leader who is reading this and looking for permission to lay people off cheaper — this essay is not for you. We will not be the firm that helps you do that. There are vendors who will. Find them. We wish you the best.
The path that works is not complicated. It is, in fact, far simpler than the path that doesn't.
A company that has decided to use AI for leverage rather than reduction begins by mapping where its talented people are spending their time. Almost without exception, a substantial fraction of that time is being spent on work that is below their talent — work that exists because the tools available to them have been inadequate to the company's ambition for the past two decades. Reconciliations. Document preparation. Compliance checking. Workflow handoffs that require humans to translate between systems that should have been talking to each other. Client onboarding paperwork. Routine reporting. The list is long, and at every company it is similar, and at every company the cost of that work is invisible because the work is doing what it has always done, paid for by the people whose talents are being underused to do it.
That work goes to AI. Not in a single dramatic deployment. In a sequence of small, targeted, measured deployments that begin in the first month of an engagement and continue every month thereafter. Each deployment frees up specific hours from specific people, and those hours get redirected at specific growth opportunities. The company does not shrink. The company's capacity to do meaningful work expands. And because the new meaningful work generates revenue at a rate the displaced grunt work never could, the company's economics improve dramatically without anyone losing their job.
By Month 18, the company that began correctly is operating with cost structures, delivery speeds, capacity per partner, and client experiences that competitors starting in Month 18 cannot replicate. Not because the technology is secret. Because the organizational learning is eighteen months deep. Late movers can buy the same software. They cannot buy the year and a half of accumulated, validated, refined operational practice. That cannot be compressed. It cannot be skipped. It can only be earned, one Week 3 build sprint at a time, and the companies that began earned it first.
The company that begins now installs operational AI infrastructure in Month 1, ships the first wave of automations by Month 3, and has measurable workflow leverage by Month 4.
By Month 12, the company has accumulated something that cannot be purchased: institutional muscle memory. Staff who have adapted their workflows around AI tooling. Internal champions who know how to identify the next opportunity. Adopted automations whose error patterns are now understood. Refined prompts and skill libraries tuned to the company's specific clients and data. Trained leadership conversations about which workflows to automate next.
None of this is technology, and none of it is acquirable in a single sprint. A competing company that begins eighteen months later will deploy the same technology, in many cases the same vendor, and find that they are eighteen months behind on the organizational curve — because the company that started early has been compounding institutional knowledge while the late one was waiting to see what worked.
The mathematical analogy is to compound interest. Time is the only input that cannot be manufactured.
This is the geometry that makes the moment matter. Not the technology curve. The compounding curve.
Let me describe what the companies that get this right will look like, eighteen months from now, so that the picture is concrete rather than abstract.
Their senior partners will have measurable, recovered capacity — capacity they have not had in years. They will be using that capacity to take on clients they previously had to turn away, develop service lines they previously did not have time to launch, and mentor the next generation of partners who will define the company's next twenty years. Their associates will be doing work worthy of their training — not because anyone gave a speech about it, but because the work that was beneath their training has been quietly absorbed by systems running in the background. Their month-end will have compressed by half or more. Their client onboarding will happen in hours, not weeks. Their delivery will be faster and more consistent than anything their competitors can match. Their error rates will have dropped to levels that would have required twice the staff to achieve through purely human effort.
Their leadership will have a dashboard, but not the kind of dashboard that consultants typically build. Not a dashboard of activity. A dashboard of trajectory. Revenue per employee, trended monthly. Cycle time compression by major workflow, charted as a curve. Cumulative hours redirected from below-talent work to revenue-focused work, expressed in the company's blended hourly rate. Capacity per partner, growing year over year. Backlog of identified value remaining to capture, replenished every month. The leadership will look at this dashboard at the start of each month and see the company's transformation made visible — not as a one-time event, but as a permanent operating reality.
Their best people will be staying. Not because of retention bonuses. Because the company has finally become the kind of place where talented people get to do the work they trained for. The talent market will notice. Their hiring will become easier, not harder. Their reputation in their industry will shift — not loudly, but unmistakably — from a company that competes well to a company that competes from a different category.
Their clients will notice too. Not in the form of explicit announcements about AI, but in the form of a company that just feels different to work with. Faster turnaround. Fewer errors. Senior people more available. Strategic conversations rather than reactive ones. The clients will tell other potential clients. The referrals will compound.
These companies will not look, to their competitors, like they got lucky or chose the right vendor. They will look like they figured something out — and they did. But what they figured out was not a technology. It was a decision. They decided to grow, and they used AI to make the growth possible.
A comet is on its way.
That is not a metaphor for technology adoption curves. It is a description of what happens to industries when a categorical capability shift collides with leadership teams that respond to it correctly and leadership teams that respond to it badly. The capability is here. The shift is real. The window for choosing how to respond is open right now, and it is closing. Eighteen months from now, the choice will have been made, by every company in your industry, whether their leadership realizes they were making it or not.
We know where the comet will hit. We know when. We know what the impact will look like, because we have been watching it begin in industry after industry for the last twelve months. We have read the papers, we have studied the companies that have already moved, and we have built the operation to install the leverage that the companies that survive will need.
We want to make sure you survive. More than that — we want to make sure that on the other side of the impact, you are one of the dominant players still standing. Not because we believe in your company more than you do. Because we believe that the companies whose leadership teams are still trying to grow, still trying to build, still trying to be useful to their clients and their staff and their communities — those companies deserve to be the ones standing. And we have built our operation to make sure they are.
The competition that will be destroyed will only be destroyed because they chose not to plan. They chose not to take the very step you took to get here, to this lunch meeting, to this page, to this conversation.
The future does not belong to those who wait. It belongs to those who choose to act, and who choose correctly when they do.
We are here for the leaders who choose to grow.
If that is you, we should talk.
If anything in it landed — the diagnosis, the moral architecture, the math — the next step is a lunch. The two of you, an hour, an unhurried table. Whatever the rest of the relationship looks like, it begins with one good conversation, and there is no other commitment attached to it.