Where does the knowledge go when 8,000 people walk out the door?

Layoffs are often talked about in terms of cost savings. They should be seen as capability destruction. When 8,000 people leave a company, the salaries come off the P&L, but the irreplaceable knowledge in workers heads, the workflows nobody documented, and the judgement built across thousands of small decisions, walks out the door uncounted. Companies cutting before capturing that knowledge are torching the very asset they believe AI is replacing.

On 20 May, roughly 8,000 Meta employees will be laid off as the company redirects more than $115 billion toward AI infrastructure. Another 6,000 open roles will be cancelled. The internal memo described it as an efficiency push.

Efficiency is one way to describe it. Capability destruction is another.

Sit with the maths for a moment. Eight thousand people. If each has spent five years inside Meta, that’s forty thousand person-years of institutional knowledge: the unwritten reasons certain decisions went the way they did, the soft handoffs between teams, the background explaining why a workaround exists. 

None of that appears on the cost-savings line. None of it is recoverable from a Confluence page. And none of it can be relearned by the model that is, in theory, replacing it.

This is the accounting fiction at the heart of the current layoff cycle. We are calling it cost reduction. The honest entry would be capability destruction.

The 80% that nobody is counting

Every business runs on two kinds of knowledge. The first is explicit. It lives in policies, contracts, dashboards, code, training materials. It’s documented, searchable and, increasingly, machine-readable. 

The second is tacit. It lives in the heads of the people who do the work: the patterns they’ve learned, the judgement they apply, the shortcuts and exceptions and unwritten norms that make the explicit stuff actually function.

McKinsey's research suggests about 42% of essential expertise inside large organisations lives only in employees' heads, never documented anywhere. Other estimates put the tacit share closer to 80%. Whichever figure you prefer, the conclusion is the same. 

The institutional knowledge most material to how your business works is the knowledge least likely to be captured anywhere when the person holding it leaves.

That’s fine when attrition is gradual. It’s catastrophic when the tech industry sheds 80,000 jobs in a single quarter, with nearly half of those positions justified as AI-driven efficiency.

What’s the actual asset being destroyed?

The asset being destroyed is the organisational substrate AI needs to run on. 

AI systems don’t invent business judgement. They learn it from your data, your processes, your decisions, your edge cases, your historical exceptions. 

If you cut the people who hold those patterns before you’ve captured them, you don’t have a leaner company that AI can now optimise. You have a company with a thinner foundation and a more expensive AI bill.

Said differently: a layoff that precedes knowledge capture is not a transition to AI. It’s a downgrade to a worse version of what you had, with higher capex.

This is why the same companies announcing layoffs are simultaneously announcing struggling AI rollouts. 

Gartner research published in February 2026 found that only 1 in 50 AI investments delivers transformational value, and only 1 in 5 delivers any measurable return — while global AI spending is forecast to reach $2.52 trillion in 2026. 

The systems are being asked to perform on an evidentiary base that has been hollowed out. You cannot orchestrate decisions you no longer remember how to make.

Why does conventional knowledge management fail here?

Because it was designed for a world where knowledge transfer was a polite courtesy, not the precondition for a multi-billion-dollar capital strategy.

Wikis, runbooks, exit interviews and offboarding checklists were built to preserve continuity for the next human hire, not to feed a system of agentic apps that needs structured, schema-aware, decision-grade input.

The result is that most "knowledge bases" are graveyards. They store the documents nobody reads in a format nothing can use. When the people leave, the explicit residue stays. The tacit core, the part that mattered, evaporates.

Capability accounting starts before the layoff, not after

Companies that take this seriously do three things differently. 

They treat tacit knowledge as a balance sheet item, not a HR concern. 

They capture it in a structure their AI systems can actually consume, which means schema and ontology, not free text. 

And they do this before workforce decisions are made, so the asset is preserved regardless of who stays.This is the work Sugarwork was built for, and the reason we joined Decidr in late 2025. 

Sugarwork captures the invisible 80%: the undocumented workflows, the tacit expertise, the decisions and exceptions that never made it into a system of record. 

The companies that come out ahead won't be the ones that cut fastest. They'll be the ones that knew what they had before they let it go.