It didn’t work.

The friction wasn’t interpersonal. Nobody was difficult. The software engineers shipped continuously with weekly goals/milestones. The ML engineers iterated on models over months, worked in notebooks. We could notice that while the stand-up gave the engineers clarity, for the ML engineers it created more noise. The review process that worked for code made no sense for a model evaluation.

We weren’t collaborating badly. We were using the wrong unit of collaboration.

What eventually got the team into rhythm wasn’t a better stand-up format. Some squads even reduced the frequency of stand-ups.

We did a few things, like: we ran a Trust Summit with policy people, PMs from other teams, engineers, data scientists in the same room. We built planning weeks that weren’t really about planning: presentations, data insights, debates about what the domain actually was and what we were trying to protect.

The goal level is what aligned us. Most teams have OKRs. Fewer have goals that actually function at the value level, where the goal is an outcome, not a task with a deadline attached. The tooling caught up too: better deployment pipelines, model monitoring, shared data infrastructure.

Once people understood the domain deeply enough, the coordination started happening at the right layer.

The honeycomb

I used to teach the Spotify testing honeycomb to every engineering team I worked with. The insight wasn’t just about testing.

When teams moved to microservices, the testing pyramid stopped working. The biggest complexity was no longer inside the service, it was in how services interacted. Spotify named that shift and built the testing honeycomb around it. The microservice became the new unit.

The same shift is happening now with collaboration. Most teams and engineering leaders are running a model designed for a specific team composition: engineers who share tools, share a delivery cycle, share a definition of done. The stand-up, the PR review, the sprint ceremony, all of it was designed around the task as the unit of work.

We are seeing that the task is no longer the unit.

In the 2036 Job Ad I wrote about what this shift looks like for the individual: away from task execution, toward end-to-end value ownership. The individual contributor who used to own a feature now owns a workflow cluster.

The same logic applies one level up.

If the individual unit of work is shifting from task to value outcome, the collaboration unit has to shift with it. You can’t coordinate effectively at the task level when the work is operating at the value level. The rituals as we know them today were designed to synchronize people around tasks. These systems were not designed to synchronize people around outcomes that go across automated systems, human judgment, and AI-generated output all at the same time.

I am hearing about exhaustion. Less collaboration, more solo work, leadership pushing harder. The instinct in such moments is to run more rigorous review, tight feedback loops, clear handoffs. And these are the right instincts. But the system they're running them on has changed.

Looking back at Spotify, I don’t think their engineers just stopped caring about quality when the honeycomb landed. They updated the layer at which quality got measured.

There’s no honeycomb for collaboration yet. We are still in the process of naming the new unit of collaboration so we build rituals around it.

Photo by Jonas Hensel on Unsplash

Then I saw: the average confidence score for talking about their own value was 4.8 out of 10, across 27 users. No nines. No tens.

And when asked what they were optimising for: career clarity. Almost uniformly. Not promotion, not salary, not a new role.

I knew it in some form. We all see it or experience it. But seeing it in numbers is always a different story.

The abundance era

There’s a lot of talk about the era of abundance. AI unlocking creativity and AI empowering everyone.

But you have to be able to see what you’re worth clearly enough to reach for it or create the abundance. Lower confidence leads to taking less risks, settling for less, not seeing options that are actually there.

Most people from the experiment had access to managers, performance reviews, some form of feedback. The infrastructure existed. But these are the moments where we realize it was built for the company.

Felt the urge to do something about it

So I built a starting point.

CareerPowers is a set of open-source skills — md files you can run with any AI system — to help you see your career more clearly, on your own terms.

It maps six dimensions: Depth, Reach, Signal, Network, Optionality, Momentum.

I ran the career capital with my partner. He’s not a reflective thinker. He is more functional, concrete, not someone who spends time on this kind of thing. He looked at the output and said:

“Things that people probably already know, but never put into words. It gives you something to reflect on. It helps to think better about the things that truly make a difference and define you.”

The four skills in the first release:

• career-capital — what you’ve built, where your leverage sits, where the gaps are

• positioning-review — what you’re known for, what you want to be known for, the gap

• goal-setting — annual, quarterly, monthly goals aligned to your trajectory

• performance-review — achievements framed with impact, narrative support for the conversation

You might have versions of some of these at work. These ones are yours.

github.com/danaaonofriesei/careerpowers

Agents are next, of course. They’ll work for you, surfacing opportunities relevant to your trajectory, helping you plant the seeds for things you’ve been putting off.

If you try it, I’d like to know what you find.

Both sides have a point. They're just looking at different companies.

TrueUp is tracking 120k tech job cuts already in 2026 and project 361k by the end of the year. Challenger, Gray & Christmas, offering outplacement service, reported 217,362 cuts in Q1 alone, with AI cited as the leading stated reason.

I think there are 5 different categories of AI layoffs.

