For decades, the safest way to hire a senior executive was to find someone who had already done a larger version of the job. A larger P&L, more employees and longer industry tenure offer legible evidence of leadership at scale. But relying on those credentials assumes that the operating conditions under which the experience was accumulated remain sufficiently similar to those the executive is being hired to navigate. As AI changes the economics of some forms of knowledge work, the hiring organisation needs to test that assumption against the actual mandate.
The failure does not look like failure
Eric Vishria, a general partner at Benchmark, described watching executives from the previous software generation struggle inside scaling AI companies. His advice was not to work harder or learn faster. It was to discard the inherited template and rebuild from the customer and the technology upward.
The obvious objection is that none of this is new. Large-company executives have always struggled inside fast-growing companies. Speed breaks people. Stage mismatch breaks people. Neither requires a technology explanation. Those explanations need to be distinguished from a failure of operating-model design. A speed or stage mismatch often appears through missed timelines, unstable execution or delayed decisions. A design error can be harder to identify because the executive may execute competently while recreating a structure whose economic rationale has changed.
Vishria illustrates the distinction with the public software companies that grew after the 2021 valuation peak, reached or approached profitability and still lost equity value as revenue multiples compressed. His formulation is deliberately severe:
"Every single day that you are hitting your plan, you are destroying equity value."
The quotation is not a general law of management. It names a recognisable failure mode: competent execution towards a design whose economic rationale is dissolving underneath it. These are people who have scaled organisations from hundreds to thousands before, doing it correctly again, from a starting point that has moved.
Good execution metrics can therefore conceal a weak design premise. The executive who owns the mandate, the talent leader and the search partner need to examine that premise before the appointment, rather than expecting a later performance review to expose it.
Experience creates two things
Experience develops judgement, but it also produces assumptions about how organisations should work. Judgement can remain valuable across environments even when some of the assumptions need to be reconsidered. A leader who has spent twenty years running large organisations has learned enormously valuable things about customers, incentives, capital allocation, execution and organisational behaviour.
They have also formed a view of what a product organisation should look like, how many people a function needs, which work requires specialists and how software gets built. Those views extend to where decisions sit, how many managerial layers are necessary and which processes need standardisation.
Those conclusions were often rational responses to the constraints under which the organisation evolved. If a constraint changes, the organisation should test which of those conclusions still applies and which now limits its options.
Why this is not the last cycle again
Hiring leaders have heard versions of this argument before. In the late 1990s, the advice was to read resumes forward and weight digital fluency over operating depth. Many of those bets failed. The scepticism is earned, and the present claim has to be specific to survive it.
The internet changed distribution. Cloud changed infrastructure. Mobile changed access. Each also changed knowledge work, but none removed its familiar staffing relationship: more parallel thinking generally required more people. AI can loosen that relationship in particular tasks. Information that required a team to gather can sometimes be synthesised by one person with a model. Software can be generated before an engineer reviews and integrates it. Specialist handoffs can compress into an AI-assisted workflow.
The important word is can. Microsoft's chief executive said in April 2025 that AI produced 20 to 30 per cent of code in some company repositories, with wide variation by language. DORA's research found benefits in individual productivity and code-review speed alongside weaker delivery stability and throughput. GitClear found more duplication and short-term churn in the codebases it studied.
The evidence does not show judgement scaling without limit. It shows that generating work and validating work no longer move together. Some capacity migrates from production to review, integration and accountability. That is enough to disturb an org chart. A handoff built because no one person could hold the customer problem and the implementation may become expensive when one AI-assisted person can hold more of both. A managerial layer built to aggregate information may need a different purpose when aggregation is cheap.
The appointment can go wrong when exceptional experience operating an existing structure is treated as sufficient evidence of the ability to redesign it. The relevant assessment has to include whether the candidate understands the constraints that produced the structure and can determine which are changing.
That does not make institutional depth worthless. It means the credential needs its constraint provenance examined: under which conditions was this earned, and do those conditions still hold?
