Your head of engineering was hired to run eight people, and now holds responsibility for 25, two of them more senior than anyone on the team when they joined. Nobody has formally changed a thing about the role.
This type of thing is almost never described as a leadership problem. It is told as a hiring problem, a retention problem, or a founder saying the team feels slower than it should for its size.
Companies in this market review their individual contributor bar constantly: interview processes get rewritten and more difficult, levelling gets recalibrated, every strong hire moves the reference point for the next person coming in. The layer above those people gets looked at when something forces it, which usually means a raise, a board conversation, or a resignation that lands badly. That difference in review frequency is where the problem grows, and in AI companies it grows faster than it has for any previous generation.
The company changes shape incredibly quickly
An AI company now reaches the revenue milestones that used to take three years in half the time, which means the job every leader was hired for is redefined inside eighteen months.
Stripe’s June 2025 analysis of its 2024 payment data compared the top 100 AI companies on its platform against the top 100 SaaS companies of 2018. The AI cohort reached $1 million in annualised revenue in a median of 11.5 months against roughly 15, and $5 million in 24 months against 37.
Bessemer’s study of 20 high-growth AI startups splits them into two archetypes, and the second one matters more than the headline-grabbing first.
The Supernovas are the companies everyone reads about, $40 million ARR in their first year of commercialisation and around $125 million in their second. Bessemer is candid that this comes with roughly 25% gross margins and fragile retention.
The Shooting Stars are the more useful benchmark, and Bessemer says so explicitly: this era will be defined by hundreds of them rather than a handful of outliers. They go from around $3 million in year one to around $12 million, then $40 million, then $103 million.
Taking the Shooting Star path, a head of engineering hired to support a $3 million business is running a $40 million business two years later. Nobody in that company changed jobs, but the job changed underneath them.
The headcount signs that used to warn you has gone
In the SaaS era you knew a new layer was needed because you crossed 50 people, then 80, then 150. Headcount has always been the indicator of complexity.
Bessemer’s Supernovas average $1.13 million of ARR per employee in year one, four to five times a typical SaaS benchmark, and even the Shooting Stars sit at around $164,000.
The trend holds well beyond the outliers. Bessemer surveyed around 175 functional leaders across more than 100 portfolio companies for its talent report and 49% said they were already delivering more without adding headcount.
Which means the old trigger never fires. You can triple revenue, double customer complexity and add fifteen people. By the measure founders have always used, nothing happened.
The management layer is thinning while IC scope expands
The clearest evidence arrived in June 2026, in SignalFire’s State of Talent report, built on data covering 650 million people and 80 million organisations.
Spans of control have widened everywhere, and fastest at exactly your stage. Each engineering manager at a major tech company now supervises around 12 engineers, up 14% on 2019. At early-stage startups it is around 15, a 34% increase. Product managers support 22% more engineers than they did. SignalFire is explicit that the thinning is not confined to engineering: the senior management layer has shrunk in nearly every core function.
At the same time the ceiling on individual contributors has lifted. SignalFire names the result the Super IC, someone operating at a scope that used to belong to managers and directors. Senior IC and staff roles are growing as a share of hiring while engineering manager roles are flat or declining, and top-of-band staff and principal packages now rival or beat director pay, reversing a management premium that held for two decades.
So the two cycles are not just running at different speeds. They are moving in opposite directions. Your best ICs are absorbing scope that leadership used to hold, while the leadership layer itself gets thinner and each remaining leader carries more. A head of engineering who felt appropriately senior in 2023 can be under-levelled by 2026 without a single thing changing on paper.
SignalFire’s own recommendation to founders is to stop planning on headcount and start measuring engineers per engineering manager, engineers per product manager, and builders per non-builder. That is a better early warning system than any headcount threshold, and almost nobody we work with tracks it.
What happens to early leaders
Funding rounds are where people leave. Pin’s July 2026 analysis of four million career moves at venture-backed companies found that 54% of the people who joined at Series A had gone by Series B, 50% of Series B joiners had gone by Series C, and 59% of all hires were out inside 24 months. Leadership is not exempt from that, and a leader who leaves at the round takes the context of everything built before it.
Felicis mapped the executive benches of 30 outlier companies at every funding round, including Stripe, Databricks, Snowflake, Atlassian, Shopify and Anduril. It was published in 2022 and covers the SaaS and marketplace generation, so it is well outside the window we would normally use.
We are still citing it as nothing published since tracks which functions break first across a company’s whole funding history. Everything newer measures turnover in aggregate, and the mechanism it describes is stage mismatch rather than performance, which is not a thing that dates. Read the direction as sound, and assume the timeline is tighter now than it was for the companies in it, given everything above about how much faster this generation scales.