The rebalancing cycle, now with an AI budget line

Some companies would have restructured regardless. The business cycle turned, the growth thesis changed, leadership decided to realign. What’s different now is that every one of those conversations includes a new line item: AI infrastructure, compute, model access. That budget has to come from somewhere.

Meta just cut 8,000 jobs — 10% of its workforce — while guiding $115–135 billion in capital expenditure for 2026, almost entirely directed at data centres and AI infrastructure. The internal memo from chief people officer Janelle Gale described the cuts as a way “to offset the other investments we’re making.” Microsoft is running voluntary buyouts alongside it. These are companies spending more than ever. The people are competing with the infrastructure budget.

Is that AI washing? Partially. Is it also real? Yes. Because the capital reallocation is real, even if the productivity gains aren’t yet proven.

The productivity dividend, taken early

These are companies that have seen productivity improvements and are adjusting headcount.

Klarna is an example. Since 2022, headcount dropped from 5,000 to around 3,000, primarily through attrition and a hiring freeze timed with its AI rollout. They reported that its AI assistant was handling 66% of customer service chats and was estimated to drive a $40 million USD in profit improvement to Klarna in 2024. The cuts weren’t announced with a press release. They happened through not replacing people who left.

This is the most honest version of AI-driven headcount reduction. But Klarna eventually started rehiring in customer support after service quality dropped.

If you’re in a function where the productivity math is starting to work, this category might tell you something about the direction of travel, independent of this particular round of cuts.

The transformation bet

Block is the clearest example for me, and the most candid. On February 26, 2026, Jack Dorsey announced a cut of over 4,000 people — more than 40% of the company — despite Q4 gross profit of $2.87 billion, up 26% year-over-year. The business wasn’t struggling. His rationale: “Intelligence tools have changed what it means to build and run a company. A significantly smaller team, using the tools we’re building, can do more and do it better. I think most companies are late.”

This is a bet placed ahead of the evidence.

It didn’t look likeDorsey is claiming he’s already seen the full productivity gains at scale. He’s claiming he can see where this goes and would rather move now than be forced to move later.

If this bet is right, a lot of companies that look stable right now are going to face a more chaotic version of the same decision in 18 months. That’s the part that should make people in employed roles pay attention, not just the people who lost their jobs at Block.

The survival pivot

Some companies are cutting because AI or changes brought by AI are threatening the relevance of their product and business model. The workflows their product sat inside are being redesigned.

Chegg is the starkest example: students replaced its homework-help platform with ChatGPT directly, and the company has been in structural decline since. ProductBoard cut 30% of its workforce in April 2026 as AI-native product tools and platform consolidation from Atlassian, Microsoft, and Google ate into its market. Reading the letter of the ProductBoard CEO, I see between the lines: “we are sorry, but we have to pivot in order to survive”.

Atlassian cut 1,600 people (10% of its workforce) while cloud revenue grew 25%+. Atlassian CEO was direct: “It would be disingenuous to pretend AI doesn’t change the mix of skills we need or the number of roles required in certain areas.”

These companies aren’t cutting from weakness in the traditional sense. They’re cutting because the competitive landscape shifted and moving slower is more dangerous than moving now. The question for people inside these companies should no longer be “is the business okay?” . It should be “is the operating model changing, or just the headcount?” Those are different situations.

Some of the companies in this category look like this YTD:

The ZIRP correction, relabeled

And then there’s the category that doesn’t fit the AI story, even though it often gets labeled that way. An Oxford Economics report released in January 2026 found that many layoffs CEOs called AI-related were actually the result of past overhiring.

Companies that over-hired in 2021 and 2022, when capital was cheap and growth assumptions were overheated, are still working through that correction. Block’s own announcement acknowledged this: Dorsey noted the cuts were partly a reaction to COVID-era overhiring, alongside the AI rationale. AI gives these companies a cleaner investor narrative than “we made a bet that didn’t pan out.”

ZIRP = Zero Interest Rate Policy

So are AI layoffs real?

Yes. But the category determines what you’re actually navigating and what comes next.

I wrote recently about the seven people in every team meeting — seven different career anxieties, shaped by a different relationship with AI and work. The company-level taxonomy and the individual-level one are the same moment viewed from different angles. The AI layoff category your company falls into might shape which of those seven people you’re about to become, or already are.

Knowing which one matters because it changes what you’re actually dealing with.

Social preview photo by Nathan Dumlao on Unsplash

So I went looking for data.

I used the European ESCO taxonomy and analysed 12,572 skills mapped across every major employment domain in Europe, across healthcare, engineering, ICT, retail, agriculture, legal services, and eight other domains simultaneously.1

The most transferable skills

The most transferable skills (skills found across domains): manage staff, create solutions to problems, quality standards, adhere to organisational guidelines, manage budgets, use different communication channels, have computer literacy, identify customer’s needs.

