Hi {{First Name|friend}}

Asking someone directly about their AI usage is like finding electrons in the quantum world.

The moment we shine a light on AI Productivity with an engagement survey or in a casual 1:1, the beam itself distorts the environment.

Employees feel the need to defend or attack their AI work while subconsciously editing their productivity, adding extra, unprompted polish to it. All along, the leader sees what they want to see.

As a result, they misread the situation and keep pushing their people & systems to the brink.

Until something cracks — either top talent quits or a huge customer trust crisis breaks out.

But there’s another way to measure our AI efforts.

The late famous investor Charlie Munger had a concept called Inversion.

It came from his days as a WW2 military meteorologist, where his job was to map out the safest flight conditions for his pilots.

But Charlie decided to do something different. Instead of tracking all the complex weather variables for a successful trip, he flipped it and asked a crazy question,

“Suppose I want to kill a lot of pilots. What is the easiest way to do it?”

It turns out that there are only 2 sure ways to kill a pilot: Put too much icing on their wings or let them run out of fuel.

So Charlie would make sure they never went through extreme cold weather or too far from a runway.

And by fanatically focusing on avoiding those 2 conditions, all his pilots came back home alive.

Similarly, in a complex organism like a business, it is very hard to tell what AI initiative will move the needle, but predicting what conditions will kill us is easy.

So we need to flip the AI-value question and ask,

What would it look like in our business if our AI initiatives were going really bad?

The answer is 3 conditions for an AI fallout.

Every AI initiative is sending you 2 bills.

The work you redo and the work you check.

In operations, this is called the Cost of Quality - the failure cost & the appraisal one.

The 1st failure cost shows up as re-work of AI slop and it’s avoidable. It is a clumsy end-product that lands on someone else’s desk and becomes their problem to fix. We can lower this cost by putting in better quality controls & filters further upstream.

The 2nd is vetting the quality of the AI work and this is unavoidable. Because the speed of production has increased, your best people, often senior leaders, are now spending a big chunk of their week confirming other people’s work.

But it’s the cost of doing AI business. All we can do is operationalize this new quality vetting step, and make it a more predictable & distributed process.

Ultimately, the trick is to know the difference between these 2 bottleneck bills.

Because if we get it backwards — throwing more bodies at the re-work & killing the quality controls, then we’ll make it much worse.

The 2nd condition that could ruin AI effectiveness is like quiet quitting.

A well-paid, not burnt-out high performer who’s hitting their numbers & looks good on paper might still give their two weeks notice by January.

And there’s a reason we didn’t fully see it coming in our 1:1s or our surveys.

It’s because nobody experiences “engagement”.

Asking about it head-on won’t give us an honest answer as feelings are noisy. People’s low work satisfaction & sense of increased burnout could stem from a whole host of reasons outside AI: a bad manager week or a sick kid at home.

The key is to look for what AI is taking away from them: the ability to do their best work.

With every task or skill we dish out to AI, we risk dimming the lights of our best & brightest.

It’s not only moving them away from their Picasso zones of genius, it’s also making them feel more like an elevator attendant. And the more we rob people of the deep sense of pleasure they derive in doing the things they excel at, the more we threaten their core identity.

So if we see a team’s self-reported time in doing their specialty trend down for a few months, then our alarms should be going off.

Because the thing that made this job their craft is slowly being drained.

An AI-driven crisis is a very real possibility. Asking someone how prepared they feel is useless because they don’t know what they don’t know.

Instead, ask what they can name in 2 ways:

1st: What does success look like?

Having a clear line of sight on a successful end-state helps people measure how much progress they’ve made against it. Without that finish line, employees are working helter-skelter, building and using AI to impress others or at the very least, to avoid feeling left behind.

If people don't know what good looks like, they hide their AI use — and hidden usage means that our AI measurement tool is corrupted. We can’t trust what people are saying anymore.

2nd: How could catastrophe enter?

If employees can’t name specific ways their AI activities could go wrong, they can’t protect against them. Because that means everyone's using it but no one knows the rules or the danger zones. Add to it that many employees have personal unmanaged accounts & we’ve got genuine exposure.

High adoption with low red-line clarity is the most dangerous state.

The clarity on both these questions is absolutely critical.

Without knowing them, we’re at risk of climbing the wrong mountain or going off a cliff.

If we put this together, here’s what this 90 second survey would look like.

Remember, this is an anonymous monthly company pulse, broken down by team and a single month's snapshot tells you almost nothing.

Just like icing or a low fuel gauge by itself didn't kill Munger’s pilots, no single one of these questions in isolation yields anything shocking.

But when they all move in concert together month after month, they reveal a lot.

Each arrow is one month’s movement for that sentiment & it’s going up (↑), down (↓) or holding steady (→). And it’s the persistence of these vectors together over a period of time that gives us the full signal here.

For example, If the AI Re-work score & Vetting score are up while the AI Success Clarity one is still low, then that means either people are still learning or there's an unaddressed structural problem. This is not the time to say the initiative failed.

Here’s a worst case scenario that tells us we’re in danger territory.

If after 3 months, the employee AI Re-work score & the AI Vetting work score are staying high, and people's Best Work sentiments keep dropping, and that’s coupled with a hiring freeze or termination event to offset that increased cost, then there's a good chance of a customer value collapse.

The key is to publish the results each time and show what leadership is working on in response to it. If not, people stop responding to the survey.

Here’s the full instrument to look at:

AI Rollout Pulse & Dashboard

If you want help running it with your team, reply back and I'll walk you through it.