On the morning of May 7, 2026, two things happened that should change how we're thinking about AI and jobs.
Cloudflare CEO Matthew Prince published a letter to employees before stepping onto his earnings call. It announced more than 1,100 layoffs, cited a 600 percent increase in internal AI usage over the prior three months, and stated that the founders were “reimagining every internal process, team, and role across the company.” Cloudflare’s stock fell roughly 15 percent as investors digested a story in which AI simultaneously delivered record efficiency and triggered large-scale restructuring.
That same morning, Challenger, Gray & Christmas, Inc. released its April 2026 job cuts report. The headline number, roughly 88,000 job cuts, up 38 percent from March, was not the most important data point. The buried lead was a short line in the methodology: for the second consecutive month, artificial intelligence was the leading cited reason for job cuts in America. AI was cited for 21,490 cuts in April alone, 26 percent of all monthly layoffs, and 49,135 year‑to‑date, making it the third‑leading cause of all 2026 layoff plans.
Andy Challenger, the firm’s chief revenue officer, offered the sentence every leaders should think deeply about: “Regardless of whether individual jobs are being replaced by AI, the money for those roles is.”
This is one of the most important shifts we're seeing as companies evolve into the agentic enterprise. It is not just a question of whether AI is actually doing the work. It is a question of whether the budget has already been moved. The CFO may have reallocated the money. The board may not know whether the function is still being performed.
Two things can be true
Two camps have been fighting about AI and jobs for the past few years. Both are largely correct. Neither is asking the question boards actually need to answer.
The AI Boomers: upside as default
The first camp belongs to the “AI Boomers” - the optimists, the ones who see all of the potential of AI and default to a glass‑half‑full view.
Andreessen Horowitz's co-founder Marc Andreessen is the emblematic voice here. In his Techno‑Optimist Manifesto, he argues that we are being lied to - that technology has never destroyed total employment, that the lump‑of‑labor fallacy embarrasses every generation that believes it, and that AI will function as a “brilliant friend” available to every person on earth, compressing centuries of progress into decades. The historical record largely supports this direction of travel: research from economists and institutions like Goldman Sachs finds that a majority of workers today are employed in occupations that did not exist in 1940 and that most net employment growth over the last 80 years has come from technology‑driven creation of new roles that no one had named in advance. The World Economic Forum’s Future of Jobs work projects a substantial net gain in jobs globally by 2030 from AI‑enabled sectors, even after accounting for displacement.
Sam Altman and Greg Brockman at OpenAI embody this hard‑boomer stance in the frontier‑model world: scale powerful systems, drive down costs, and assume that new categories of work will emerge as they are deployed. Jensen Huang at Nvidia tells a parallel story in silicon, framing AI as a new industrial revolution in accelerated computing that will “rewrite” every industry and generate massive productivity and employment gains as enterprises replatform on GPUs. Reid Hoffman, in his “bloomers vs doomers” framing, leans into the idea of AI‑augmented founders and workers as the engine of abundance, treating AI primarily as a growth and opportunity story.
Even Elon Musk, who often sounds alarms about long‑term AI risk, behaves like a boomer in practice at xAI - pushing aggressively to build more capable models and take them to market quickly. In this camp, the implicit governance assumption is that if boards back the right AI strategy, the labor market will eventually take care of itself.
The optimist case points to roughly 250 years of economic history. It is compelling story...over the long run.
The AI Doomers: structural and existential risk
The second camp is what you might call the “AI Doomers.” They are not anti‑technology, per se. They are focused on who captures the gains, who absorbs the costs, and what happens if governance fails at scale.
Daron Acemoglu, the MIT economist and 2024 Nobel laureate, is a central figure in this camp. His research shows that institutions - not technology itself - determine how the gains from innovation are distributed. He emphasizes that every transformational technology in history has eventually delivered widely shared prosperity, but never automatically and never quickly. The Industrial Revolution created unprecedented wealth along with child labor, 14‑hour shifts, and urban slums; productivity gains flowed to capital until institutions forced redistribution.