You could bucket almost all of them under one word: judgment. The ability to read a situation, make a call, and move people through it.

But the one at the top is manage staff. Essential in 257 occupations, spanning all other skill domains.

Which is both obvious and worth saying out loud.

Managing staff

Managing people is not the same as managing a team. It’s not a headcount threshold you cross at five direct reports. It’s a set of instincts you develop every time you have to move something through humans.

When I was building the Trust & Safety engineering function at Trustpilot, the domain was new territory. For the company, and partly for me. What really moved the work forward wasn’t engineering depth alone. It was everything surrounding it: managing change across teams who hadn’t asked for a new function, getting alignment from stakeholders with competing interests, holding people through enough uncertainty that something real got built.

That’s the work. And it transferred when I moved across domains.

This isn’t an argument against technical depth. The ESCO data isn’t saying specialization doesn’t matter. It’s saying that what compounds across a career, across domain moves, across the unpredictable pivots most of us will make over a forty-year working life, is something different.

The skills at the top of the portability ranking are also, not coincidentally, the skills AI replicates worst. Models can pass engineering exams. How well can they manage a team through a restructuring? Or hold alignment across a leadership group that has stopped agreeing?

Most career development frameworks help us build depth. Almost none help us build the portable layer or name it clearly enough to develop it deliberately.

If you mapped your own career capital honestly, how much of what you’ve built travels?

But that’s roughly how most of us manage our careers.

And a career isn't a side investment. For most people, their ability to work and earn is their single largest financial asset. A 25-year-old earning €80,000 a year is sitting on potentially millions in lifetime earnings. Yet someone five or eight years in, has already made real bets on that asset (an industry, a skill set, a type of company) and has no way to tell whether those bets are compounding or stalling.

I’ve been thinking about this for a while. What would it look like if we applied the principles of financial management to how we think about careers? Concretely. The way someone would sit down with a spreadsheet, not a journal.

I have 3 principles for you.

Start with clarity

The most basic principle in finance is: know what you own. Every financial advisor starts the same way: let’s get everything on the table. Assets, debts, cash flow, exposure.

Now try doing that with your career.

Most people would describe what they hold as something like “I’m a senior product manager with ten years of experience.” Which is a bit like saying “I have some money in a bank somewhere.” It doesn’t tell you what’s actually in the portfolio, what’s appreciating, what’s stagnant, what might be quietly losing value.

No judging. This is how we were incentivised to describe ourselves. And knowing what you actually own in career terms isn't a straightforward task.

Career assets might include:

• deep domain expertise,

• a network of people who’d take your call,

• a public signal that attracts opportunities,

• a track record of judgment in high-stakes situations,

• or the optionality that comes from transferable skills.

If you are looking for tools that help individuals see their career capital, there is nothing. There are dozens of platforms that help organizations manage employees: Workday, Lattice and many more.

Liquidity

In finance, liquidity means how quickly you can convert an asset to cash. In career terms, it’s how quickly you can convert what you have into an opportunity: a job offer, a consulting engagement, a seat at a table.

Some people have deep career capital that’s completely illiquid. Their expertise, their judgment, their track record, all locked inside one company’s context, invisible to the outside market.

The relevant career liquidity question isn’t “how much capital do I have?” It’s “how much of it could I use tomorrow if I needed to?”

This is part of why we’re seeing such dramatic growth in fractional work. The number of fractional professionals doubled from 60,000 in 2022 to 120,000 in 2024. What I see in those numbers isn’t just a trend toward flexibility, it’s people actively trying to make their career capital more liquid. More portable. Less dependent on a single employer’s context.

Rebalancing

Any financial advisor will tell you to rebalance periodically. Check your allocation. Make sure you’re not overexposed to one sector, one asset class, one bet. Adjust based on where the market is heading.

A quarterly portfolio review is standard in finance. A career portfolio review? Most people have never done one.

has an article about how organizations are starting to grapple with this: that retention can no longer be about holding people in place or pushing them up a predefined ladder. It increasingly means helping people move forward, even when that movement doesn’t follow a linear path inside the company.

However, even some organizations start rethinking how they support careers, the individual still has almost no infrastructure of their own. No way to see what they hold, how it's changing, or whether anyone is even looking out for them.

The missing infrastructure

We have an entire infrastructure built around managing financial capital (advisors, dashboards, frameworks, quarterly reviews, risk models). And that makes sense. For career capital, which for most people is worth ten or fifty times more? We have a CV and an annual review if we're lucky.

We’re living longer, working longer, and increasingly not doing one thing for one employer across a single domain. Having more careers is no longer the exception. The career capital is getting bigger, more complex, and harder to see. And we’re still managing it the way we did when a career was a forty-year straight line.

I’ve been watching this pattern for months — across teams, companies, and conversations that start professional and end personal. The same anxiety, wearing different masks. And I think we’re making a mistake by treating this moment as one story — “AI is changing work” — when it’s actually seven.

