6 AI Truths for Leaders (2025)

Hi, folks! In November 2025, I had an opportunity to offer my perspective on AI and its future. Here's the summary slide, followed by some commentary:

1. AI isn't human, Randy!

Over one million users talk to ChatGPT instead of their loved one, sometimes with grim outcomes. Wired reports:

In recent months, a growing number of people have ended up hospitalized, divorced, or dead after having long, intense conversations with ChatGPT. Some of their loved ones allege the chatbot fueled their delusions and paranoia.

Why are humans talking to ChatGPT instead of their loved ones?

South Park nailed the reason: sycophancy. Watch this short clip:

AI sycophancy occurs when chatbots flatter their users to endear themselves. In the clip, Sharon calls Randy out, ”Just because she kisses your ass doesn’t mean you have good ideas!”

AI isn’t human, Randy!

2. We're still in the Wild West phase

I’m old enough to remember when internet access was new. Dozens of companies were vying for market share. They competed by mail. Every week I’d get a CD from AOL, Netscape, Internet Explorer, Earthlink, or some other startup with the promise of 30 days of free (better) internet. It was the Wild West of the Internet. Companies were making it up as they went.

The same thing is happening with AI today. Things are moving fast. There are lots of AI companies, lots of large language models (LLMs), lots of use cases, lots of experimentation, lots of hand waving, but no clear “winner” or business model.

Case in point, Google has rebounded from laughing stock (remember their cringe release of Bard?) to talk of the town with their release of Gemini 3 this week.

What do smart leaders to when things are still frothy?

• Run experiments.
• Measure what’s working.
• Stay up to date.

But fer cryin' out loud, keep your hand on the wheel.

3. When someone says, "AI," ask, "Which one?"

AI isn’t one thing.

Like “transportation” isn’t one thing. It’s cars, buses, trains, airplanes, bikes, sidewalks, and all the infrastructure required to move people from A to B. Each type of transportation requires different types of planning and executing.

Same with AI. AI is a suite of tools including those on the list above.

The question isn’t “Should we use AI?”

The question is “Which AI solves an actual problem we have?”

4. 5% of AI projects work. 95% don't.

MIT reports:

Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95% of organizations are getting zero return. The outcomes are so starkly divided across both buyers (enterprises, mid-market, SMBs) and builders (startups, vendors, consultancies) that we call it the GenAI Divide. Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact. This divide does not seem to be driven by model quality or regulation, but seems to be determined by approach.

So which approach works?

Those rooted in admin and finance. The stuff with structure, rules, repetition, and clear data.

The ones that don’t work are the sales and customer service experiments. The human stuff.

This is good news. It means you don’t need a moonshot to get value from AI.

Start your AI experiments where structure already exists.

5. China may have already won the race

Nvidia CEO Jensen Huang announced at the 2025 AI Summit in London, “China is going to win the AI race.” Data backs him up.

China has:

• A national economic strategy called “AI+” focused on leveraging AI to power their economy. The U.S. does not have an economic strategy, let alone an AI-centered strategy.
• DeepSeek and dozens of other AI companies that are mostly open source and run on less energy-dependent chips.
• Positioned itself as a leader on AI regulation. The U.S. is trying to deregulate AI.
• More published research, PhDs, and influential national institutions focused on AI. At one influential conference, over half of the papers presented were from Chinese think tanks and universities; only 17% of the papers were from American institutions.
• Unmatched scale: with 1.4B people and massive digital platforms, China generates data that improves AI models at a breakneck pace. The U.S. has 330M people, one-fourth China’s population.

This isn’t a reason to panic. It’s a reason to focus.

6. Human + AI is magic

Will AI replace humans in some jobs? Yes, in some. Have you seen an Amazon warehouse?

More likely, AI will replace parts of every job.

Yes, machines can “think,” but they can’t care.

They can analyze, but they can’t empathize.

The can predict, but they can’t decide what matters.

There are thousands of examples where AI has helped humans make better decisions. Here’s one: at a Boston City Council meeting, AI mapped a month of road complaints. The City Council looked at the maps and saw their districts differently. AI produced the data; the Council made decisions.

AI will continue to get better. So humans will have to be increasingly discerning: which AIs are useful, and for what? AI can help us optimize, analyze, evaluate, and even improve. But only humans can say WHY these improvements matter.

What do you think? What AI Truths are true for you? Drop me a line.

Thank-yous!

• Becky Bartoszek and Laura Wiegert and the award-winning Fox Cities Chamber for inviting me to their AI Summit. I learned a lot - and had a ball (see photo above)!
• Ethan Mollick, for always teaching One Useful Thing and your review of Gemini 3.
• Kara Swisher for smart interviews with AI’s living legends on On.