Commercial leadership is where stage mismatch bites first and hardest: none of the Series B sales leaders were still head of the function by IPO or Series E. Technology has the most road, and Felicis’s read is that external technical advisors or engineering operations can extend an early CTO’s runway considerably, in a way that does not work in revenue-generating functions.
Three questions worth separating before you hire
When a founder tells us a leader is struggling, they are usually describing one of three different situations that happen to feel identical from the outside. The fix for each is different, and picking the wrong one is expensive.
Is it capacity? The person is doing the right job well and there is more of it than one person can hold. The most common and most fixable. Usually resolves with a layer underneath, not a replacement above.
Is it scope? The role has quietly absorbed things that were never in it. A head of engineering who now owns security, data, infrastructure and vendor relationships is not underperforming. They are doing four jobs, probably one of them badly, and usually the one they care least about. Splitting the remit often beats changing the person.
Is it capability at this stage? The genuine version, where the company needs judgement the person has not had to exercise yet. Less frequent than the other two, and worth naming honestly when it is the answer, because the other two fixes will not touch it.
The order matters. Companies that jump to question three hire over the top of people who were solving a capacity problem, and lose them within a year.
Hire for twelve months out, not for the exit
A mistake we see most often is hiring for a company two stages ahead of the one you have.
It usually comes from the board, and the logic sounds reasonable. If you will eventually need someone who has run a 200-person organisation, why not hire them now and skip a transition?
Because the job in front of them is not that job. At seed to Series B you need someone who will personally run 25 people and personally hire the next 25, write the operating cadence from scratch, and hold a credible technical conversation with a staff engineer. Someone whose last three years were spent managing through directors often finds that work unrecognisable, and leaves before the company grows into the role they were hired for.
The stage distinction is real, and worth being precise about. Bessemer’s advice for growth-stage companies points the other way: once you are scaling hard, hiring a commercial leader who has carried a revenue remit in the hundreds of millions becomes important. Both can be true. The error is applying the growth-stage answer at Series A.
Hire for the next twelve to eighteen months, with enough headroom that the person is not at their ceiling on day one. You will have this conversation again, and planning for that is cheaper than avoiding it.
Promote, reshape, or add above
Once you know which of the three problems you have, there are three ways out, and most founders only seriously consider one of them.
Promotion is undervalued, particularly on the technical side, with the caveat the Felicis numbers attach: plan for what happens to that person at the next stage rather than treating the promotion as the end of the conversation.
Reshaping is the route founders forget. Writing in 2024, Deel’s CEO Alex Bouaziz told Fortune he was advised to replace his whole leadership team as the company scaled and ignored it, keeping 80% of the people who were there at $0 ARR. His approach was to narrow senior people back to what they were genuinely excellent at and hire around them. His head of growth had been running all of marketing; refocusing her on top-of-funnel growth and hiring for the rest worked better than replacing her and losing what she knew.
Adding above is the route founders raise last and the one that often fits best. It keeps institutional knowledge, demotes nobody, and brings in experience the company will need soon. It also needs the most care, because the people underneath need to understand why before they hear it from someone else.
In AI, your leaders are a retention lever
There is a reason to move on this that does not apply in most other markets.
Your strongest ICs can leave. Not vaguely. The people you most want to keep could have a conversation with a frontier lab this month if they chose to.
What keeps them is rarely only the package. SignalFire’s 2025 State of Talent report found Anthropic leading the field on two-year retention at 80%, against 67% at OpenAI and 64% at Meta, and credits the draw to intellectual discourse, researcher autonomy and clear routes for career growth. All three are produced by the leadership layer. They are what a good manager of technical people creates and what a mismatched one quietly dismantles.
So a leadership hire in an AI company is a retention decision as much as an execution one. Candidates at this level ask who they would report to, and they check. The answer travels: these communities are small, people have worked together before, and a reputation for strong technical leadership does more sourcing work than most employer brand spend.
That is also the argument for moving early. In our experience these searches run three to six months from brief to start date once notice periods are counted, and the best candidates are not looking. By the time the gap is obvious enough that your team has noticed it, you are hiring under pressure, against a visible problem, with the people you are trying to retain watching how you handle it.
Where to start
A straightforward review, once or twice a year, well away from any crisis:
- For each leader, which of the three questions applies? Capacity, scope or stage.
- Which function will constrain the next eighteen months most? That is where the next hire goes, regardless of which gap feels loudest today.
- Who could hold more if something were taken off them?
- If a key leader resigned tomorrow, how long would the search take? If that number worries you, start a conversation now rather than a search later.
None of this needs a reorganisation. It needs an hour, honestly spent, before the market forces it.
These are the conversations we have most days with founders and talent leaders across AI companies, usually somewhere between “we think we’re fine” and “we needed this person three months ago”. If you are somewhere on that line, we are easy to find.
Related reading: Forward Deployed Engineers. What is the role, why is demand spiking, and who can perform in it?