Subscribe now

1. The Enhancer

Employed. AI multiplying their output. Quietly unsettled.

Their role looks better than ever on paper. They’re shipping faster, handling more, producing at a level that would have required a team two years ago. But if AI is making them 3x more productive, nobody’s paying them 3x more. The tool making them look good could eventually make them look replaceable.

The Enhancer’s problem isn’t capability — it’s articulation. “I’m using AI well” doesn’t translate into a raise. “My judgment about what to automate and what to keep human is the actual differentiator” — that’s closer, but who’s helping them frame it that way?

2. The Invisible Expansion

Employed. Scope doubled silently. No one noticed.

What used to be split across roles has quietly collapsed onto their desk, because they’re the person who figured out how to make it work. The reward for being competent right now is more work at the same pay. From the outside, things look smooth. That’s the whole problem — things are running smoothly because this person is absorbing unsustainable load. What they need isn’t a productivity hack. It’s a mirror — something that reflects the actual scope of what they’re carrying, in language their manager can hear and act on. Without that mirror, their choices narrow to burnout or exit.

3. The Identity Fracture

Employed. Craft devalued. Professionally disoriented.

This one hurts to watch. AI now does a version of the thing that made them proud. The copywriter whose first drafts are generated. The designer whose initial concepts come from Midjourney. They’re not unemployed. They’re something harder to talk about: underpurposed. The craft that formed their identity happens in seconds now, and the new value they’re supposed to provide, “curation,” “strategy,” “oversight” feels thinner, harder to point at. What they need isn't reskilling in the traditional sense. They need a new narrative about where their value lives.

4. The Competitive Displaced

Laid off. The role still exists. Someone else got it.

The most psychologically brutal position. They didn’t lose their job because the function disappeared — the hiring bar shifted and they were on the wrong side of a line that moved overnight. The job posting now reads differently: “experience with AI-augmented workflows,” “comfort with ambiguity in human-AI collaboration.” The fear is specific: not “will I find work?” but “am I already obsolete and nobody told me?”

5. The Structurally Displaced

Laid off. The role is disappearing.

Different in a way that matters: this person’s function is genuinely contracting. The advice they’re getting — “reskill into AI” — ranges from insufficient to insulting. They don’t need a bootcamp. They need a bridge — a way to translate twenty years of accumulated judgment, relationships, and domain knowledge into a role that hasn't been invented yet, or at least hasn't been named yet. They often have capabilities that are genuinely valuable in the new landscape, but those capabilities are locked inside job titles and role descriptions that no longer map to anything.

6. The Re-entry Cohort

Coming back. The world moved.

Almost nobody is talking about this one. People returning from parental leave, medical absence, or career breaks are walking into a workplace that transformed while they were away. Every other persona at least watched the shift happen. The Re-entry Cohort missed it entirely. They're managing the imposter syndrome that career breaks already trigger — on top of an AI capability gap. The AI tools that could accelerate their catch-up are themselves the thing they need to catch up on.

7. The Architect

Employed. Quietly redesigning everything.

Systematically rebuilding their own workflow — delegating execution to AI, keeping judgment for themselves. Often the most valuable person in the organisation. And often incentivised to hide it. An Architect who says "I automated 60% of my role and now spend my time on the 40% that requires human judgment" is one management conversation away from being told they should take on three more projects, or that perhaps the team doesn't need that headcount anymore.

So they stay quiet and the redesign happens invisibly. The judgment calls go undocumented. And the organisation doesn’t learn from what might be its most important transformation experiment.

The thread connecting all seven

Every one of these people has a value problem that is actually a language problem. In every case, the person has more value than they can communicate — and no infrastructure to close the gap.

I ran a survey recently. When I asked people what they think about when they think about their career, the majority said the same thing: comparing myself to others. The mirror most people reach for is someone else’s LinkedIn profile. Which tells you everything about why the value gap is so hard to close — the only tool we have for self-assessment is relative positioning, not actual articulation.

I see this in the data too. When people try to capture what they’ve done, the overwhelming pattern is execution language — what they did, not what it changed. The vocabulary for judgment and impact barely exists.

That’s the 70/30 flip in action. We built our entire career infrastructure — CVs, interviews, performance reviews, compensation frameworks — around execution. Nobody built the infrastructure for judgment.

Seven people in every team meeting. Seven versions of the same crisis. Seven flavours of “I know I’m valuable, but I can’t prove it in the ways the system recognises.”

Which one are you?

If this resonates, you’re not imagining it — and you’re not alone in it. I’m working on this problem. More soon.

I write about the intersection of AI, career capital, and professional judgment. Follow along if you want to be part of the conversation.

I surprise myself feeling relief when a chatbot gives me a short answer.

Because it costs me less to process. Less to read, less to evaluate. The short answer respects something the long answer doesn’t: I have a finite Tuesday.