Acemoglu’s specific critique of current AI deployment is what he calls “so‑so automation”: systems that displace human labor without generating proportional productivity gains or new task creation. His estimates suggest that the net contribution of AI to productivity over the next decade could be modest - fractions of a percentage point of GDP - even as firms use it to justify large layoffs and wage pressure. In 2025, U.S. companies announced roughly 1.2 million layoffs, up significantly year‑over‑year, with tens of thousands of roles explicitly attributed to AI.
Around Acemoglu are the x‑risk‑oriented doomers. Eliezer Yudkowsky has argued for extreme slowdowns or moratoria on advanced AI development, warning that misaligned superintelligence could pose an existential threat. Yoshua Bengio and Stuart Russell warn that misaligned or poorly governed AI can both hollow out middle‑skill jobs and create systemic safety risks that current institutions are not ready to manage. Max Tegmark and PauseAI US activists press for aggressive constraints and global governance mechanisms before scaling frontier systems further.
In this camp, the governance assumption is reversed: unless boards and policymakers deliberately build structures to protect workers and the public, the gains will be captured by shareholders and senior executives, and the downsides will be concentrated on workers and communities.
Both camps are right about outcomes. Boomers are right about the long‑term potential. Doomers are right about the short‑term distribution and risk. Neither, on their own, resolves the board’s central problem.
The insight that changes the board’s job
This is more nuanced than either camp wants to admit. It is not zero-sum, with one side 100 percent right and the other 100 percent wrong. The part the AI jobs debate consistently misses, and the reason the transition to the agentic enterprise is a harder governance problem than anyone admits, sits in the space between them.
AI is not (yet) primarily taking jobs. It is taking "parts" of jobs.
Alexis Krivkovich, a senior partner at McKinsey & Company who leads the firm’s People and Organizational Performance practice, put it bluntly this year: very few jobs are being entirely automated away by current AI and robotics; instead, technology is slicing off tasks across many roles. McKinsey’s research finds that today’s technologies are technically capable of automating about 57 percent of work‑related activities in the U.S. economy - but that percentage is spread across “pieces and parts” of roles, not concentrated in whole positions. Another way to think about it: You can’t take half of person X, one quarter of person Y, and one quarter of person Z and make it one person.
Boris Cherny, the creator of Claude Code at Anthropic, gives this a concrete face in software. He told CNN that by the end of the year, “software engineer” as a role label will begin to fade, replaced by something closer to “builder” - someone who directs agents, reviews AI‑generated code, and designs systems at a higher level of abstraction while writing fewer lines of code personally. The job does not disappear. It transforms into something that does not yet have a standard title, compensation benchmark, or oversight framework.
For boards, this is not a more reassuring picture than classic automation. It is a more complex governance problem.
When AI eliminates whole roles, a board can count the headcount, audit the savings, verify that critical functions are still covered, and assign committee accountability for the transition. When AI takes 25 percent of dozens of roles simultaneously, none of those standard governance moves apply cleanly. The productivity gain is real, yet it's also diffuse, invisible, and hard to verify. The displacement is real, however it's also fractional, borne by people who are still on payroll and therefore invisible in layoff statistics. The accountability gap is real, and it's much harder to locate than a headcount reduction line on a CFO slide.
Andy Challenger’s observation cuts directly to the audit committee’s mandate: the money for those role‑fragments has been reallocated whether or not the AI is actually doing the work. The CFO has moved the budget. The question of whether the function is being executed - at what quality, under what controls, with what accountability if it fails - is exactly the question an audit committee exists to answer. Most audit committees are not yet asking it.
The Challenger data makes the stakes explicit for boards. AI has been cited for 49,135 job cuts in 2026 through April, making it the third‑leading cause of layoffs year‑to‑date and the top reason for cuts for two consecutive months. AI‑cited cuts rose from about 13 percent of total cuts through March to around 16 percent through April. But those figures capture only the declared rationale. They do not capture the thousands of roles that have been quietly reshaped rather than eliminated - where 25 percent of the work migrated to an AI system, the budget for that work was reallocated, and the human is now doing 75 percent of the job plus managing the AI, often without a change in title, pay, or oversight framework. None of that shows up in standard reports. It is inside the enterprise, invisible to the board, compounding into what you can think of as Governance Debt.
