The end of optimization

Some thoughts on a transition point in recent internet history,

Here’s an observation that, I suspect, has ceased to be true:

There was a time period in recent internet history — call it the era of Big Data, or the platform era — when the large digital platforms (Google, Facebook, Amazon, Twitter, Netflix) focused on optimization. The platforms had an immutable comparative advantage over their potential competitors. They had more data, more user engagement. They leaned on all that data and activity to refine and improve their products.

(To be clear, the platforms still have more data and more user engagement than any potential competitors today. It’s just that they don’t bother to improve their products anymore. Optimization is so very ten-years-ago.)

This was one of the themes in my friend and GWU colleague Matt Hindman’s 2018 book, The Internet Trap. It is good to be an Internet King. Google’s search results were better, and delivered faster, than its competitors. Netflix constantly fiddled with its recommendation algorithm, introducing customers to their next favorite show. Amazon could tell, based on your purchase and search history, what products to show you next. None of these services were perfect, but all of them were better-than-the-competition. Data optimization was a race to the top. The big platforms had a self-reinforcing advantage. And they took that challenge seriously.

These same observations were baked into my 2016 book, Analytic Activism. I was studying digital advocacy groups, not tech platforms. But the fundamental behavioral assumptions were all there — the bigger your list, the better you could use data analytics to engage in “digital listening.” The challenge for activists lay in thinking carefully about theories of change, and making sure you measured and optimized for the right things.)

I do not think we were wrong about these observations, back then. The research was solid. Matt’s book won awards, and rightly so. But what I have now come to recognize is that the focus on optimization was a time-limited social fact. Platform executives and their senior managers believed optimization was important, and they built internal reward structures that rendered it true. But this only lasted until they decided to discard it.

From the vantage point of 2025, optimization is clearly no longer a priority for the tech platforms. Google’s search results have gotten worse. Google doesn’t care. Facebook is awash in AI slop. It welcomes the slop. Amazon is filled with fake products and fake reviews. All of these companies still dominate their categories. Degrading the user experience isn’t costing them. The motivating belief that these companies had to optimize, or else they would be out-competed, no longer drives Silicon Valley behavior. Optimization was an era. That era has ended.

Let me jump on Cory Doctorow’s bandwagon here (his new book is really good, btw). We can go ahead and label the current internet era as the Enshittification era.

Doctorow treats enshittification as a process, rather than a time period. And he makes a strong argument. But optimization is a process too. So if the Internet of the ‘10s was characterized by the platforms’ shared commitment to optimization, then the Internet of the early ‘20s has been defined by the platforms’ shared embrace of enshittification.

History is complicated. Plenty of factors combined to pave the way for the enshittification age. My back-of-the-envelope list includes, at a minimum, (1) skyrocketing post-COVID tech valuations, (2) blockchain bullshit hastening the financialization of everything, (3) the surge of gambling-on-everything apps, (4) the bulldozing of the regulatory state, and (4) the tech billionaires having their precious feelings hurt by a workforce that sought to make collective demands and a public that didn’t applaud loud enough.

But also, it really kind of is Elon’s fault.

Elon Musk purchased Twitter in October 2022. He didn’t have some grand plan. (Hell, he spent the summer trying to get out of buying the thing.) But he did have a working theory: the company was spending too much on delivering a high-quality product. You could fire most of the workforce, shut down the servers, turn off the peripheral services, and everything would still be *fine, basically*.

The counterarguments ranged from “lol it might all crash and burn” (It did, but only temporarily) to “everything will keep getting marginally worse and people will leave and oh also you’ll face millions in regulatory penalties.” (bingo)

The core of Elon’s hypothesis was that optimization was overrated. So what if the website loads slower, and the ads are less well-targeted? Elon saw opimization as a very expensive moat, and he quite publicly announced to his tech-billionaire peers that it was unnecessary. You can degrade the product. Your users will have a hard time going anywhere.

I publicly predicted Twitter would go bankrupt. I was wrong. But the reason I was wrong was that I expected the administrative state and the courts to move faster than they actually did. (I thought the FTC fines and the employment lawsuits would take six months. They took years. And Trump shut down all the FTC investigations as a favor to Elon, his largest investor.)

Instead of bankruptcy, X stayed upright long enough for Elon to sell it to sell it TO HIMSELF in an all-stock acquisition. And this matters quite a bit, because in the context of the tech billionaire groupchats, it proved Elon to be functionally right. Twitter/X is obviously worse now than when he acquired it, but not in ways that matter to the owners. The lesson for his peers was “Elon nailed it. We’ve been spending unnecessarily on making our products better. We don’t need to do that anymore.”

Eras rise and eras fall. I can’t say how or when the enshittification era will conclude, but it sure seems like it cannot last indefinitely. My hunch is that it ends with the collapse of the AI bubble. Much like the post-dotcom bubble years, a number of industry-defining companies will be bulldozed once the financial gimmicks fall apart.

I have trouble envisioning what comes after it. Too much depends on the how and the when of the crash (including whether David Sacks just bails all his buddies out). Just this week, a group of techies published the Resonant Computing Manifesto, which strikes me as a gesture toward what well-meaning people with skills will hopefully set out to build in the aftermath. I like the idea, but can’t quite wrap my head around the reality quite yet.

So, for the moment, I’ll leave this here as a partially-formed idea: Optimization was an era. It did not last. It was replaced by enshittification, as an era. It too, cannot last forever.

Here’s hoping the next era is something better.