Why AI Disappoints At Productivity - But Excels At Ambition

Best AI users don't spend less time on their projects, they spend more. We thought we would get a speed machine. We got a depth machine instead.

I generated the first draft of this article in ten minutes.

Then I spent five hours reading, cutting, restructuring, asking another AI for opinion, re-writing, finetuning the title.

And fact checking.

—exactly the opposite of what the brochures promise.

Whether in coding, finance, or content creation, the pattern persists: lightning-fast draft generation followed by painstaking verification and adjustments.

We thought we were getting a time-saving productivity device.

We got something entirely different. And we’ve been setting the wrong expectation.

What Can You Do With Gen AI ?

Would you use a nail gun that occasionally shoots your thumb?

A GPS that takes you to the wrong city once per trip?

A calculator that randomly multiplies instead of adds?

What do you do with a tool that’s brilliant 90% of the time and confidently wrong the other 10%?

The challenge with Gen AI is those invisible mistakes, wrapped in perfect grammar and unshakeable confidence. If it was wrong very often, you’d naturally ignore Gen AI altogether. Like asking a creative but unreliable friend for advice—entertaining, occasionally insightful, never trusted blindly.

But today’s AI occupies an uncanny valley of reliability. Too accurate to ignore, too fallible to trust.

The Expert’s Prescription (That Gets You Nowhere)

Visit any AI productivity forum, and you’ll find the same advice, repeated like a mantra:

Use AI for the right tasks.

After eighteen months of collective experimentation, the community has converged on three “safe zones”:

Zone 1: Creative Exploration
Brainstorming. Ideation. Blue-sky thinking. Where “wrong” might spark something interesting.

Zone 2: Drafting
First passes. Rough outlines. Initial research. Meeting notes. Where errors are cheap to fix and perfection isn’t the goal.

Zone 3: Good-Enough Work
Routine updates. Form emails. Where 90% accuracy is genuinely sufficient.

“Stay in these zones,” the experts promise, “and AI becomes your superpower.”

The logic seems unassailable.

This is exactly where our problems begin.

The Productivity Trap - When Faster Just Burns You Out

So you follow the expert playbook. You delegate every low-stakes task to AI. Email responses. Meeting summaries. First drafts. Status updates.

The AI handles it magnificently. Tasks that took hours now take minutes. Your output explodes. You’re a productivity machine.

Then something unsettling happens.

Oliver Burkeman documented this phenomenon in Four Thousand Weeks, before ChatGPT existed. He calls it the efficiency trap: the better you get at clearing your plate, the more the world piles on it.[1]

Email demonstrates this perfectly. People who achieve “Inbox Zero” actually spend more time on emails than the others. Why? Faster responses trigger more messages. It’s conversational physics—send the ball back quickly, it returns just as fast.

AI supercharges this dynamic. That podcast backlog? AI transcribes and summarizes it—now you have thirty must-read summaries. That newsletter pile? AI digests them all—now you’re drowning in digests. Those meeting recordings? AI extracts action items—now you have three times as many tasks to track.

You’re not saving time.

You’re operating at AI velocity while your capacity remains stubbornly human.

Even when AI makes us faster, the finish line moves. Microsoft’s 2025 Work Trend Index shows that productivity gains are real—but over half of AI-leading firms say the result is taking on more work, not less. [2]

You’ve optimized yourself into a hamster wheel spinning at AI speed.

The Selection Strategy: Surely This Works

“Fine,” you think. “I’ll be strategic.”

No more using AI for everything. Just for work that genuinely matters. The deep projects. The meaningful challenges. The stuff you actually care about.

This way, you dodge the volume trap. You’re not drowning in shallow tasks. You’re deploying AI surgically for maximum impact.

Finally, a framework that makes sense.

Except it doesn’t work either.