Direction is not magnitude
If AI changes only a trivial share of the rationale for current organisational design, none of this warrants a different appointment. The available evidence is uneven. Software-development figures are measures of generated code, not eliminated roles. DORA's findings suggest that faster production can increase downstream work. The bottleneck can move rather than disappear.
Sales offers a more provocative signal, although the public evidence remains observational. Vishria reports individual representatives at AI-native companies carrying many times the revenue capacity associated with conventional enterprise-sales planning. If those cases prove repeatable beyond exceptional companies and unusually favourable markets, they do not merely improve a sales organisation. They alter the headcount formula used to size one.
Finance, legal, procurement and operations may move differently. Some activities may prove far more resistant than current enthusiasm suggests. A few functions do not settle an organisation. But a senior appointment is a multi-year commitment. The relevant question is not whether every function has already changed. It is whether the mandate requires someone to determine which constraints are moving during the term.
The credential may describe the old constraint
Consider one of the most legible executive credentials:
Managed 3,000 people.
It tells a hiring committee something important about leadership at scale. It can also hide an assumption: that the future organisation should continue to require something resembling those 3,000 people. If the mandate is to operate that organisation more effectively, the credential is highly relevant.
If the mandate is to determine what that organisation should become as AI changes the economics of knowledge work, the inference is less obvious. The more relevant question becomes:
Can this person distinguish what must be preserved from what no longer needs to exist?
Vishria describes the assets that matter as understanding customer problems, taste and fluency with the jagged edge of AI capability: where models excel, where they fail and how fast that boundary moves. A title does not reliably certify those things.
At senior levels, the candidate needs to connect an understanding of how the business works with a clear judgement of what good looks like and a practical understanding of what technology makes possible. Where the technology changes assumptions within the strategy itself, that understanding needs to sit with the executive accountable for the strategy, alongside the expertise of the CIO or CTO.
Where the credential gets institutionalised
It would be convenient if this were only a board problem. It is not. A board directly owns a small number of appointments. The operating committee, function heads, the layer beneath them and the succession slates feeding all three also run through executives hiring their own reports, a talent organisation that owns the process and search partners that build the market map.
The selection process can reproduce the existing model at each of those levels, even without an explicit decision to preserve it.
Search methodology reproduces the market's visible shape. Retained search commonly maps people in equivalent roles across competitors and adjacent organisations. Done well, it is rigorous. But a map of existing profiles does not automatically identify someone for a role whose design is meant to depart from the existing market. Ask for a leader who has redesigned a function around AI and the easiest proxy remains someone who has run that function at scale. The shortlist can be accurate and still answer yesterday's question.
Search economics do not correct the bias. Public filings from Korn Ferry describe executive-search fee revenue as generally one-third of estimated first-year cash compensation. That does not make the search partner's advice suspect. It does mean the fee model contains no natural incentive to question whether a role should carry less organisational scope than its predecessor.
Job architecture can price the constraint into the organisation. Job-evaluation systems connect accountability, problem-solving, scope and reporting relationships to levels and pay structures. Headcount is not always the primary input, and practices vary. Where people managed remains a strong proxy for role size, however, the architecture may discount a leader who creates more enterprise value with a materially smaller structure.
The error compounds downward. A function head appointed on scale credentials can reasonably apply the same criteria to direct reports. Within a few appointments, an assumption can become a leadership model without anyone explicitly approving it.
The correction therefore cannot sit only with a nomination committee. It belongs with the executive who owns the mandate, the talent leader who writes the scorecard and the search partner who decides which careers count as evidence.
The opposite error is also expensive
The assessment also needs to guard against the opposite error: removing a structure without understanding the function it performs or the risk it controls. The ability to identify what should change is valuable only alongside the ability to recognise what still needs to be preserved.
Geoffrey Hinton's 2016 prediction that organisations should stop training radiologists remains the cleanest illustration. Imaging models continued to improve, but the profession did not disappear. NBER researchers have documented how reimbursement, liability, workflow, demand and institutional structure complicate the translation from task performance to employment.
Klarna is the operational version. In February 2024, the company reported that its AI assistant had handled 2.3 million conversations in a month, performed work equivalent to about 700 full-time agents and reduced average resolution time from eleven minutes to under two. It projected US$40 million in annual profit improvement.
