AI Is Doing 75% of the Work — So Why Are You Working 5x More?

A study out of UC Berkeley's Haas School of Business followed 200 employees at a U.S. tech company for eight months — a company that had fully embraced generative AI. What they found should be mandatory reading for every founder, operator, and automation consultant alive right now.

AI was handling a significant chunk of the work. Productivity metrics looked strong. And employees were burning out at an accelerating rate, working longer hours and absorbing more responsibilities than ever before.

5× — More tasks absorbed by employees in heavily AI-integrated companies 31% — Of workers say AI increased their workload after implementation 2.2 hrs — Extra work per week added for every step up in AI exposure (CEPR)

This isn't a fringe finding. Multiple independent research bodies — from Berkeley to the Centre for Economic Policy Research — are converging on the same uncomfortable truth: automation without architecture doesn't reduce work. It amplifies it.

Why This Happens: The Three Traps

I've consulted with dozens of businesses implementing AI across their operations. Before I understood this dynamic fully, I saw it play out firsthand. Here are the three traps that turn your automation investment into a burnout machine.

Trap 1 — Task Expansion (The Scope Creep Spiral)

When AI makes an employee faster, the natural organizational response is to give them more to do. No one intends for this to happen — it's an emergent property of productivity gains meeting human ambition. Work that previously justified hiring a new person suddenly gets absorbed by whoever has AI capacity. Headcount freezes. Scope expands silently.

"Employees absorbed work that might previously have justified additional help or headcount. They usurped each other's roles, took on coaching responsibilities, and corrected AI-generated output — all without a pay increase or a new job title."

I call this the Invisible Promotion Problem: your people level up in output but not in compensation, recognition, or role definition. The resentment follows eventually.

Trap 2 — Boundary Dissolution (The Always-On Effect)

The Berkeley study found employees were prompting AI tools during lunch, between meetings, while waiting for files to load, and during breaks. The friction that once created natural stopping points — "I can't do this until I'm back at my desk" — evaporated. The result was a workday with no real edges.

AI surveillance tooling in some organizations is exacerbating this. Workers in AI-monitored roles are extending their hours not because they're asked to, but because algorithm-driven performance scores create ambient pressure to stay visible and productive at all times.

Trap 3 — The Ratchet of Expectations

This is the most structural problem of the three. Once AI makes a team capable of delivering at a higher output level, that output level becomes the baseline expectation. There is no going back. Clients, managers, and market competitors all recalibrate upward. The productivity gain that was supposed to give your team breathing room instead becomes the new minimum viable standard.

⚠ Founder Warning: If you're running a small team and using AI to do the work of a larger one, you're not building leverage — you're borrowing against your team's wellbeing. The debt comes due in attrition, burnout, and degraded decision-making quality.

What the Data Actually Says About AI Productivity

Here's where I want to be precise, because the picture is genuinely complicated.

A National Bureau of Economic Research study surveying over 6,000 executives across the US, UK, Germany, and Australia found that the vast majority reported little to no measurable impact from AI on employment or productivity. Meanwhile, a separate analysis suggested U.S. productivity grew by 2.7% last year — with strong indicators that AI adoption is beginning to bear fruit at the macro level.

Both things can be true: AI is starting to move the aggregate productivity needle while simultaneously, at the ground level, creating unsustainable conditions for the individual workers powering it.

This is the gap I spend my entire professional life trying to close. The macro story can be positive and the micro reality can still be painful — unless you design for both deliberately.

The Fix: Automation That Actually Liberates

This is the part of the conversation that rarely gets airtime, because it's harder to solve than it is to diagnose. Here's the framework I use with every business I work with.

Step 1 — Map Capacity Gains Before You Reassign Work

When you implement an automation that saves your team 6 hours a week, that time should be documented, tracked, and intentionally allocated — not quietly swallowed by the next thing on the backlog. Treat time savings the same way you'd treat recovered budget.

Step 2 — Define What "Done" Looks Like in an AI-Augmented Role

Most job descriptions haven't been updated for the AI era. They still list tasks that AI now handles while adding no clarity on the new judgment-intensive, AI-supervisory responsibilities that replaced them. Ambiguity about role scope is where burnout breeds fastest.

Step 3 — Build an AI Practice, Not Just an AI Stack

The Berkeley researchers concluded that companies need what they called an "AI practice"—deliberate norms for how, when, and at what pace AI tools are used. This means:

Protecting time blocks explicitly not for AI-assisted work to rebuild judgment and creativity muscles.

Building in regular team rituals that are human-only — not because AI can't be involved, but because connection and trust require the friction of unaugmented interaction.

Setting output ceilings, not just floors. If a team can now produce 3× the content, that doesn't mean they should. It means you can produce better content with more time for quality control.

Auditing AI workload quarterly — not just AI efficiency. Ask: "Is this making my people's working lives better?" as often as you ask, "Is this saving time?"

Step 4 — Hire Into the Capacity You've Created

The most expensive mistake I see growth-stage founders make is using AI savings to freeze headcount indefinitely. AI-driven efficiency creates organizational slack. That slack is for strategic expansion — new products, new markets, better service quality — not for running a 10-person team at a 30-person output forever.

My Take: This Is a Design Problem, Not a Technology Problem

AI is not the villain in this story. The tools are extraordinary. What's missing — in most businesses I encounter — is the intentional design layer between the technology and the people using it.

The companies that are actually winning with AI right now are not the ones with the most tools in their stack. They're the ones who have thought carefully about what they're optimizing for — and it isn't just output volume.

They're optimizing for decision quality. For team retention. For the kind of deep work that AI genuinely cannot do. They're using automation to create the conditions for humans to operate at their highest level — not to squeeze maximum throughput from every working hour.

The goal of automation was never to make you busier. It was to make you more capable of doing work that matters. If your AI implementation isn't doing that yet, the problem isn't the AI — it's the system around it.

This is exactly the work I do. Not just deploying automations — designing the entire operational architecture so that the technology serves the people, not the other way around.

If your team is running harder than before your AI rollout, something is wrong — and it's fixable.