SerpApi.Org
You can 10x or 100x or 1000x all you want, but that’s not the productivity boost you’re looking for.
I’m gonna hammer the final nail in the coffin of the AI productivity argument.
This whole myth started to unravel over a year ago, when reports started surfacing that once the actual measurement of AI productivity had been made, the results were—well, let’s say far from spectacular.
How about that oft-quoted MIT study?
“Just 5 percent of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact.”
Thanks, nerds! Emphasis mine because you guys are so non-confrontational.
And I know what some of you are thinking. You’ve never been more productive using AI—it’s like sprinting a marathon on Red Bull and steroids and meth, twice a day, but, you know, with business.
I get it. I do. And I believe that guy. And I’m rooting for him. But his coding of 30 apps in a weekend or his 1,000 personalized sales emails sent in 10 seconds is not why everyone got fired. Well, it is why everyone got fired, but it was never a suitable replacement.
So I’m going to negate the 10x or 100x or 1000x AI productivity argument with two words: Breadth. And scale.
Here’s why we’re not ready for the kind of productivity AI has bequeathed to us.
AI Agents of Chaos
The infamous MIT study of June 2025 will go down as the record scratch heard round the AI world. But that study was, shockingly, explained away: AI was new. Humans were stupid.
And those critics of the AI critics weren’t necessarily wrong. A good percentage of the problems that the enterprise is having with their AI implementations is not AI’s fault. The truth, the quiet part I’m about to say out loud: We’re simply not ready for that much productivity.
I’ll start making my case by quoting the insanely wise Heath Ledger’s Joker.
“Do I really look like a guy with a plan? You know what I am? I’m a dog chasing cars. I wouldn’t know what to do with one if I caught it!”
That’s it. Right there. The enterprise is the dog. AI is the car. Early adopting CEOs are the “psychotic” agents of chaos. No, seriously. I didn’t say it, the CEO of Box did.
The truth as we know it now is, even if the often-ridiculous productivity claims of AI are true, even if it replaces all the customer service agents and all the drive-thru order takers and all the coders and all the middle managers… what we’ve discovered is that the breadth of AI as a replacement for human labor leaves a lot to be desired, and it also breaks at scale
OK. No problem. We can fix that.
Enter the Forward Deployed Engineer
There’s a reason why the role of “Forward Deployed Engineer” is suddenly being touted by the AI makers and the AI stakeholders, like AWS and OpenAI. See, AI implementations failed at that 95 percent rate because no one had one of those.
Fine. I had never heard of that role before 2026—specifically June 2026, which is a couple weeks ago—but whatever. I like the idea. These are the fixers, the cleaners. They are Winston Wolfe, or those guys in John Wick, cleaning up after Wick “sorted out some stuff.”
This is a good start, a step in the right direction, assuming these FDEs are not just glorified project managers or expert ninja guru prompt engineers (Ha! Remember those? I’ve got an “I told you so” post coming for them).
And by that I mean, the FDE is critical and necessary, in most cases, but on Day 0, when an FDE shows up at the door like Ryland Grace banging on the airlock of Blip A—sorry, spoiler alert—they can’t just start spinning up agents and unleashing them on the architecture in search of that mythical 100x productivity.
Because breadth and scale.
We figured this out 15 years ago.
Eff It. We’re Doing 52,000,000x
At Automated Insights, where we released the first commercially available generative AI engine in 2010, one of our first customer wins was Yahoo Fantasy Football, where we created fantasy football matchup recaps overnight after the Monday night game. For Yahoo, this meant 13 million new pages of highly personalized and actionable content every Tuesday, seen by 52 million eyeballs, with all kinds of opportunities to sell sponsorship and ads within.
So 52,000,000x. They were very happy.
But not every customer of ours was set up for any kind of million-eyeball windfall. In some cases, like the Associated Press for their Quarterly Earnings Reports, we took them from 400 companies covered to 4,400, a “mere” 11x.
For others, where the volume was smaller but the payoff higher, maybe we got them to 2x. And I’ll let you in on a little secret. No one was ready for that kind of bump in productivity, whether it was 52 million new eyeballs or a couple hundred.
So we eventually deployed “managed services,” our own version of forward deployed engineers, to be able to harvest the gains made by our nascent but powerful technology. This worked, but it didn’t magically turn a 2x company into a 52,000,000x company
Why? Because breadth and scale.
Every Employee, Every Function, All at Once
We don’t operate in the business multiverse.
Look, I’ll give the enthusiasts the benefit of the doubt. Maybe AI can emulate the rote tasks that make up the repeatable functions of parts of a very junior level employee’s role. Maybe. But it’s going to take more than a wicked prompt or a few lines of Python to make that happen.
And even if you get AI there, it’s stuck, frozen in the context of the time it was initiated. And yes, you can create “memory,” and yes, it can “learn” and “predict.” Of course it can. But seriously, ask any programmer who’s still left—the more complexity you add, the more performance deteriorates.
With AI, it happens slightly at first, then quickly, then all at once.
Getting 100x productivity out of a single function, like 1,000 sales emails, for even a single employee, like a junior developer, does not translate to 100x or 10x or in many cases even 2x for the team or the org or the company context in which those functions sit.
Because those incremental gains have to add value elsewhere.
The JoeBot Is Not Going to Save You
Here’s the simple case I make.
I like to think I write posts that people enjoy and sometimes find helpful. If you RAG me and turn me into a JoeBot, you can only ingest so many of my posts before you hit the max limit of being me, and then you’re also stuck in time from when the JoeBot replaced me, so you don’t get anything new, like my take on Anthropic’s super-dangerous Fable 5. You get the Opus version applied to my Fable take.
Breadth.
But even if you’re cool with this—and I’m not, by the way—I write two or three times a week, and even if you could get JoeBot to crank out a new post every hour on the hour—who is that entertaining and helping?
Is it more people than I could entertain or help on my own? Certainly. But not “once-an-hour” divided by “two-times-a-week” more people (84x). (Sorry for the math.) There’s a limit.
Scale.
And then, even after all that, the earnings on that “84x” productivity deteriorate. Just because JoeBot can write 24 posts a day, every day, doesn’t mean it writes 24-posts-a-day worth of value. Maybe one. Sometimes two. But eventually, it loops into itself, even with new data, because it’s not the words or the headline or even the content of the articles people pay for.
It’s the ideas.
As productive as AI can make a meatbag, it can’t do the meatbag ideas part. That’s not how AI works.
And the productivity increase in ideas generated is the productivity that pays the bills, regardless of how many times AI can photocopy a single idea.
Argument over. Now go out there and use AI to be 1.5x.
Please join my email list to get experienced, 100-percent human-written tech and business guidance peppered with jokes and obscure references.
This post originally appeared at inc.com.
“Click here to subscribe to the Inc. newsletter: inc.com/newsletters"