I’ve been using AI tools daily — for writing, research, thinking through problems, building. I’m not an AI skeptic. But I’ve started paying attention to a specific friction: the tools are optimized for helpfulness, but their model of helpfulness is misaligned with how I actually work.

And I think this misalignment is creating a new kind of cognitive load that we’re all absorbing without realizing it.

Subscribe now

3 places where I feel it most:

The tool has no perspective on time

When I ask a colleague a question at 16:47 on a Friday, they calibrate. They give me the short version. They know I’m wrapping up, that I probably need a direction not a dissertation. That calibration isn’t laziness, it’s judgment. They’re reading the context around the question, not just the question itself.

AI tools don’t do this. Almost every answer arrives as if you have unlimited time, unlimited attention, and no other demands competing for either. Ask a simple question, get 800 words. Ask a complex question, same.

I noticed this even more sharply when I ran the clone experiment — my clone was far more verbose than I’d ever be.

The workaround people suggest is to add context to your prompt: “I have two minutes for this” or “give me the short version.” Sometimes that works. But notice what just happened: you absorbed the burden of managing the tool’s inability to read your situation.

You have to calibrate their perspective every time

With human colleagues, I build a mental model over time. I know that Sara tends toward progressive interpretations and Jon skews conservative. When either gives me input, I’m not just hearing what they say — I’m calibrating it against how they think. That calibration is effortless because I’ve built it over hundreds of interactions.

With a chatbot, I never build that model. I don’t know what it’s optimizing for beyond providing an answer. Is it giving me the balanced take? The safe take? The risky one?The take most likely to seem helpful? I’m doing extra cognitive work on every response — evaluating the answer while simultaneously reviewing the reasoning behind it.

With a colleague, that evaluation becomes effortless over time. With AI, it stays deliberate. Every interaction, from scratch.

The tool is rarely proactive

Good colleagues are proactive. Not in a dramatic way — in small, practical ways. They notice the project is creeping in scope and flag it. They save a document you’ll need later. They handle admin work for themselves without being asked because they can see it needs doing.

AI tools have the opposite design philosophy. Full agency stays with the user. The tool does nothing until prompted. In theory, this sounds respectful. In practice, it means you're carrying the full mental load — staying on track, noticing when the scope drifts, deciding what to keep, knowing when to stop. The work a good colleague absorbs quietly.

The irony is that the more capable these tools get, the more this passivity costs you. A tool that can do almost anything but initiates nothing is a tool that requires you to be the project manager of every interaction.

This isn’t a prompting problem

The standard advice right now is: learn to prompt better. Give more context. Set constraints. Build scaffolding.

That advice isn’t wrong, but it frames the entire misalignment as a user skill gap — as if the burden of making these interactions work should sit entirely with the person. Knowing how much to say, in what context, to whom — that’s judgment. It’s what makes a colleague genuinely helpful versus merely knowledgeable.

With AI tools becoming permanent parts of how we work, this misalignment isn’t a temporary inconvenience. It’s a design question. What does “helpful” actually mean when the person you’re helping has a finite Tuesday, can’t calibrate your perspective, and is already managing a dozen other demands on their attention?

“Maximally helpful in the moment” and “actually helpful across a working day” might be two very different things.

That gap is where the real alignment work is.

My AI content agent advised me against including this. I'm including it anyway.

• Social platforms were built on the premise that humans hold full agency — that people would decide how much time to spend on them. But when teenagers spend 9 hours a day by design, we can see that this was never really about human agency. I'd rather we didn't repeat that assumption with AI tools.

In 2024, Stanford researchers interviewed 1,052 people — two hours each, covering "lives and views on controversial issues" — then built AI agents from those transcripts that could simulate each person’s responses. When tested, the agents matched participants’ answers 85% as accurately as the participants matched their own answers two weeks later.

The researchers built this for policy simulation. I had a different question: what happens when you apply the same architecture to careers?

Specifically — could a structured interview, processed through the right synthesis layer, reveal how a professional actually thinks, decides, and operates? Including things they can’t see about themselves.

I decided to find out. On myself.

Subscribe now

How I did it

I recorded a 30-minute interview with myself. Not a CV walkthrough. Questions designed to surface behavior and moments rather than self-description.

Describe a time things were genuinely ambiguous and high-stakes. How did you navigate it?

When have you held a position most people disagreed with — and were you right?

The design principle comes from the Stanford research. Their interview-based agents outperformed agents given only demographic data — not because interviews were longer, but because they captured the specific, unguarded moments that break stereotypes.

I ran the transcript through four analytical lenses: leadership psychology, career strategy, behavioral economics, and systems thinking. The output was a set of what I’m calling Judgment Currents — stable patterns in how I reason, create value, and navigate difficulty. Not skills. Not personality traits. Something closer to operating signatures: the recurring logic underneath decisions that shows up consistently across different contexts and years.

