The human side of AI-era work
Adoption is counted. Adaptation has to be demonstrated.
Organizations count who has AI. Almost none can show what changed once they did — in how people work with the machine, with each other, and with the job itself. TalentVector studies that change and builds the instruments that make it observable. Demonstrated, not self-reported. Observed, not surveilled.
Read the evidence See what we built
One hundred people, counted. The structure between them only appears where something is observed — and what has been observed stays on the record. Move the cursor to look.
100 counted 0 in view 0 edges observed 0 on the record
The gap
The tools arrived in weeks. The work is still catching up.
Most organizations can tell you exactly how many people have AI. Far fewer can tell you what those people now do differently — which decisions they still make themselves, what they hand off, how they check what comes back, and whether their managers are normalizing any of it or quietly waiting it out.
That isn't a reporting gap. It's a category error. Adoption asks whether people use the technology. Adaptation asks what changed in the work. The first is easy to count and almost universally reported. The second is what leadership is actually paying for, and almost nobody can show it.
76% → 2%
Employees who report using AI in some capacity, against those who say all or most of their work is actually done with it. That second number has not moved since 2024.
76 of 100 counted · 2 of them (the wide marks) work with it McKinsey, April 2026 · Pew Research Center, October 2025 · single-source
84%
Organizations that have not redesigned jobs or workflows around AI, two years into having it.
84 of 100 organizations Deloitte, State of AI in the Enterprise 2026, January 2026, n=3,235 leaders across 24 countries · single-source
15 points
How much more optimistic executives are than their own employees about the difference AI is making.
The width of the gap between employees and their executives Google Workspace / Hypothesis Group, Beyond AI Optimism, December 2025, n=2,643 · single-source
Adoption counts the nodes. Adaptation lives in the edges.
adoption — counted
Access
Licenses issued, seats provisioned
Proves someone bought it.
adoption — counted
Activity
Logins, prompts, usage minutes
Proves someone opened it.
← most AI reporting stops here
adoption — counted
Attendance
Course completions, workshop seats
Proves someone was told about it.
← most L&D reporting stops here
adoption — counted
Attitude
Confidence surveys, readiness scores
Proves someone says they feel ready.
← most “adoption” studies stop here
adaptation — demonstrated
Demonstration
Behavior observed against a named standard
Proves someone can do it.
Where evidence begins. What NextPass produces.
adaptation — demonstrated
Transfer
Behavior observed in real work, over time
Proves the work changed.
What leadership actually needs.
Adoption theater is reporting the bottom four rungs as if they were the top two. Our work starts at rung four — and we've built the first instrument that gets there.
How we think about evidenceFirst instrument, in production
The thesis, built.
In 2025, a management program with fifteen years of clients had a problem every serious program has: it could prove people learned the method, and couldn't prove they used it. Rung-two evidence, when the client was paying for rung five. We built the instrument that closes that gap. It's called NextPass.
first observation Illustrative. A single observation is a snapshot; adaptation is a trajectory. weeks later
Built for
A proprietary management curriculum that needed evidence its behaviors transferred out of the workshop and into real conversations. The method stayed theirs. We built the instrument underneath it.
What it produces
Rung-four and rung-five evidence: each learner's behavior, demonstrated out loud, observed against the program's own standard, tracked across weeks — with the learner's practice kept private from their employer. Demonstrated, not self-reported. Observed, not surveilled.
What it proved
That the instrument thesis holds for one method — a program can show what changed, not just who attended. It now runs for training companies, certification bodies, consultancies, and internal academies, each on their own curriculum. The next question is how many methods it generalizes across.
NextPass
The practice-and-proof platform for training programs.
Three edges
Where adaptation actually shows up
Adoption metrics count the nodes: who has it, who logged in, who attended. Adaptation lives in the edges — in how a person now works with the machine, with the people around them, and with the job itself. Each of those can be observed.
With the machine
What people delegate, verify, challenge, and escalate. The most consequential AI behaviors are small, spoken, and habitual: the question asked before an output is trusted, the moment someone says “that's wrong,” the decision to keep a call for yourself. These are observable. They are rarely observed.
With each other
How work moves between people once AI is in the room — what gets expected, modeled, coached, and said out loud. This is what management actually is: the layer where one person's adaptation becomes a team's, or doesn't. Managers are the highest-leverage edge here; leaders set the conditions the edges sit in. Neither shows up in a license count.
With the work itself
What a role now is, and how the person holds it. As tasks move to machines, what's left is a new combination of judgment, exception handling, and ownership — and whether someone can perform it is a different question from whether a skills taxonomy says they should. How people relate to that work is visible in what they take on, what they bring forward, and what they let go. It is not visible in a survey.
The first instrument was built for the edge where adaptation transmits — the relationship between a manager and the people who report to them. The other two are where we're looking next.
How we build
We build instruments. Others run the observatory.
An instrument has to do four things: capture a behavior out loud rather than a claim about it, compare it to a standard the practitioner chose, show change over time, and do all of that without surveilling the person being observed. That is a product problem, not a program problem — which is why most training providers can't solve it and shouldn't have to.
Training companies, certification bodies, consultancies, and internal academies bring the method. We build what sits underneath it.
Our instrumentsRequirement
Captured out loud
A behavior, not a claim about one.
Requirement
Against a chosen standard
The bar belongs to the method, not to us.
Requirement
Over time
A snapshot is not a trajectory.
Requirement
Without surveillance
The person observed owns their practice.
Principles
Boundaries we keep on purpose
Why this, why now
I watched the same thing happen everywhere. The tools showed up. The training ran. The reporting filled with usage numbers. A month later, most people were working exactly the way they always had.
It wasn't the technology, and it wasn't resistance. Nobody had a way to answer the only question that mattered: did the work change?
That question is worth answering well. This is the biggest change in how people work in a generation, and we are measuring it with license counts. We built our first instrument for one management program that needed proof its method actually transferred. It did. Now we're after the larger pattern.
Someone should be able to show what changed. That's the whole company.
— Brett VanTil, Founder