I come across as someone who is quite skeptical of AI (see my post “Oh no, oh dear, what have we done to ourselves?” for that perspective). So I’d like to clarify a little bit.

I’m not skeptical of AI the technology. AI the technology, is just vector databases and algorithms and attention. It has some very interesting potential applications — and some potential applications that would be terrible ideas. Just like any technology.

No, I’m skeptical of AI boosters, the people. I’ve expressed that opinion before, and I remain convinced of it. AI by itself is inert code. AI in the right hands can lead to incredible translation and dictation results, interesting developments in mathematics, and, ideally, the ability to automate away drudgery so people’s live’s become more meaningful and fulfilling.

AI in the wrong hands? AI pushed by uncritical boosters? Well, let’s look at Exhibit A.

Meta’s plan to replace 60% of their staff with AI

Reuters has some excellent reporting around Meta’s (AKA Facebook’s) AI strategy. Their plan (it isn’t really a strategy, but we’ll get to that later) looked like this:

1. Install software on every employees’ computer to track everything they do
2. Use that tracking data to train AI on the jobs of all your employees
3. Mandate that your employees use AI to further train it
4. Eliminate up to 60% of some teams through layoffs
5. You now have an “AI native” workforce composed of a much smaller, “talent-dense” staff

Now, before we get into the specifics of why this plan was doomed to failure, just from a logical point of view, I want to point something out.

Mark Zuckerberg and Co have been AWFUL at predicting the future!

Facebook itself was a copy of a different website — not an attempt to create a future that would be used around the world. And Facebook largely remained on top by ruthlessly buying out competitors or copying them. Again, no prescience required. And in their few attempts to get ahead of the pack? Well …

First up is “Libra” (later rebranded Diem), a worldwide stablecoin that would bring crypto to the unbanked masses.

(Now that I write that out, I realize Libra wasn’t actually that original either)

The fact that you went “Libra?” is indicative of the success of the project.

But the more famous attempt at predicting the future, and one that predates Libra, was the Metaverse, starting with the acquisition of Oculus VR in 2014. Since that time Facebook has spent roughly 80 billion dollars trying to make the “metaverse” happen.

It has not — even with Facebook constantly moving the goal posts of what would constitute success in that endeavor.

Given their history, Meta’s predictions of the future could be viewed as a kind of anti-prophet, confidently stating what will happen, spending billions of dollars trying to MAKE it happen, and then watching as it does not happen.

Likewise, their attempt at rolling out AI in their own company according to how they think companies will run in the future (the tiny “talent-dense” staff overseeing armies of bots) hasn’t gone well.

That’s not to say we won’t ever get to that kind of future, but we’ll get there if and when the world is ready for it. You can’t manifest a sci fi future like the metaverse or AI anymore than you can force a pregnancy to move faster. It takes as long as it takes, and in the mean time, you have to deal with reality as it IS.

And that’s the first mistake.

So let’s get to dissecting Facebook’s failure, and planning your own more successful strategy.

1. Know what’s going on

In Good Strategy/Bad Strategy, Rumelt says:

A great deal of strategy work is trying to figure out what is going on.

Rumelt calls this the “diagnosis” — it is the definition of the problem you face as an organization.

In order to have a solid diagnosis, you have to have a solid understanding of the actual state of reality as it exists.

Almost every major tech figure has come out with some kind of blog post or interview or podcast or manifesto about where AI is leading us — Zuckerberg included.

Notably, almost none of them talk about where we are today. What is AI? What is an LLM? How does it work? What can it do right now? What risks are there? What opportunities?

You cannot possibly make a good strategy if you can’t answer those questions, and most tech leaders don’t bother answering them because they’re way more boring than questions like “Where are we going? What will a job be in 20 years? When will I get my very own Jarvis?”

But if you want to make a strategy you have to do the boring stuff. Sorry.

And the truth of the matter is this: AI (referring mostly to LLMs) is an incredibly impressive technology. Today, this very day, it can do some things, like create routine code, very well. It can somewhat accurately summarize vast quantities of information.

But AI is also unproven and constantly changing. We don’t know how it will effect our own minds, and the minds of our children. We don’t know if “AI Psychosis” is just a blip, or something we need to be genuinely concerned about. We don’t know how much biases from its training data are inserted into unrelated prompt responses.

We don’t know how risky it is to use long term — psychologically, mentally, or technologically. We don’t yet have a good handle on the attack surface it generates. I mean, we still can’t deal with prompt injection — probably the first flaw identified.

We don’t know how much it’s going to cost long term (I mean that in two senses — some organizations blow through their token budget in months (hello Uber), which is one uncertainty, but we also don’t know if the cost will balloon as frontier labs need to demonstrate profitability for an IPO).

To a certain extent, we don’t even know how LLMs work. That’s crazy.

We do know that people are using it. We know that there’s a fear that we’ll get left behind, although studies have shown that productivity gains very widely.