That synthesis alone told me more about how I operate than a 200 EUR coaching session. And it took an hour, not six weeks.

I loaded both layers — raw transcript and synthesis — into an AI agent. For the experiment I used a Claude Project. Nothing exotic.

Then I designed 10 questions to test accuracy. Not generic questions any smart AI could answer plausibly — questions requiring knowledge of my specific history, values, and reasoning patterns. I answered all 10 in writing first, sealed them, then ran the clone through the same questions. No cross-contamination.

The results

The results were 3 full matches, 5 partials, 2 misses.

The matches were unsettling in their own way. On behavioral questions — how I navigate ambiguity, how I build trust with teams — the clone reconstructed my thinking accurately enough to be uncomfortable. It had absorbed the pattern from the evidence.

The partials were more interesting. The clone was consistently more impressive than me. More articulate, more resolved, more philosophically elegant. It moved through every question to a conclusion.

I found that genuinely fascinating — and also suspicious.

Three patterns kept showing up in where it diverged.

• First, verbosity. I answered one question in a single sentence: feedback, honest conversation, one month max. The clone built an architecture around the same conclusion. Knowing when you’re done is its own form of judgment — and it’s one the clone didn’t have.

• Second, the clone skipped the hesitation. I sometimes don’t know my answer until I’ve sat with a question for a moment. The clone moved straight through to somewhere polished.

• Third, orientation. I locate myself before I evaluate a situation — my constraints, my actual available moves — and then I assess what’s in front of me. The clone looked outward first. It was reasoning from an idealized position rather than an actual one.

The places it missed weren’t random. They were consistently the questions where the honest answer is "I don't know yet."

What the experiment revealed

When writing, analysis, coding, and coordination become fast and affordable, what’s left that’s distinctively valuable?

The answer I keep returning to is judgment. The specific patterns in how a person reasons through ambiguity, recovers from failure, holds uncertainty, and makes calls when the situation is genuinely unclear. That becomes scarce when execution becomes abundant.

But judgment has a problem skills don’t: it’s still largely invisible — even to the person who has it.

It lives in the accumulated evidence of how someone has actually operated across years of real work. In the decisions made at ambiguous junctures. In the positions held and revised. The clone revealed something specific about where that evidence is richest and where it goes thin. Observable behavior — how you navigate conflict, how you build trust — leaves traces. The harder stuff doesn't leave the same traces. How you orient yourself before deciding. What you're willing to say you don't know yet.

That’s not a model problem. That’s a capture problem.

The clone’s limitations weren’t a failure of the model. They were a failure of inputs.

Most professionals I talk to can describe their career. Very few can describe how they think. The CV captures the what. The how stays invisible, even to them. And that’s exactly the thing that’s becoming most valuable.

The experiment worked well enough to be worth repeating. And the gaps were more interesting than the matches because that's where judgment actually lives.

This is the second article in a series on judgment as a career asset in the age of AI. The first, on the 70/30 Flip, is here:

If you’re curious what your own evidence layer looks like — the patterns in how you actually operate, not how you’d describe yourself in an interview — I’m running this with a small group. DM me.

To understand why, it helps to think about what most of our work actually is:

• Infrastructure — the systems and platforms everything runs on

• Workflows — the processes and rituals that coordinate how work gets done

• Judgment — the calls people make about what to build, what to prioritize, and what "good" looks like

What I’m watching collapse right now is the middle layer. The workflows.

Take project management tools. Jira is the obvious example, but this applies broadly. These tools were built to solve a specific problem: execution was the hard, slow, expensive part of building software. You needed somewhere to track what was left, who owned what, how confident you were shipping on time. Tickets, sprints, status updates, standups — the whole ritual existed because the gap between “we decided to build this” and “this is shipped” was long and unpredictable. The workflow was a coordination layer for that gap.

That gap is compressing. Not closed — but compressing.

This is my read on what’s happening to project management software — not that someone will rebuild Jira with Claude Code (though they will), but that the teams who used to live in Jira are starting to work differently. The rituals are breaking. Not replaced by something cleaner. Just no longer the thing holding everything together.

What replaces them? I genuinely don’t know, and I don’t think anyone does yet. I’m watching engineering teams run experiments in real time — some dropping sprint ceremonies entirely, some delegating all of that to agents, some running hybrid systems that don’t have a name, some moving toward no tickets or fewer tickets and more principles. None of these is the new workflow. They’re all mid-renegotiation.

What’s collapsing is the middle — the part that touches how people actually spend their days.

Think about what Jira-era work actually required: knowing how to break down a project into tickets, managing sprint ceremonies, keeping status visible to stakeholders. Those skills were real. But Jira taught you what to value, what to measure, what "good" looked like. Take Jira away, and a lot of people discover they've been describing themselves in workflow language:

“I run sprints.” “I manage delivery.” “I keep teams aligned.”

