The Most Honest AI Slide I've Seen This Year

It's not every day that an AI company tells you where not to use AI.

A few weeks ago, I attended an event hosted by Anthropic.

The talks were good. The technology was impressive. The hands-on experience was exciting. Like every AI event these days, there was plenty of discussion about agents, reasoning, and where the technology is heading next.

Then one slide appeared.

It wasn't the flashiest slide of the day. There were no benchmark charts. No predictions about AGI. No bold claims about replacing entire workforces.

In fact, it looked surprisingly simple.

And that's probably why it stayed with me.

It quietly summarized something many organizations are beginning to realize after three years of AI hype.

Not everything should be delegated to AI. And some things simply shouldn't be.

For the last few years we've heard the same promises everywhere.

• Let AI run your business.
• Replace people with agents.
• Give AI access to everything.
• Fully autonomous companies are just around the corner.

They were compelling ideas. Investors embraced them, vendors built around them and the media amplified them.

Reality, however, has turned out to be more nuanced.

What's particularly interesting is that this message is now coming from the companies building the world's leading AI models. For a long time we asked whether AI could do something.

The more useful question is whether AI should.

The slide breaks the answer into three simple categories.

1. Hand it off

These are the tasks AI already handles remarkably well.

• Drafting emails
• Summarizing documents
• Pulling together information from different sources
• Cleaning and restructuring data
• Running first-pass research

None of these require perfect judgment.

They benefit from speed more than certainty.

This is where AI delivers immediate value.

2. Supervise

This is where many organizations underestimate the risk.

• Customer communications
• Reports that leave the company
• Anything involving numbers
• Multi-step workflows

AI can absolutely help with all of them, but someone still needs to review the output. Not because AI is bad. Because LLMs are probabilistic systems. They generate the most likely answer, not the guaranteed one. As the impact of a decision increases, so does the importance of human review.

3. Keep it

This may be the most important category.

• Final pricing
• Contracts
• Legal decisions
• Hiring and firing
• The financial numbers you'll certify

Every one of these comes down to accountability and accountability cannot be delegated to a language model.

• A person signs the contract.
• A person approves the numbers.
• A person makes the hiring decision.

For the foreseeable future, that person remains human.

There is another lesson hidden inside this slide.

The difficult part of AI is no longer building smarter models.

It's deciding where those models should and shouldn't be trusted.

The organizations that create the most value with AI won't necessarily be the ones deploying the largest models or the highest number of agents. They'll be the ones with clear answers to three questions:

• What should AI own?
• What should humans review?
• What should never leave human hands?

That's not an AI strategy. That's a business strategy.

This is also very close to what we've been discussing with our customers for quite some time.

• AI shouldn't replace business processes. It should strengthen them.
• Don't build your business around AI. Build AI around your business.
• Let AI remove repetitive work.
• Let it surface insights.
• Let it recommend actions.

But don't confuse a convincing answer with an accountable decision.

Those are not the same thing.

I believe the AI industry is entering a more mature phase.

Less attention is being paid to impressive demos and benchmark scores.

More attention is being paid to measurable business outcomes, governance, and responsible delegation.

That's a healthy shift.

Because business value has never really been about asking,

"What can AI do?"

The harder and ultimately more valuable question is:

"What is the right role for AI inside this business?"

To me, that's what moving From AI Hype to Business Reality actually looks like.