(this is a great spot for a sidebar about one of my favorite details from this whole saga, which was that Meta’s usage of AI coding tools led to an increase in code changes of 220%, which sounds great, until they also point out that “new or upgraded features reaching Meta users” were only up 36 percent, and that there was a 40% increase in major technical or security incidents, and time spent resolving those was up 70%. We don’t know how those numbers pan out because we don’t know the baseline, but it would seem to indicate that a large portion (maybe 75%) of coding time is more-or-less wasted, and that the 36% gain in features could’ve been offset by the 40% increase in problems and 70% increase in time to resolve — see how important it is to have a good idea of what’s going on??)

Can you see how complicated reality really is? And how utterly premature it is to try and force yourself into an “AI Native” workforce by just … firing people?

Even if AI were currently capable of what Zuckerberg believed it could do, though, their strategy was doomed from the start thanks to our second issue.

2. I can’t believe I have to say this: don’t deceive your team

Zuckerberg and Co, well, let’s just quote the article:

On March 13 … Reuters reported that Meta was planning layoffs that could affect 20% or more of its workforce. It wouldn’t have been the first culling on that scale: Between late 2022 and early 2023, Meta slashed around 25% of its staff.

Still, the report caused alarm. Many rank-and-file employees were rattled. And they weren’t yet supposed to know about the cuts, annoying Zuckerberg and leaving executives unprepared for the backlash, according to a person familiar with the matter. A Meta spokesperson at the time called the story “speculative reporting about theoretical approaches.”

Internally, high-level executives hunkered down. They opted not to discuss the article with the rank-and-file and quietly dismissed it to senior managers, instructing them to tell their teams they should expect roles to “evolve” as a result of AI, according to a talking-points document seen by Reuters.

In April, Reuters published more details: Meta was set to cut about 10% of its workforce in a first wave of layoffs on May 20 and was aiming to shed more staff in the second half of the year. Meta soon confirmed the 10% reduction plan to staff, and Zuckerberg later told employees the cuts were due to heavy capital expenditures.

SO, Meta put software on everyone’s computers specifically to train their AI replacement. When called on this incredibly obvious ploy, they denied it and lied about it. When forced to admit it, instead of saying the truth (that they’d been planning layoffs to help their company become AI First) they blamed it on Capital Expenditures — which, come on. They burned 80 billion dollars over the course of a decade to try and make “Snow Crash” a reality, and now they’re feeling fiscally responsible?

People aren’t dumb. They see what’s coming, and if they know that AI is going to lead to layoffs, how are they going to react to it?

I don’t need to spell this one out because it’s so obvious. I mean, not obvious enough for the executive team at Meta to understand, but I’m pretty sure you understand how people will react when told to train the person (or bot) who will take their job and potentially make it so they never get a job again.

A strategy that requires that you deceive your own team is doomed to failure, not just for the logical reasons we listed above in this specific instance, but also because the purpose of strategy is to get everyone rowing in the same direction. How can strategy “coordinate action to address a specific challenge” (Rumelt, again) if people don’t know what the strategy is? As he points out “Strategic actions that are not coherent are either in conflict with one another or taken in pursuit of unrelated challenges.”

In other words, the purpose of a strategy is to coordinate the actions of your team. If some people don’t know the strategy (because you are deceiving them as to what the strategy is), they’ll be working at cross purposes with the rest of your team. At best it will be inefficient, at worst, self-destructive.

3. You can’t cut your way to a “talent dense” team

As a reminder, Facebook’s goal was to deploy software to everyone that was clearly meant to eventually replace them (which they lied about but uh … they haven’t shown themselves to be particularly trustworthy in the past). They didn’t tell anyone what was going on, but people put it together.

Now, if your goal is to keep the most talented team members, and you send a big red flashing sign to everyone that they’re getting replaced, two things are likely true:

1. Your most talented team members are going to figure it out very quickly (duh)
2. Your most talented team members will find other work where they won’t be replaced

Not only that, but studies have shown that layoffs lead to an increase in voluntary departures after the layoff, especially among high performers, and that when one high performer leaves, others tend to follow.

In other words, they sent a very clear message to their most talented people to leave. Then they were planning on firing people, which would inevitably lead to MORE departures, especially among high performers, who would leave as a group because when high performers leave, they take other high performers with them.

(and I want to take a second to point out that it’s been my experience that almost anyone can be a high performer, if they’re put in the right situation with the right support, so all this “fire low performers” talk should probably actually be “train our managers better” talk, but that’s a different blog post)

What are you left with? Well, a workforce made up of the people you haven’t already fired who couldn’t find a different job. Are they going to feel secure about their position? Are they going to put out good work? Are they going to be enthusiastic about AI, which was just used to fire all their friends and hangs ever over their head?

Probably not.

Again, I wouldn’t call this a strategy, but even as just a plan it’s awful.

So what is a good strategy for deploying AI?

