More Is Less

Kevin Mannix

Oct 27, 2026

Share

I had a different post in mind for this week, but as I sat down to write it some pretty shocking news came across my desk and I decided to dash off some quick thoughts as a followup.

Microsoft has offered 60,000 employees, or about 27% of its global workforce, voluntary buyouts as the software giant spends heavily on artificial intelligence.

The company’s voluntary exit offer was available to eligible employees in Canada, India, the United Kingdom and the United States, the company’s Chief People Officer Amy Coleman wrote in a blog post Monday.

What truly bonkers news. A massive corporation with a market cap of ~$4 trillion, the third largest company in the S&P 500, revenue growth in the high teens over the last several quarters, marking a fourth of its workforce as expendable in an era of massive potential growth.

Now, to be precise, a voluntary buyout program isn’t technically marking a specific individual as expendable - that’s a bit too far. Microsoft knows that the participation rate will not be 100% and they undoubtedly had some HR actuaries in a Redmond backroom zeroing this in to hit a specific shedding target - something closer to 20% or 30% participation seems more likely. The message, however, is clear: we don’t need as many people even as we grow. Microsoft refutes this as the intention, but I don’t see how you can expect any message other than that to land. Even if participation is low, this seems like softening the ground for a later advance if and when needed.

Software’s economic magic has always been the operating leverage it has from being capital-light & effectively zero marginal cost in the long run. AI has supercharged the leverage per employee to an unthinkable extent. The upper bound per employee has evaporated, and a lot more can be done with a smaller set of people. Even with its fantastic operational leverage, there was still a relationship between business growth and headcount growth; a sense that in order to grow the balance sheet and valuation, one must grow headcount. This may not only be broken, but may actually be reversed! More headcount may be detrimental.

AI is the obvious catalyst here, likely both in the carrot of extreme increases in productivity per person and the stick of large CapEx investments that impact cash flow, etc. It has dramatically expanded the range of business functions that can be performed without a proportional expansion of the organization. One of my favorite Civilization quotes (via Oscar Wilde) no longer applies: “The bureaucracy is expanding to meet the needs of the expanding bureaucracy.”

While Microsoft is among the constellation of special companies that have dominated the market for the last decade and a half, this reworking of a fundamental relationship is not isolated to just them. I expect this correlation between headcount and revenue potential to be generally applicable to most companies. Anthropic released its S-1 the same day (exciting times) revealing a revenue per employee of $6.6m, a pretty striking example of what’s possible in this new paradigm. These are extreme examples that come from gigantic, world-class companies, but they hint at what might trickle down to startups of all shapes and sizes, existing and future. Foundation model companies keep overturning fundamental assumptions across multiple fields; they have also ushered in tools that challenge the long-standing assumption that organizational expansion is a necessary consequence of business growth.

This should send shivers down the spines of most people that work at a pre-AI company. I’m mostly focused on startups here - my guess is that most large corporations (Mag 7 notwithstanding) have enough business inertia and internal hang ups that individuals will largely be inoculated from fundamental AI-related changes for a while. The downstream effects at both a company design level and an individual level could both be pretty heavy and rework some of the core truths we take for granted.

Existing (legacy) startups face an existential threat: your non-AI-native company structure will slow down your pace and leave you vulnerable. This may be avoidable if you’re seed / Series A as your PMF is weak enough, distribution limited enough, & team small enough to muddle through a complete reset of the company to look AI-native from a far enough distance. Any size beyond that suffers from the same disease that incumbents face, that competitors are both better equipped from a product standpoint & that your cost structure is inherently weaker than AI native companies.

The growth cycle for startups has typically been something like 1) traction/growth leads to funding 2) which unlocks more people 3) which enables more growth 4) repeat. Organizations have to naturally grow headcount to handle growing operational complexity: deeper and wider product lines, customer support, compliance, billing, enterprise sales, internationalization, and so much more. Headcount growth transforms a founding team’s loose undergrad-group-project management dynamics to hierarchical structures and reporting lines. There are teams and divisions and orgs all with their own responsibilities. What starts off as a small team of builders crystallizes to a certain shape, and the organizational inertia is a meaningful force; it’s not easy or quick to break apart and reassemble.

However, that growth path may not be necessary anymore. A common software quip shared with first-time EMs goes something like: “software is easy, people are hard.” AI has enabled an alternative where companies can skip the “hard” part and maintain an extremely lean team without sacrificing growth. This is phenomenal for an individual or a small founding team just starting out, who can delay incurring the overhead of additional hires for far longer than before.