The subsequent record resists a simple reversal story. Klarna's chief executive later acknowledged that excessive emphasis on cost had reduced service quality and that the company was adding human capacity. Its 2025 annual report also says the assistant handled 80 per cent of customer-service chats with no decline in consumer satisfaction, while company filings report substantial cost savings.
The technology worked at scale. The organisation still needed to decide where a person created value that the aggregate service metrics did not capture. Gartner predicts that by 2027 half of companies attributing customer-service headcount reductions to AI will rehire people into similar functions, often under different titles. It is a forecast, not an observed result, but it identifies the same risk: task automation is not proof that the surrounding role has no residual purpose.
The two errors present differently. Preserving too much can look like prudence and surface slowly, as drift. Cutting too much can look like decisiveness and surface suddenly, as breakage. The useful selection criterion is the ability to establish why a structure exists before deciding whether it still needs to. Testing that requires candidates to explain the operating purpose, dependencies and consequences of a proposed change. Enthusiasm for AI or caution about it is an inadequate proxy.
This changes the value of breadth
Senior selection has generally rewarded depth. Long tenure inside an industry or institution can provide pattern recognition, credibility and an understanding of consequences that outsiders miss. During discontinuity, breadth offers a different signal. Someone who has repeatedly moved across industries, companies, functions or business models has had to learn new systems rather than inherit one institution's answers.
The objection is obvious: frequent movers can carry one template and apply it with more confidence each time. The distinction is what the template is for. A template held as a conclusion is portable baggage. It arrives finished and looks for somewhere to land.
A template held as a hypothesis is a starting position revised by every organisation it meets. The person has been wrong in several environments and has had to notice. Career shape therefore supplies a question, not a verdict. Breadth can indicate exposure to different constraint sets; tenure can indicate deep knowledge of one. Neither proves whether the candidate can separate an operating principle from the environment that made it useful.
The emerging advantage may belong neither to the traditional operator nor to the pure technologist. It may belong to a third profile:
the leader with enough institutional experience to understand why complex organisations work the way they do, enough AI fluency to see which constraints are changing, and enough independence from the existing model to redesign it rather than merely optimise it.
The mandate has to be honest
When choosing between the person who has operated the largest version of yesterday's organisation and the person better equipped to design tomorrow's, which experience should carry the premium? It depends on the mandate. Prior operating scale may carry more weight in a preservation mandate, while a redesign mandate requires stronger evidence that the candidate can examine and change the underlying model.
That conditional helps only when the organisation is honest about which mandate it is running. Transformation language is common in executive specifications, even where the practical requirement is continuity with controlled improvement. The word can become positioning rather than a decision.
If the mandate is preservation - a stable business, a regulated environment, a structure that works - scale experience may be the right answer and the specification should say so plainly. If it is redesign, the credential describing the old constraint should not decide the appointment by itself.
The failure is writing redesign into the brief, hiring for preservation and discovering the mismatch three years later, after the executive has competently delivered what their experience taught them to build. The candidate's record should therefore be assessed against the organisation the role is expected to create, including which prior lessons still apply and which require challenge. That makes the mandate, rather than the most legible credential, the basis of the appointment.
The decision this should change
Whoever owns the mandate - the board for the chief executive, the chief executive for the operating committee, the function head for the layer beneath - states whether the role is principally preservation or redesign before the specification is written. That choice determines which credentials carry weight.
The talent function changes the scorecard before the shortlist arrives. Ask the candidate why a structure in a previous organisation existed, what it protected against and what would have broken had they removed it. The answer helps distinguish someone who inherited an organisation's assumptions from someone who examined them. Separately, review whether people managed still belongs as a primary scope input wherever AI-assisted work changes the relationship between headcount and value.
The search partner is asked to include profiles the conventional market map might miss: candidates whose prior headcount was smaller but whose remit was genuine redesign. A shortlist of larger equivalents is not automatically responsive to that mandate.
Hiring executives apply the same test one level down, before the assumption embeds itself across the leadership population.