The week that made the pattern visible
Cloudflare’s letter and the Challenger report landed on the same morning, but they sit inside a wider pattern that matters for boards.
Cloudflare’s founders chose a level of transparency most CEOs have so far avoided. Their communication described internal AI usage increasing more than 600 percent over three months, with employees in engineering, finance, HR, marketing, and other functions running thousands of AI agent sessions per day and large portions of internal workflows now mediated by AI. That is the “fractional displacement at scale” model: AI did not eliminate Cloudflare’s marketing department. It changed what each marketer does, repeatedly, faster than a governance system built for annual reviews can track. The 1,100‑plus cuts are the visible residue of a transformation that was already underway.
Elsewhere, the governance questions look different but rhyme.
Coinbase has been restructuring toward what it calls “AI‑native pods,” where one person directs agents performing the work of multiple cross‑functional teams, while cutting layers of “pure managers” and flattening the org chart. That is not just an efficiency play; it is a redesign of the company’s delegation‑of‑authority architecture — the mechanism by which boards hold management accountable. When a management layer disappears, the accountability does not vanish. It migrates upward or goes missing.
In the same week, Coinbase reported a first‑quarter 2026 loss of roughly $1.49 per share against consensus expectations of a modest profit, and transaction revenue that missed estimates by tens of millions of dollars. Sam Altman has warned publicly that some firms will use “AI restructuring” narratives to reframe underlying execution problems as innovation stories. Aleksandar Tomic at Boston College makes the point plainly: “We’re restructuring for AI efficiency” lands very differently with investors than “We have a performance problem.”
Cloudflare and Coinbase are not alone. BILL, Upwork, and PayPal are among the firms announcing material workforce reductions in 2026 while citing AI‑driven efficiency and operating model shifts. Challenger data show that technology sector cuts totaled over 85,000 jobs year‑to‑date through April 2026, up about one‑third from the same period last year and the highest year‑to‑date total for the sector since 2023. The shared governance question is not whether each CEO made the right call. It is whether each board can verify the claims, track execution against the AI thesis, and name the committee that owns the answer.
Most cannot. An ISS review of 3,048 U.S. companies earlier this year found that only 8 percent disclosed any board‑level AI oversight, only 9 percent had formal AI policies, and only 16 percent had even one AI‑skilled director. Those numbers describe most boards evaluating AI‑rationalized restructuring today.
The jobs no one is counting
The optimists are right that new jobs are emerging. They are wrong to assume those jobs will govern themselves.
The World Economic Forum’s Future of Jobs 2025 analysis points to big data specialists, AI and machine learning engineers, fintech engineers, and information security analysts as some of the fastest‑growing roles. Challenger’s hiring and sector data already show some of this demand: aerospace and defense hiring is up sharply year‑over‑year, driven in part by AI infrastructure and autonomy investments, and automotive hiring is growing as manufacturers retool around AI‑enabled production. Goldman Sachs estimates that AI‑driven data center build‑out in the U.S. alone could create hundreds of thousands of net new jobs by 2030.
Those are the visible supply‑side jobs: the people who build and maintain AI systems. The invisible jobs are the governance‑side roles that enterprises now need and mostly lack.
Five years ago, your company did not need people who could authorize AI agent actions in production workflows. It now does. It did not need human oversight engineers to evaluate whether autonomous systems are operating within their authorized scope. It now does. It did not need AI disclosure officers capable of translating internal AI operating model changes into language sufficient for audit committee review and securities disclosure. It now does. It did not need governance architects who can map AI accountability to existing committee structures while avoiding new bureaucratic layers. It now does.
Boris Cherny’s “builder” is one example of an emerging hybrid role that sits directly in the board’s line of sight. A software engineer who now spends much of their time directing AI agents, reviewing AI‑generated code, and designing systems at a higher level of abstraction is operating in a different risk and accountability context than the engineer of five years ago. They need a new accountability framework, a new compensation benchmark, and a new oversight structure. Most compensation and audit committees do not yet have those in place.
