A transformative technology (AI) has been put in our hands for $20 a month – and most knowledge workers are too anxious about its potential to use it.

For any executive looking to unlock AI’s potential, this is a big problem. But for some reason, a lot of execs are still focused on much smaller problems:

• Which LLM will emerge as leader
• When AI innovation will slow so they can avoid re-training employees
• Waiting for AI performance to peak before deployment

The pace of innovation will not slow down for years – if not decades – so these people will be waiting a long time. And the reality is that the performance of LLMs today is good enough to deploy. It’s the people using it that aren’t good enough yet.

You want a successful AI deployment? Focus on what you can control – the people operating it.

Model differences are overstated

It’s no secret that AI has faults. Hugging Face’s hallucination leader board shows that some  models hallucinate nearly 30% of the time. And with newer, faster, smarter models being released all the time, I understand the instinct to wait for something even better to come along.

But as of today, most of the major players in the LLM space are basically on par with each other – with a few pulling slightly ahead (o1 and DeepSeek)

So while some models do better in math skills or response speed, the quality of overall responses is about the same.

The thing that does make a big difference in output is a user's ability to prompt the AI – and that’s universal across models right now. So again, the tech is not your problem.

Why employee performance is the big opportunity

If your team doesn’t know how to prompt, you’ll never get amazing outputs even if model performance peaks.

Here’s the difference in outcome between a lazy prompt and a great prompt – in this example, I’m looking for a list of competitors for an AI-powered coach:

Which of these would you rather your team turn in?

Right now, the state of AI prompting in the workforce is abysmal. According to our research, the average knowledge worker scores 1.9 out of 10 in prompting ability – an F by any reasonable standard. We're handing the vast majority of knowledge workers keys to Ferraris and letting them drive into trees.

Maybe we’ll have agentic AI at some point (we just have the hype now), and we won’t have to worry about prompting anymore. But that future isn’t here yet – so these prompting scores should concern you.

What’s standing in the way of great prompters

Prompting isn’t rocket science – but it does take practice. People aren’t good at it yet for some very fair psychological reasons.

• Anxiety about job loss. Workers don’t want to teach their employers how to outsource their work to AI.
• It’s an act of resistance. Undermining AI’s value keeps workers in a position of safety and sticks it to the cost-cutting execs.
• It feels wrong. Psychologically, we’ve all got deeply entrenched feelings about what makes us valuable. These feelings have been reaffirmed through years of traditional education, and AI challenges them.

If you’re not sure if your employees are freaked out about AI, assume they are. In fact, assume that some are actively sabotaging your deployment because they don’t want to use it.

You don’t just need to get your team proficient in AI use – you need to get them on board with the decision in the first place.

Getting past the friction honestly

To be clear, you’re not going to make everyone at your organization feel warm and fuzzy about using AI. But you can make them feel respected by being honest with them.

If you need a cheat sheet, here’s my talk track about AI as a CEO who is 2 years into building an AI-native team:

• We are going to apply AI more and more at work – both in our daily workflows and also in the products we’re offering.
• Some teams will get bigger because they are that much more impactful with AI. If we can make in-demand teams like engineering, sales, and marketing more productive, we’re going to want more of these highly productive people.
• Some teams are going to stay the same size – they’ll be able to do more work and we won’t need to increase headcount in order to scale the business.
• And yes, some teams are going to get smaller. We’ll need less people on certain teams as AI takes over key tasks. But it won’t be that many people and it’s not my goal.

Putting it all on AI is a mistake

This human friction is a big reason why a lot of people are excited about agents. You build them and they do whole workflows without the need for a human in the loop. It won’t matter if your team is on board with AI or if they are good at prompting or not.

You don’t need a whole complicated change management strategy for agents. And that’s why Big Tech is excited about them, too. Enterprise AI adoption has been slowing as time goes on. Humans are the bottleneck in AI growth.

Yes, at some point agents will come, and we won’t be having these conversations anymore. But that’s not for a good long while. And in the meantime, all this gnashing of teeth about the best AI model is just an attempt to delay dealing with the friction.

Don’t be avoidant. While you’re nickel and diming hallucinations and model performance, other teams are navigating the tradeoffs of AI interactions to get to MBA-level insights faster and more cheaply than you.

Here’s your to-do list:

1. Just commit to an AI model to roll out to your team – it’s hard to pick wrong
2. Make a plan to transparently address the psychological impacts of your deployment (try our AI Manifesto template)
3. Invest in training the skills that make your team good at AI today and worry about re-skilling later