The Expertise Paradox

When you use AI for work that matters, you encounter a perfect catch-22 perfectly identified by Daria Cupareanu [5] :

Scenario A: The Learning Curve

You’re exploring unfamiliar territory. Building new capabilities. Venturing beyond your expertise.

The AI seems invaluable. It explains concepts, provides frameworks, generates examples. You’re moving fast, feeling powerful.

But here’s the trap: you can’t see the 10% that’s wrong.

Carnegie Mellon University research found that AI chatbots remain overconfident even when they’re wrong - unlike humans who adjust their confidence after seeing poor results, AI models “tended, if anything, to get more overconfident, even when they didn’t do so well” [3] The very expertise you’re trying to build gets contaminated at the source.

So you slow down. Cross-reference everything. And suddenly, you’re spending more time fact-checking the AI than you would have spent learning it properly the first time.

Scenario B: The Expert’s Curse

Experts don’t benefit either in terms of time saving.

Imagine, you know this domain intimately. Years of experience have developed your judgment, refined your taste, established your standards.

Now you see every flaw in the AI’s output. Each interaction triggers a correction cascade:

Generate → Review → Identify errors → Refine prompt → Regenerate → Find new errors → Manually merge versions → Realize you’re now middle-managing a robot → Wonder why this feels like more work

That might explain why a July 2025 RCT found experienced developers took ~19% longer with AI vs without AI (quality held steady) .[4]

The paradox is perfectly symmetric:

• Know too little? You can’t spot the errors fast enough

• Know too much? You can’t fix them fast enough

Either way, you’re not saving time.

We Bought the Wrong Machine

We thought we were buying a time-saving device. We got something completely different.

AI doesn’t save time. It trades time for capability.

You can now attempt work that was previously impossible. Explore territories previously inaccessible.

But it costs MORE time, not less.

This follows a familiar pattern. Consider the parallel from photography. When digital cameras eliminated film costs and developing time, did photographers spend less time on pictures ? No—they took thousands more pictures, spent hours editing, created work previously impossible. The technology didn’t save time; it transformed what was possible.

The Depth Machine Protocol

First, abandon the time-saving fantasy entirely.

Stop measuring productivity by speed. Start measuring by depth. Not “How much did I produce?” but “How good was what I produced?”

When you stop trying to save time with AI, you’re augmenting intelligence, not replacing effort.

Second, conduct The Ambition Audit.

List three things you’ve always wanted to create but couldn’t—a novel, a research paper, a complex analysis, a new product. Pick one. This is where AI earns its keep: making the impossible merely difficult.

Budget 10x the time you think it needs. It’s the price of reaching beyond your previous limits.

Third, embrace The Iteration Investment.

Stop counting drafts as waste. Budget for multiple iterations on anything that matters:

• Early drafts explore the territory

• Middle drafts reveal what you actually want to say

• Cross-check with another AI to correct for bias

• Apply human judgment and manual rewrites

• The real work emerges

Like a sculptor, each iteration excavates excellence.

Fourth, recognize The Human Handoff.

After 3-4 iterations, AI stops adding value and starts adding friction. This is your signal to take full ownership. Your taste and judgment should dominate. The project becomes yours, not the AI’s.

Finally, accept the rest is exploration, not productivity.

Most AI use isn’t about efficiency—it’s intellectual play. And that’s perfectly fine.

The Bottom Line

GPT-5 won’t save you time. Neither will GPT-6, or whatever comes next.

If you’re using AI to go faster, you’re using it wrong. If you’re drowning in AI-accelerated tasks, you’ve fallen into the trap. If efficiency is your goal, you’ve missed the point entirely. The 90%-accurate robot is here to stay. It’s a capability amplifier that demands more time, not less.

Use AI to go deeper into fewer things.

Use AI to attempt the impossible.

To create the previously unimaginable.

Just don’t expect to save time doing it.

---

If your organization is measuring AI adoption — prompts used, content generated, tools deployed — it's tracking the productivity trap, not the depth machine. Same paradox, different scale. Read my latest article.