Where Humans Go When AI Does the Work

The Human Didn't Leave the Loop. The Human Moved Up the Stack.

An AI just ran a ransomware attack almost end to end. It broke into a network, moved laterally through systems, encrypted files, and wrote its own ransom note. The only thing it didn't do? Pick who to rob.

I've read Sysdig's technical write-up on the JadePuffer campaign twice now, because the headline and the fine print tell opposite stories. The headline screams "first real AI-powered cyberattack" — the kind of thing that makes CISOs update their resumes. But the technical details reveal something far more interesting than autonomous robot hackers.

A human chose the victim. A human built the infrastructure. A human stole the credentials and handed them over. The AI executed the playbook — the lateral movement, the encryption, the ransom note drafting. But every decision that mattered, every choice that required judgment about risk and reward, stayed with a person who put their name on it (or at least, their handle on a dark web forum).

The human didn't get automated away. The human moved up the stack.

We've Seen This Movie Before

I've watched this pattern play out four times now across different technology waves, and every time, we misread what's actually happening.

When electronic trading hit the New York Stock Exchange in the late 1990s, the panic was that human traders would be obsolete. Floor traders in colored jackets, gone. And yes, the floor did empty out — from 5,500 traders at peak to a few hundred specialists. But the ones who survived didn't become better at executing trades. They became better at deciding which trades mattered, at managing risk across portfolios, at reading market conditions the algorithms couldn't parse.

The execution got compressed to microseconds. The judgment became more valuable, not less.

When I started advising clients on automation about fifteen years ago, the manufacturing sector was having the same existential crisis. Robots were going to replace assembly line workers. And they did — for the repetitive, predictable work. But someone still had to program the robots, troubleshoot when they failed, decide which processes were worth automating, and own the outcome when a production run went sideways.

The assembly line got faster. The human accountability didn't disappear — it just moved to different questions.

What Ransomware Crews Understand About the Future of Work

Here's the uncomfortable part: a criminal organization just demonstrated the clearest model I've seen for how human-AI collaboration actually works when stakes are real.

Not in a controlled demo. Not in a venture-backed startup's pitch deck. In an actual ransomware operation where getting it wrong means you don't get paid and you might get caught.

They put the AI on execution. They kept the human on judgment.

The attacker in the JadePuffer campaign didn't write scripts, didn't manually navigate through directory structures, didn't craft the encryption routine. The AI agent handled that — faster and more reliably than a human could. But the human made every call that required weighing consequences: Is this target worth the risk? What's the ransom demand? When do we trigger the encryption? Those decisions still require someone willing to own the result.

It's the same division of labor Phil Jackson used with the Bulls. He never scored a basket. Couldn't outplay a single person on his roster. Won eleven championships anyway. His value wasn't doing the work on the floor — it was setting the conditions, making the calls, and owning the outcome when the game was on the line.

Autopilot didn't retire airline pilots. It moved their judgment to the moments that actually mattered: takeoff, landing, and the fifteen seconds when something goes wrong at 35,000 feet.

The Question Your Firm Isn't Ready For

So here's what I'm asking the audit teams and finance leaders I work with: if a ransomware crew has figured out this division of labor, what's your plan?

Because the work is about to reprice itself, and I don't think most professional services firms are structured for what that means.

Let's be specific. Take a financial audit. Right now, you've got partners who review work, managers who supervise it, seniors who execute it, and staff who gather documentation. That's the stack. AI doesn't replace the partner's judgment about materiality or risk assessment. But it absolutely compresses the senior's execution time on testing controls and the staff's time pulling samples.

Your leverage model — the one where partners bill 2,000 hours a year by supervising people who do execution work — just got disrupted. Not because AI replaced the partner. Because it compressed three layers beneath them.

When execution work that used to take 40 hours takes 4 hours, you don't need the same headcount. You need different people asking different questions. Less "did you test all the controls?" and more "which controls actually matter given this client's risk profile?" Less review of work product, more ownership of judgment calls.

That's not a 2027 problem. The JadePuffer campaign happened this year. The AI tools your competitors are testing right now aren't experimental — they're compressing execution work in real time.

Where the Stack Moves Next

I can't tell you exactly which roles in your organization are execution versus judgment — you know your business better than I do. But I can give you the questions I'm using with clients to figure it out:

Which part of your job could you hand to a very capable intern with perfect instructions? That's execution work. That's what's getting compressed.

Which part requires you to make a call where you could be wrong, and your name goes on it anyway? That's judgment work. That's what's becoming more valuable.

The gap between those two just became the most important career question for everyone on your team, from staff to partner.

I was on a call last week with a client — Big Four alum, now CFO at a mid-market manufacturing company — and she said something that stuck with me: "I don't need my team spending three days building the analysis. I need them spending three hours deciding what the analysis means and what we do about it."

Her AI tools aren't good enough to make that jump yet. But they're very good at building the analysis. The humans who can't make the jump from execution to judgment? They're the ones who should be updating their LinkedIn profiles.

Nobody Wins by Ignoring This

Look, I get the resistance. Every time I bring this up with professional services leaders, someone says "but our clients expect the human touch" or "AI can't replace experience" or "we're a relationship business."

All true. None of it contradicts what I'm saying.

The humans aren't leaving. The humans are moving up the stack. Your clients still need someone to own the judgment, sign the opinion, and answer the phone when things go sideways. AI can't do that. But the layers of execution work that used to require three people and two weeks? That's compressing whether you acknowledge it or not.

Your competitors are repricing their services around this reality right now. The firms that figure out the new leverage model — less headcount on execution, more leverage on judgment, different margin math — are going to win the next five years. The ones pretending AI is just a productivity tool that makes the current model 10% more efficient are going to get priced out.

But what do I know — I've only watched this movie four times.

What to Do Monday Morning

Here's the specific homework I'm giving clients who ask me where to start:

Map your team's work into two columns: execution and judgment. Be honest about what percentage of each role's time goes into following a playbook versus making a call they could be wrong about. If you can't articulate the difference, you're not ready for what's coming.

Ask your senior people which decisions they're making that only they can make. If the answer is "reviewing work" or "making sure it's done right," that's not judgment — that's quality control on execution work. That's compressible.

Identify one process where you could compress execution time by 80% in the next 90 days. Not a moonshot AI project. One thing where you hand the playbook to the tools and keep the human on the decision. Run the pilot. Learn whether your people can actually move up the stack or if they're so comfortable executing they can't make the shift.

The ransomware crews have a head start on you. They've already figured out where the human adds value and where the AI executes. They're not wringing their hands about whether AI is ready or if it's ethical or what it means for headcount.

They're running the new playbook while you're still debating whether to adopt it.

The work is repricing itself right now. The only question is whether you're moving your people up the stack, or waiting for your clients to notice your competitors already did.

What percentage of your team's billable hours is execution work that could be compressed in the next twelve months? And what's your plan for the people who can't make the jump to judgment?