McKinsey’s 2025 State of AI work found that only 13 percent of organizations have hired AI compliance specialists and just 6 percent have AI ethics specialists, even though a majority of firms report experiencing at least one negative consequence from AI use. Microsoft, in a survey of 20,000 workers across 10 countries, found that while employees are rapidly adopting AI tools, most companies have not yet updated metrics, incentives, or performance management to reflect how AI is changing work. That gap, between AI‑enabled operating models and AI‑aware governance staffing, is widening.
For boards, that gap is not theoretical. It belongs squarely to the compensation committee and the nominating/governance committee.
Who owns what: a committee‑level map
Most boards lack a clear answer to which committee owns an AI‑driven operating model restructuring - whether through whole‑job elimination or fractional displacement. The ambiguity is itself an accountability gap.
- The full board owns the strategic question: does the restructuring make economic sense, and does the capital allocation behind it match the board’s risk appetite? The board does not run the restructuring. It sets the standards and holds management to them.
- The audit committee owns Andy Challenger’s question: if budget has moved from human roles to AI systems, is there evidence that the functions previously funded by that budget are being performed at equivalent or better quality - and a clear line of accountability when they are not?
- The compensation committee owns the transition question at both ends. On the displacement end: who is bearing the cost and are reskilling efforts proportionate? On the transformation end: when “builders” absorb AI‑directed work on top of existing duties, are they being compensated for the expanded scope or doing more for the same pay while savings accrue to the P&L?
- The risk committee owns the failure architecture: what happens when an AI agent’s output propagates into a financial, HR, or legal decision before human review and is wrong? McKinsey reports that around a third of organizations have already seen AI‑related negative outcomes; the risk committee owns the AI‑specific incident response plan, distinct from cybersecurity.
- The nominating/governance committee owns the composition and charter question. If directors cannot distinguish a sound AI productivity claim from narrative, or evaluate whether the AI systems replacing management layers are operating within their authorized scope, there is a skills gap on the board.
Outside the U.S., this is becoming not just best practice but legal exposure. The EU AI Act’s phased enforcement, with full effect around August 2026 and penalties up to 7 percent of global revenue for serious violations, will make explicit AI governance structures a requirement for many companies operating in Europe. Boards that are restructuring around AI now will cross that regulatory threshold while still in transition.
Continuous governance, not episodic approval
Traditional governance is episodic. Management presents a plan, the board approves it, management executes, and the board reviews outcomes at the next checkpoint. That cadence assumes the operating model changes slowly.
The AI operating model does not.
Cloudflare reported a 600 percent internal AI usage increase in three months. Independent tracking from groups like Epoch AI shows inference prices for large language models halving roughly every couple of months and overall compute used in frontier training runs growing at multiples per year. The restructuring you approve this quarter will be powered by more capable systems six months from now. The risk surface, productivity potential, and displacement pattern will all move under a static governance framework.
Challenger’s monthly data show that AI‑cited cuts are rising as a share of total cuts while announced hiring plans fall sharply in the same period. The technology sector is simultaneously the largest source of AI‑cited layoffs and one of the weakest sources of new job announcements in early 2026. That asymmetry will not self‑correct. It requires continuous governance infrastructure: real‑time inventories of AI systems, exception reporting when agents operate outside scope, and recurring committee‑level review of AI‑driven budget reallocations.
Every quarter of deployment without that infrastructure is Governance Debt. Unlike financial leverage, this debt does not accrue linearly. It compounds. A board that approved major AI contracts three years ago and has not updated its committee charters, incident response plans, or disclosure protocols since owns that debt - whether or not it recognizes it.
The organizations that are governing AI effectively are not slowing down as a result. They are moving faster. They know where their AI systems are, what they are authorized to do, who can override them, and which committee will receive a report when something goes wrong. When a regulator, auditor, or investor asks what their AI systems were doing during a specific period, they can answer. That is not just a compliance posture. It is a competitive posture.
