The 80/20 Token Economy: PE Firms Are Already Targeting 80% AI, 20% Human

A PE investor revealed on Moonshots EP #242 this week that he has already implemented a target across all his portfolio companies: 80% of operating costs in AI tokens, 20% in human salary.

Not as a prediction. As an implementation.

Dave Blundin, GP of Link Ventures: "We already implemented targets across all of our companies on this. And we're targeting 80% token 20% salary."

Then the sentence that reframes everything: "An employee that can improve the efficiency of our AI, even 1%, is worth a lot more than cost cutting."

This is not a workforce reduction story. It is a workforce inversion story. The 20% who remain become the most valuable people in the company. Not because they do the most work — AI does 80%. Because they direct the work. They are the articulation layer. They are what makes the 80% effective.

The Stepping Stones Are Already Mapped

Blundin did not describe this as a single leap. He described a progression that his portfolio companies are already walking:

Current state → 50/50 (immediate target) → two-thirds token, one-third salary (Jensen Huang's benchmark: a $500K engineer who does not spend at least $250K on AI tokens should alarm you) → 80/20 (Blundin's target) → endpoint: "token costs much, much bigger than payroll."

Each step has a different organisational implication. At 50/50, you are augmenting existing roles. At two-thirds/one-third, you are restructuring teams. At 80/20, you are redesigning the entire operating model around a fundamentally different question: not "what work do our people do?" but "how effectively do our people direct AI?"

Dr. Alexander Wissner-Gross made the measurement dimension explicit: "For the first time, we have legible, defensible, analyzable inputs for employee productivity. And that is a sea change." Tokens are measurable. Hours are not — not honestly, not at the level of productive output. For the first time, a PE firm can look at per-employee token spend and know, with data, who is directing AI well and who is coasting.

The advice from EP #242 was specific: "Just measure the spend on tokens. And then gather all of the prompt history for each and every user. Because that is the only way the AI can analyze the efficiency."

Why $120B Cannot Buy Adoption

Here is the uncomfortable context. OpenAI has raised $120 billion. It is offering PE firms 17.5% guaranteed minimum returns to deploy AI across their portfolios. Anthropic is running a comparable pitch through a Blackstone JV.

The most-capitalised AI company in history is paying for enterprise distribution. That is not confidence. That is a confession. Enterprise AI adoption remains, structurally, very hard.

The data confirms the difficulty:

These numbers do not describe a technology problem. They describe an organisational architecture problem. The models work. The infrastructure is being built at industrial scale — ARM just broke a 35-year business model to ship its first-ever CPU specifically for AI agent workloads. The bottleneck is not compute, not capability, not capital.

The bottleneck is that most organisations cannot describe what they do in a form that AI can execute. We call this the Articulation Gap, and it is the binding constraint that $120 billion has not solved.

The Articulation Gap Is the Real 80/20 Problem

Consider what the 80/20 ratio actually requires. For AI to handle 80% of operating costs, it must be able to execute 80% of the work. For it to execute the work, someone must have articulated what the work is — decomposed it, described it, codified it in enough detail for an agent to act on it.

Most organisations have never done this. Not because they are lazy or resistant, but because they never needed to. Human workers fill ambiguity gaps instinctively. You give a competent person a vague brief and they figure it out. AI systems cannot. They need the brief to be precise, structured, and complete.

This is why 90% of AI deployments show zero impact. The technology is deployed into organisations that have not done the articulation work. It is the equivalent of buying a fleet of delivery vehicles for a company that has not mapped its routes.

The PE firms moving to 80/20 have understood this implicitly. Blundin's stepping stones are not just cost targets — they are articulation milestones. At each step, the organisation must have articulated more of its work into a form AI can execute. 50/50 requires partial articulation. 80/20 requires near-complete articulation. The ratio measures how much of the work has been codified, not just how much AI costs.

What the Remaining 20% Actually Do

If 80% of operating costs flow through AI tokens, what does the remaining 20% of human workforce do?

They do the three things AI cannot: they articulate, they judge, and they redirect.

Articulate: They describe what needs to happen in enough detail for AI to execute. This is not prompt engineering — it is organisational design. The articulator understands the business process, the edge cases, the context that never made it into any documentation, and they make that knowledge explicit.

Judge: They evaluate AI output against standards that cannot be fully specified in advance. The balance sheet that "didn't balance" in the Boardroom Brief this week was built by a CFO who could articulate what they wanted but could not judge what they got. Judgment requires domain expertise and the willingness to override confident-looking AI output.

Redirect: They notice when the context has changed and the existing articulation no longer fits. Markets shift. Regulations change. Competitors move. The AI keeps executing the old brief perfectly. The human notices that the brief is wrong.

An employee who can do all three — articulate, judge, redirect — is what Blundin means by "worth more than cost cutting." They are not doing the work. They are making the AI's work valuable. That is a fundamentally different job description than anything most organisations have today.

The Speed Differential Is Staggering

Blundin described what happens when PE firms apply this methodology: they can retool a company "in one-one thousandth the time it would have taken a year ago." Diamandis added: "You're going to have teams cherry-picking companies and reinventing them and disrupting them."

Salim Ismail, who is currently rewriting the Exponential Organizations framework because AI breaks it, was blunter: "Most large companies out there are walking dead. Their business models will be fundamentally disrupted in the next two to five years."

The ExO index — which ranked Fortune 100 companies on organisational characteristics — showed the top 10 outperformed the bottom 10 by 40x in shareholder returns over seven years. Ismail is rebuilding those criteria for the agent era: intelligence factor, MTP architecture, trust framework. The organisations that score well on the new criteria will be the ones where the 80/20 ratio works because the articulation architecture exists.

The PE firm that builds the agent-based company teardown and retooling methodology first captures a massive advantage. Not because they have better AI — everyone has the same models. Because they have the organisational architecture to deploy it at 80/20 speed.

The Governance Gap Is Already Visible

There is a cautionary signal in this week's intelligence that deserves attention. Secret CFO's Boardroom Brief reported that CFOs building their own AI tools are finding them "riddled with significant errors — the balance sheet didn't balance."

Stack Overflow's 2025 developer survey found that 46% of developers distrust AI coding accuracy and 66% are frustrated by AI-generated bugs. The people who build with AI professionally do not trust its output without verification.

Now consider: if the developers who build AI tools distrust AI accuracy, what happens when a CFO who does not build tools vibe-codes a financial model? The enthusiasm is real. The governance is absent. And the 80/20 ratio amplifies the problem — at 80% token cost, 80% of the output is AI-generated. If the remaining 20% of humans cannot validate that output, the organisation is not efficient. It is blind.

This is why the 80/20 transition is not just a cost target. It is a governance design problem. The 20% must include people who can judge AI output, not just direct it. And most organisations have not thought about who those people are.

Your Waypoint

Before your next leadership meeting, ask one question: if our competitors moved to an 80% token, 20% salary operating model over the next 18 months, which of our people would be in the valuable 20%?

Not which roles. Which people. By name.

The articulators — people who can describe what the organisation does in enough detail for AI to execute. The judges — people with enough domain expertise to catch AI output that looks right but is not. The redirectors — people who notice when the context has changed and the old brief no longer fits.

If you cannot name them, the Articulation Gap is your binding constraint. Start there. Identify the articulators. Build the architecture that makes the rest of your workforce articulators too.

That is the transformation that matters now. Not deploying more AI. Making the AI you have deployable.

If you want to track this shift weekly, our Transformation Intelligence Briefings cover the signals, the data, and the actionable implications — published free every Monday morning.