Well, we’ve already talked about what a good strategy isn’t, but let’s sum up:

1. A good strategy engages with reality as it is and diagnosis a clear problem. A bad strategy is built on what people hope reality will become, and it isn’t really engaging with any clear problem.
2. A good strategy brings your team together to work towards a common goal in a common way. A bad strategy is a secret that most of your team doesn’t even know about.
3. A good strategy is something everyone can get behind — management, employees and customers. A bad strategy has to be kept a secret because it only serves a subset of those engaged (in Facebook’s case, just management, really. It can’t even be said to serve their customers)

So here’s the steps to creating your own AI strategy

Step One: Engage with Reality and Solve a Problem

First, you have to understand what AI is, what it’s capable of, and what the risks are.

This takes time and effort — it’s not something that you can read an article (or a blog post, sorry) and feel like you’re prepared. Because the risks of AI are very different not just between models, but in different industries and circumstances as well.

So first, come to a good understanding of AI.

Second, as Rumelt points out, good strategy addresses a problem by drawing on sources of advantage. AI is a tool — it is a how, not a what. Your AI strategy can’t be “Use AI more.” What problem are you solving with that? The problem of AI adoption? Why are you adopting AI?

I was having a chat with a friend who works in a highly regulated field and we were discussing the layers of bureaucracy they have to work through to get anything done. That is a problem — how do we handle the red tape better and more consistently than our competitors? And AI, with the right deployment, is a decent solution to that problem.

But if you don’t know what the problem is, AI is just a solution looking for a problem.

Step Two: Bring people together working towards a common goal

A good strategy is inspirational. It makes people want to engage with the problem it defines, and the method of engaging (outlined in your “guiding policies” — you should read Rumelt’s book to learn about that) feels workable to people.

If Meta’s strategy had been “create the next facebook that will connect the people of the rising generations in a way that brightens their lives and improves mental health” (or even “fix our own stuff so it doesn’t hurt people’s mental health”) then your employees would’ve been engaged. And if they’d then said “We’re going to do this using AI because it allows us to comb through so much more information and find patterns we’d otherwise missed” then their employees would’ve gone “OK, yeah, I see the utility there. I want Facebook to not cause young girls to have eating disorders! I’ll see if I can’t make this AI thing work!”

Step Three: Your strategy has to serve everyone

In order to get everyone rowing in the same direction, everyone has the know and buy in to the strategy. And for everyone to know and buy in to the strategy, they must all feel well served by it.

That doesn’t mean the strategy includes raises for your employees — in fact, if your strategy says “we’ll have to pay our people a lot more to keep them” (which Facebook’s strategy admitted) then you’re already in trouble.

Employees, or team members in any organization, are often there for two reasons:

1. They need a paycheck (of course)
2. There’s something about the organization that they connect with

People like to find the nobility in what they do — even if it’s a stretch. They want to feel like they’re doing something meaningful with their time.

A good strategy engages that instinct by helping everyone understand what you do, why you do it, and why it matters.

Some people doubtless work for facebook because of the large paychecks, and they don’t care about anything else. However, a lot of people are there because they like the idea of connecting other people — of doing something that has a real world impact. Maybe they even found their significant other through Facebook! Or they found connection in a niche Facebook group when they were feeling alone.

Whatever the case, some people will stick around if you keep filling their moneybags, and some people stick around because they want to feel like they’re doing something important.

BUT! And this is the important part.

BUT, if you don’t know what problem you’re solving with AI then you can’t possibly know why you’re deploying it. And if you’re not sure why you’re deploying it, then the public default rationale for deploying AI will rule — people will assume that you’re deploying it for the same reason other companies have said they’re deploying it, and for the same reason the sellers claim it will destroy the economy: to replace human beings with bots.

If you can’t connect the dots for your employees about how AI will make their jobs better, and their customers live’s or products better, then they will connect the dots for you and assume (as they rightly assumed at facebook) that you’re deploying it in order to fire them.

Just be trustworthy

I read a management book once that talked about how you need to “build a personal brand of integrity” and talked about all these things that people with integrity do to signify their integrity to others so that people reading the book could copy those outward displays in order to build their personal brand.

And as I read it I wanted to shout “OR JUST HAVE INTEGRITY!” You don’t need to build a personal brand, you just need to be a person who has integrity.

AI is a cool tool, but it’s just a tool. And yet so many people are losing their jobs, or just sleeping poorly at night, because of all the hype around what is ultimately a shiny new tool and nothing more. It’s not AGI. It’s not the singularity. It is a very powerful tool.

If leaders at companies like facebook were just … people with integrity, they wouldn’t be hiding their plans to fire their people. They wouldn’t be trying to fire people at all! They would be finding new ways to use this very cool tool to solve problems for their customers and their team mates.

Instead they are firing people by the thousand and causing untold amounts of stress and heartache to thousands, tens of thousands, or even millions more (all those people who see what’s happening at Facebook and go “yeah, probably going to happen to me to”).

If they were just people with integrity they wouldn’t be causing all that stress and heartache and pain. They would be working with their team mates to get through what will end up being a pretty turbulent time together.