Those are true — but they’re descriptions of the coordination layer, not of the judgment underneath it.

The judgment is the part that travels. The ability to read where work is actually stuck. To know when alignment is real versus performed. To make a call with incomplete information. None of that lives in Jira. It never did. But Jira gave it a shape, and now that shape is changing.

How much of your role was shaped by a workflow that’s now shifting? What’s actually yours?

Not the sprint ceremony. Not the status update. The judgment you developed while doing all of it.

That part doesn’t expire. But you have to be able to see it separately first.

About the role
You will own three automated workflow clusters generating €2.4M in annual recurring value. Your primary job is not to execute — the workflows handle that. Your job is to decide, fast, when the workflows can’t.

What you’ll be responsible for

• Making 40+ judgment calls per day — often with incomplete data, always on the clock

• End-to-end ownership of value creation: you will know, at any moment, exactly what your work is worth to the business

• Coordinating between human collaborators and automated systems — you are the judgment layer

What we’re looking for

• Demonstrated efficiency improvements on hybrid human-automated workflows

• Comfort with ambiguity at scale — not occasionally, but as your default operating mode

• Prior experience taking something from zero to running and knowing exactly how it performed

Managerial experience not required. Managerial instincts: mandatory.

Read that back.

If your first reaction was “that’s not a real job” — you’re right, it isn’t. Not yet. But you’ve probably done some version of almost everything in that ad. The job title is made up. The skills aren’t.

Here’s what the ad is actually measuring.

Decision velocity is what happens when you’re managing automated workflows instead of doing the tasks yourself. You’re no longer the person writing the report — you’re deciding whether the report the system generated is good enough, fast enough, for the next ten things waiting behind it. Ambiguity doesn’t disappear. It multiplies. The volume of moments where you have incomplete data and still have to call it will go up, not down.

Most people have never tracked how fast they make decisions. Automated workflows that pause for human approval will change that — every decision logged, timestamped, ranked by impact.

End-to-end value ownership is already walking through the door. I had a Senior Software Engineer on my team who didn’t want to just write good code. He wanted to own the whole thing — the product, the business line, the hiring, the launch. His title said one thing. His instincts said another. He collaborated, but the drive to own the full arc was just there, in him. AI didn’t create that desire. It will just make it possible for far more people.

Managerial instincts without the title is the coordination piece. The ad isn’t asking for someone who has managed people. It’s asking for someone who can run systems: understand what a workflow is optimizing for, notice when it’s drifting, when it’s becoming too expensive, know when to override it and when to let it run.

You’ve probably done this already. The time you realized a process your team relied on was quietly producing the wrong output and you fixed it before anyone asked you to. That’s operational judgment. Most people have exercised it. They just haven’t called it that.

The actually useful question this fake ad raises isn’t whether this role will exist. It will in some shape. The question is: do you know how to talk about yourself in these terms?

Most people don’t, not because they lack the skills, but because they’ve never had to name them. That project you drove without a clear mandate? Decision velocity. The time you owned something nobody asked you to own? Value ownership. You’ve been building this. You just filed it under the wrong name.

The gap isn’t capability. It’s translation.

The room nodded. Everyone agreed. Not everyone took action.

What I was actually observing

This wasn't just something I noticed in my own teams. Boards were getting more demanding, investors were asking harder questions about engineering ROI, and the leaders who held their ground were the ones who could talk about their work in terms the rest of the business recognized. Someone would say about them: they're getting it.

It’s been a few years now that engineering leaders have been expected to behave more like business leaders, communicate in impact, justify in revenue terms, use the currency the rest of the company runs on. That expectation isn’t new. What AI is doing is amplifying it, and compressing the timeline for leaders who’ve been slow to close the gap.

The “trust me” era is quietly ending

Here’s what I’m seeing: the complexity that once gave engineering leaders a certain amount of cover is becoming more visible to everyone else. CEOs and CFOs can get reasonably educated on what AI enablement should produce, what developer productivity actually means in output terms, what good looks like. The technical mystique isn’t gone, but it’s thinner than it was.

Right now, the market is measuring AI success mostly through adoption metrics. How many engineers are using the tools? What’s the activation rate? That’s where we are. But that phase has a shelf life. The next round of questions will be about what it produced, in revenue, in speed, in customer impact. And those are conversations that happen across functions, not just inside engineering.

The leaders who can show up fluently in those conversations will have real leverage. The ones still relying on “just trust me on this investment” will find that move doesn’t land the way it used to. The information gap closed.

The same technology creating this pressure is also the fastest path through it.

Look at what’s happening with junior engineers right now. They’re getting exposure to architecture decisions, complex tradeoffs, cross-functional context — these things used to take years to accumulate. They’re not skipping the learning. They’re accelerating it. The judgment that used to require five years of scar tissue is developing faster. DX research found that newly hired engineers using AI daily hit their 10th pull request in 49 days, compared to 91 days for those without it. And that productivity gain continues well past onboarding.