It has the same potential power for legacy startups; however, because all of this is a quick reversal from previous truths, it requires a much more painful path.

In some cases, legacy startups may continue to hold onto the previous era: The large team that was filled out in order to grow might now instead be an anchor. The code quality that engineering teams have meticulously been maintaining means that the team doesn’t want to adapt AI tools for fear of inferior code output. The potential energy from these new tools isn’t captured, and things move much more slowly relative to peers who embraced them.

In others, tension might arise from embracing new tools without updating the existing structure: Product managers don’t need to rely on engineers to get an MVP out the door. Marketing can generate assets and full campaigns in-house. A single engineer can effectively build across team boundaries without depending on others shipping. When this goes against the grain of established hierarchy, chaos follows and personnel disagreements flare.

Fundamental structure and culture will need to adapt to the AI era, which is a painful transition to go through. Coarsely-grained, this probably means at best just a re-org but more likely a reduction in force. Fine-grained, it might be reevaluating what quality work is and resetting incentives to maximize the advantages of AI & agents without shirking quality or accountability of their outcomes. It’s not clear yet what makes sense here, as there are limited examples of durable success within this new era. It will be hard to get right and even harder to keep the faith as new flavors of problems emerge.

There is some serious inertia around large orgs: organizations can subtly shift to prioritize their own self-preservation over the stated goal, especially when faced with an existential threat. There will be large pushback against any consolidation or restructure, and it has a serious risk of washing out cultural goodwill. It took months to finally fill that precise mid-level role that was sorely needed. A long-time, loyal IC has finally been given the chance to have a handful of direct reports as a career growth opportunity. We just rewrote the company core values after our last funding round. Do we really want to burn it all down?

This appears to be a uniquely bad job market (though, perhaps like Tolstoy’s unhappy families, all bad job markets are bad in their own way?). The strange bit about it, not unlike most of the period since COVID and 2020, is that it’s mostly just a sense of foreboding even when things are good. The pall has been from the horror stories of well-experienced individuals in typically high-demand industries unable to find a job for months, or from new grads taking service jobs to stay afloat instead of leveraging their degree. It’s Final Destination: Career, knowing that even though somehow death has been narrowly missed so far, fate has not been cancelled, only delayed.

The battle around AI impact has mostly been fought in ink and conference keynotes; lots of wild-sounding assertions from bleeding-edge CEOs without really clear data to back that up. It goes without saying that the technology has felt transformational, and for those of us not in the research weeds it’s felt completely sudden. While we’re bad at grasping exponential growth, the base case is that at some point we’d see the creative destruction. That the pie might be getting smaller this time, and that we may not even get a slice. Maybe that time is finally here.

If legacy startups, tech companies, etc. - pretty much anyone not AI-native - align their org structure and headcount to fit with the leverage of AI tools, traditional career paths will be broken for individuals as well.

Startups have always offered an opportunity for enterprising young folk, career switchers, and high-agency ICs to jump in without requiring all of the prerequisite experience or time-in-seat, and wield enough power and influence to be dangerous. Trial by fire is a great teacher, and has helped forge a new generation of workers, many of whom have gone on to repeat the cycle by starting and growing companies of their own.

In a world where less headcount is needed to achieve great heights, those opportunities will be more scarce, and expectations raised. That’s better for individual companies, but threatens to cut off oxygen to a valuable on-ramp for individuals who might not fit in the traditional candidate box at a more established corporation, and prevents the next class of well-seasoned employees from forming.

The paradox is that the industry is demanding high-agency, experienced workers while eliminating the ecosystem that produced them.

For those already within the system, smaller teams mean less career advancement in the traditional sense: fewer promotions to give, people to manage, feudal estates to acquire. Cutting the other way, organizations might be flattened and more decentralized as the reach of an individual can double or triple what it is today. More responsibility and scope per person is likely - the traditional requirements of career advancement - but not manifested in the traditional form of a people leader.

The skill sets needed to perform may coalesce on just a single attribute: the ability to get things done. For engineers, this means thinking about problems from a holistic sense, starting from the business value and working backwards to the technical bits.

For PMs, this means wrangling with code and architecture and design rather than delegating that to others, succeeding through action rather than coordination. EMs will need to get back in the saddle and be able to code - it’s unclear how well the management approach from the last decade will translate to orchestrating machines. Each individual has been given their own team in the form of AI agents, and all will need to understand how to effectively leverage that team to succeed. Agency has always been important for startups, but now it may be the only thing.

Who knows where Microsoft’s Voluntary Buyout Program ends up in terms of partici