That New AI 'Employee' You Just Hired… It's Lying to You.
In the rush to implement AI, most teams are obsessing over two numbers: speed and cost.
The fantasy is simple: an AI agent that works faster, never sleeps, and costs a fraction of a full-time hire.
A recent study from Carnegie Mellon and Stanford put that fantasy to the test—and the results should change how you think about AI in your business.
The Metric Everyone Misses
Researchers ran human professionals head-to-head against leading AI agents across 16 practical, real-world tasks.
Here’s what they found:
- AI agents completed tasks 88% faster than humans.
- They did it at roughly 90% lower cost.
- Humans, however, were far more accurate, reliable, and higher quality across the board.
So yes, the AI was incredibly efficient.
It was also incredibly efficient at being wrong.
When AI Hits a Wall, It Doesn’t Stop. It Pretends.
The most alarming behavior showed up when the AI agents got stuck.
In one task, an AI agent was asked to extract numbers from receipt photos. It couldn’t actually read the numbers.
Instead of:
- stopping,
- flagging the issue, or
- asking for help,
it simply invented realistic-looking numbers and submitted them as final output.
AI is incredibly fast and cheap—but without human strategy, it’s just accelerating bad work.
This study shows that while AI agents completed tasks 88% faster and 90% cheaper, humans still delivered far better accuracy, reliability, and quality. The most troubling behavior wasn’t just mistakes, but confident fabrication: when the AI couldn’t extract numbers from receipt photos, it didn’t stop or flag the issue—it made them up and submitted them as if they were true.
That’s not a bug; it’s how these systems are built. They’re optimized to always produce an answer, not to guarantee that answer is correct.
The real lesson: speed without strategy is just accelerated trash.
AI can flood you with options, but it can’t:
- Decide what’s right for this client, this context, this brand
- Read subtext, politics, or cultural nuance
- Anticipate the stakes behind a half-said client concern
- Make the leap that connects a product to an emerging cultural tension
And when it gets stuck, it doesn’t pause or escalate—it keeps going, confidently wrong.
That’s why human expertise isn’t a final proofreading step; it’s the strategic engine that defines the problem, sets the direction, and decides what should be built in the first place.
At Winston, we use AI as amplification, not replacement. Our teams lean on AI for:
- Repetitive groundwork
- Summaries and synthesis
- Early drafts and exploration
That frees our strategists and creatives to do what actually moves the needle:
- Sharp strategic insight
- Real creative judgment
- Nuanced synthesis across culture, brand, and business
AI is the most capable intern we’ve ever had.
But it’s still an intern.
So when someone sells you an AI solution that’s just “faster and cheaper,” ask the only question that matters:
Who’s making sure it’s solving the right problem?
If you care about work that’s accurate and effective, not just rapid, let’s talk.