Engineering leaders can close the business fluency gap the same way. Not by outsourcing the thinking, but by increasing the reps or getting more exposure. More simulations of the CFO conversation before you’re in it. More pressure-testing of your own logic before someone else does. More exposure to how finance, sales, and product read the same information differently than you do. The skill was always buildable. What’s changed is how quickly you can build it.

The engineering leaders coming out of this moment with more influence probably won’t be the ones who waited for the organization to meet them halfway. They’ll be the ones who closed the business fluency gap while it was still optional.

It doesn’t show the feature you killed before it became a costly mistake. The strategic pivot you pushed for when everyone else wanted to stay the course. The moment you realized you didn’t know the right answer, so you ran small experiments to figure it out instead of committing to the obvious path.

Your resume shows what you did. It doesn’t show how you figured out what to do: the judgment calls, the navigation through uncertainty, the resourcefulness when the answer wasn’t obvious.

AI is making that gap matter.

The Ratio Flip

Hiring has always evaluated two things: skills and judgment. Skills are your ability to execute: write code, design interfaces, analyze data, manage projects. Judgment is your ability to decide what’s worth executing in the first place.

If I had to put numbers to it, the ratio has been roughly 70% skills, 30% judgment for most of modern work history. Employers primarily wanted to know: can you do the job? Judgment mattered, but it was secondary.

AI is flipping that ratio—not removing skills from the equation, but changing their weight.

This shift was already underway and anyone who’s been in the workforce for a decade has felt it. The nature of work has been moving from execution to decision-making, from individual contribution to strategic thinking. AI is just accelerating what was already happening, compressing a shift that might have taken two decades into something like five years.

We’re moving toward a 30% skills, 70% judgment world. But the infrastructure we use to demonstrate capability? That hasn’t caught up.

What Judgment Actually Looks Like

Judgment is harder to define than skills, but you recognize it in your own work.

It’s when a product manager realizes a feature shouldn’t be built at all because it solves the wrong problem—even though building it would look good on their performance review.

It’s when a designer recognizes the real issue isn’t the interface but the business process behind it, and redirects the conversation accordingly.

It’s when a marketer sees technically correct data pointing toward a campaign direction, but knows from market context that the numbers are leading them astray.

Judgment is the meta-skill of knowing what to do with your skills. It’s context-dependent decision-making. It’s navigating tradeoffs and choosing between competing goods rather than obvious rights and wrongs.

If you’ve been working for more than a few years, you’ve developed this. You know it when you override the obvious answer because you see something others miss. You know it when you choose not to do something, even though you could do it well.

The problem? You have no good way to demonstrate it.

The Infrastructure Problem

Here’s where we are: the criteria for what makes someone valuable at work is shifting, but our infrastructure hasn’t caught up.

You can feel the mismatch. When you update your resume and realize it captures almost nothing about your actual value. When interview questions feel disconnected from what you actually do. When job descriptions miss what the role is really about.

I’ve talked to mid-career professionals who look at their resumes and feel deflated. “It doesn’t show how much I’ve grown,” one person told me. “It doesn’t capture the challenges I navigated. It makes me feel like I didn’t achieve much, even though I know I did.”

They’re right. Resumes show credentials, titles, and responsibilities. They show outputs, not decision-making. They’re optimized for a world where execution was the scarce resource.

Portfolios have the same limitation. They show the artifact—the code, the design, the campaign results. They don’t show the judgment that created it. The three other approaches you considered and why you chose this one. How you navigated conflicting constraints or stakeholder needs.

Even interviews struggle with this. “Tell me about a time you did X” can surface judgment, if the interviewer digs into the context, the alternatives you considered, how you figured out what mattered. But that requires skill and intention. Many interviews settle for confirming you did the thing.

Great hiring managers have always looked for judgment, but not always formally. Now judgment needs to be the primary criterion, and we’re still using tools designed to evaluate skills. We keep using them partly from inertia, partly because we don’t know what replaces them.

What’s Next

We’re in between paradigms. The old model is breaking down but the new one doesn’t exist yet.

But watch for the early signals. Some people are writing narrative case studies that show their decision-making process, not just outcomes. Others are building in public—documenting not just what they shipped but how they figured out what to ship. A few companies are experimenting with work trials and case-based interviews that surface judgment directly.

These are workarounds, not solutions. Nobody’s built the infrastructure yet.

But here's what I'm confident about: the companies that figure out how to evaluate judgment rather than just credentials will hire better. And the individuals who figure out how to demonstrate judgment will have an advantage that compounds.

The question isn’t whether the shift is happening. It’s whether you’ll position yourself for the world that’s coming, or keep optimizing for the one that’s fading.