Six practical actions for leaders before the next meeting
- ACTION 1: THE 48-HOUR INVENTORY: Ask your General Counsel or Chief Risk Officer for a complete list of every AI system currently operating with autonomous or semi-autonomous action capability inside the company. A raw inventory delivered in 48 hours. Ask for both fully automated functions and partially automated ones - the fractional displacements are harder to see than whole-job eliminations and more likely to be missing. The completeness of the response is your governance assessment. The gaps in the list are your Governance Debt made visible. Most management teams will find this question harder to answer than expected. That difficulty is the finding.
- ACTION 2: THE BUDGET AUDIT: Ask your audit committee chair whether management can account for every dollar reallocated from eliminated or reduced human roles to AI systems in the current budget cycle. The mandate is Andy Challenger's sentence: the money moved regardless of whether the AI is performing the work. The question is whether there is documented evidence that the functions previously funded by that budget are being performed at equivalent quality, and who bears accountability when they are not. If management cannot provide that evidence within a reasonable timeframe, the board is overseeing a narrative, not a transformation.
- ACTION 3: THE TRANSITION EQUITY QUESTION: Ask your compensation committee chair two questions: who in the organization is bearing the cost of the AI restructuring, and are employees whose roles have been expanded - absorbing AI-directed work on top of existing responsibilities - being compensated proportionately? The fractional displacement model creates a specific and underexamined risk: productivity gains are captured in the budget reallocation while increased workload and expanded accountability land on employees whose compensation has not changed. Name it before the proxy season does.
- ACTION 4: THE DISCLOSURE REVIEW: Ask legal whether every public statement about AI restructuring, earnings call language, CEO letters, press releases, was reviewed by the audit committee for disclosure adequacy before publication. The test is materiality: if AI is now performing functions previously disclosed as human-operated, the change in operating architecture may require disclosure. Sam Altman's AI-washing warning is the relevant frame. The audit committee's review of restructuring disclosures is what separates a verifiable productivity claim from a narrative that lifts the stock price in the short term and creates liability in the long one.
- ACTION 5: THE INCIDENT RESPONSE GAP: Ask your risk committee chair for the company's AI-specific incident response plan. Not the cybersecurity plan. A distinct document addressing what happens when an autonomous or semi-autonomous agent session produces an incorrect output that propagates through a downstream financial, HR, legal, or operational decision before any human reviews it. Ask when it was last tested and who owns it. McKinsey's finding that 51 percent of organizations have already experienced at least one AI-related negative consequence means this is a current operating condition, not a theoretical risk. If the plan does not exist as a distinct document, the risk committee has found its gap.
- ACTION 6: THE COMPOSITION RECKONING: Ask yourself whether you have the governance context to evaluate management's AI productivity claims. Not AI education. Governance context: can you distinguish a sound AI productivity claim from a narrative construct? Can you evaluate whether the AI systems replacing a management layer are operating within their authorized scope? Can you assess whether the incident response architecture is adequate for the failure modes of the specific systems deployed? If the answer is no, the nom/gov committee has a composition question to address before August 2026, when EU AI Act enforcement arrives and the gap becomes a regulatory exposure, not just a competency observation.
The Next Step
The auditor in the room always asks the question the optimist does not want asked and the pessimist cannot make actionable. That is the board director's role in an AI restructuring. Not to arbitrate between Andreessen and Acemoglu. Not to decide whether the long arc of technology history vindicates the restructuring. But to ask, on behalf of shareholders, employees, and the institution: can you prove it?
Can you prove the functions are being performed? Can you prove the budget reallocation is producing the outcomes it was supposed to produce? Can you prove that the accountability structure eliminated in the name of AI-native efficiency has been replaced by something equally accountable, not just something faster?
The board that cannot answer those questions, or cannot name the committee responsible for answering them, is not being optimistic about technology. It is taking a fiduciary vacancy. As the saying goes, "follow the money" - and in this case, your job is to know where the money went, whether the work is still being done, and who answers when it is not.
Governance Is Alpha.
