The Talent Chaos Tax: The Value of Human IP

Part 3 of The Chaos Tax series

In the first two parts of this series, we mapped the Chaos Tax: the compounding cost of operating in a market that reinvents itself every quarter, and where it lands — in the P&L, in the workforce, and in marketing. This piece is about tax avoidance. It starts with the most valuable asset your company owns: your people. We’ll trace it across four dimensions — the individual expertise each person carries, the collective intelligence they build together, how you measure both, and how you design AI to compound them.

But first, the layoffs.

Companies continue to announce AI-driven workforce reductions, mostly in anticipation of capabilities that don’t yet exist. We covered the numbers in Part 1: executives cutting roles ahead of any mature AI replacement, and 55% of employers already regretting it.[1] In May, Jensen Huang said blaming AI for layoffs is “lazy,” and that executives invoke it “to sound smart.” “AI has just arrived. How is it possible they’re already losing jobs?”[2]

Huang’s point is about timing: the technology is not ready to replace these jobs. Our point is about what gets lost when companies cut talent in favor of AI anyway. It doesn’t mean every role is sacred. Plenty of tasks should move to AI, and some roles will shrink or disappear as it does. The work is to figure out the right collaboration between human and AI, based on what each does best. So before you cut, know which work is your edge, because that is the work you cannot buy back. Gina Raimondo made the business case at TED this year: displace your people to win on AI and you lose anyway. “If we’re the best in the world at [AI], but we’ve displaced millions of American workers, then we’re going to lose the global AI race. In fact, we will have automated our decline.”[3]

Your best unmeasured asset

Every company runs on several kinds of intelligence. The data you collect or buy, and the AI you license, are available to your competitors too. The intelligence that separates one company from another is human, and it comes in two layers.

The first is individual: the expertise, reasoning, experience, taste, and judgment each person brings to their role. The scientist who knows which result is worth exploring, and which obscures the root cause.

The second is collective. It is what happens when individuals work together: the shorthand, the trust, the unwritten knowledge of how things get done. It is why companies exist — you build not for individual tasks but for what a particular combination of people produces. That combination is unique to your company.

The two layers together are your Human IP. It is alive, and the source of everything that makes your company unique. The market already notices a unique edge. Intangible assets — brand, know-how, relationships — now make up roughly 92% of the S&P 500’s value, up from 17% in 1975: a wholesale inversion from worth you can touch to worth you can think.[4]

That is why the emerging idea of coding human intelligence is wrong: capture what your experts know into a system, then let the experts go. Record only an expert’s conclusions and you have a snapshot that is immediately out of date. What is worth capturing is the methodology behind the judgment — how your best people reason and work — so the rest of the organization can work with it too. It lets AI take the work it does better off their hands, freeing them for their sharpest thinking. And it puts that reasoning into a system you keep improving, so your Human IP compounds instead of decaying: always improved by humans, always amplifying them. None of this means you stop investing in the experts who generate that reasoning; that is what keeps the system alive.

Winemakers have a formula for why the same grape, grown the same way, produces a different wine in each vineyard: terroir + the winemaker. The soil, the slope, the weather, the accumulated character of a specific place — that is the terroir. The judgment, decisions, and vision come from the winemaker. Companies have a similar formula. Everyone now has the same equipment: the same technology, the same data. What makes valuable wine is your winemaker (the judgment of your people) and your terroir (the existing IP and competitive edge you already have). Let the right AI tools square that formula and your uniqueness scales. Use AI for efficiency alone and you get slop swill. The wine world already lived this: as producers everywhere chased the same critics’ scores, regions that once tasted of their own place converged on one bold, oaky “international style” — technically polished and largely interchangeable.

AI hands us an opportunity we did not have a few years ago. You can take those two layers of human intelligence and multiply them instead of replacing them. We call the result collaborative advantage, and it is where this series goes next. First, though, you have to know what you are amplifying.

Sameness won’t win

MIT Sloan made the point before the current wave crested: when AI becomes ubiquitous, it lifts markets but rarely benefits one company uniquely.[5] Content volume is exploding while two-thirds of customers say most brands seem the same.[6] When every company automates toward the same outputs, sameness becomes the default.

Most companies hold a register of every software license and depreciate every laptop. Almost none hold any account of the expertise, judgment, relationships, and collaborative patterns that win them customers. We treat the commodity as an asset and the asset as a cost.

And there is no generic fix, because Human IP is specific by definition. BCG’s Vinciane Beauchene refused to give her audience a list of forever-human skills: “There is no universal list. Each company needs to figure it out based on its strategic positioning.”[7] That is exactly why off-the-shelf AI playbooks produce off-the-shelf results. What is irreplaceable at a pharmaceutical company is not what is irreplaceable at a fashion brand, or at yours.

Put a number on it

In Part 2, we made the case that organizations protect what is easy to measure, not what creates the most value. Human IP is the clearest example. What it can’t measure, it can’t defend in a budget meeting. It moves the numbers; it just doesn’t get the credit.

Start with attrition. Replacing an employee costs as much as 213% of salary for key roles, once you account for lost productivity and the knowledge that leaves with them. Reduced productivity alone makes up 42% to 66% of that cost.[8] At company scale the figure is large: firms in the S&P 500 lose between $228 million and $355 million a year to attrition and disengagement, and large companies lose roughly $47 million a year to knowledge that wasn’t passed on.[9] Markets price this too. Key-person risk lowers a company’s valuation, and leadership departures move share prices the day they are announced.[10]

The companies that design AI around their people, rather than against them, are pulling ahead on the metrics boards watch. BCG found that the 5% of firms it calls “future-built” — the ones designing AI around human and machine collaboration — posted 1.7 times the revenue growth and 3.6 times the three-year shareholder return of their peers.[11]

So measure it the way you measure anything you intend to grow. Track the attrition of your highest-judgment roles, the collaborations that produce your best work, and the decisions where a person changed the outcome. What you measure, you can protect. What you protect, you can compound — now with the help of AI.

Design for human and AI collaboration

The early evidence is consistent. Harvard Business School researchers found this spring that job postings have fallen 17% in automation-heavy roles but risen 22% in roles built around human and AI collaboration.[12] A companion Harvard Business Review analysis concluded that automation shows early gains, while augmentation wins long-term.[13] The market is repricing collaboration above replacement.

The clearest case is the profession AI was supposed to kill first. In 2016, Geoffrey Hinton said we should stop training radiologists. The Mayo Clinic did the opposite. Its radiology staff has grown 55% since then, to around 400, while the institution runs more than 250 AI models that sharpen images, flag abnormalities, and clear routine work.[14] It did not hire more radiologists despite the AI. It hired them because of it: every model made the human judgment on top of it more valuable, and those radiologists now reach more patients and save more lives.

Pfizer put AI to work across drug research, searching more than 20,000 documents per drug and saving roughly 16,000 hours a year across 1,500 scientists.[15] That is the right work to give a machine: document search and pattern matching, the parts that aren’t the scientists’ edge. The expertise and judgment stayed with the humans. The thesis was not “automate the lab.” It was to make the most valuable expertise in the industry faster and braver.

We see the same logic in our own work. One client — a global media brand with one of the most recognizable voices in the world — used AI not to generate that voice but to build an organizational muscle to use it more consistently: the skill, governance, and daily habits that let a small team protect and scale it. The starting question was not “what can the AI do?” It was “what must the humans keep doing, and how does the AI multiply it?”

We run Atmos on the same idea. The layer we put on our AI tools integrates the models with our own terroir: our judgment, our workflows, and the way we reason through a client’s problem. It captures how we think so it can amplify how we work, not stand in for us.

One more thing worth naming — not as an asset to file away but as a deeply human trait: ego. Competitive drive, conviction, the stubborn belief that you are right when the data says you are early. No system has captured it, and none will. But you can build to amplify it. Make the AI safe, testable, and reversible, so your people can take bigger risks with their ideas. The machine’s job is to lower the cost of being wrong; the ambition stays human.

Ask these before you automate

Microsoft’s research team posed a great question: would you rather have “a tool that thinks for you, or a tool that makes you think?” Their studies found that badly designed AI assistance produces measurably fewer ideas and less critical thinking — what they describe as a hive mind that “keeps suggesting the same five ideas.” Designed differently, the same technology has the opposite effect.[16]

So before the next automation decision, a few questions.

• If an AI could take over all of your team’s tasks, who would you keep, and why?
• Have you captured how your best people reason and work, in a way that makes them and everyone around them better? Or did you lay those people off already?
• When you look at your AI roadmap, is it designed to amplify what makes your company distinctive, or around what the vendor’s product does currently?

Nobody is arguing against efficiency. The argument is for knowing what value you have before you automate it away. The companies that come through this era with a real edge will be the ones that treated their Human IP as an asset class: they identified it, measured it, invested in it, and designed their AI to compound it. How to do that is where this series goes next.

Your vineyard is yours to harvest, or to pave over. Make 2026 your best vintage yet.

Emma Cochrane is co-founder of Atmos Partners, a strategic advisory firm helping PE-backed and global enterprises design for AI value. She writes about what separates the companies that compound their human advantage from the ones that automate it away. If you’re ready to find out what’s irreplaceable in your business — before you automate it — get in touch at [email protected].