it's not the future until it's boring
5 min read
For a few years now, I've been chipping away at a historical research project about Will Rogers, the early 20th century humorist and one of the most widely read newspaper columnists in America. I find his story and his role in American cultural history fascinating, and as a fellow Oklahoman I feel somewhat obliged to help tell it. He died 91 years ago in 1935, so the vast majority of his life's work has already rolled into the public domain, and the rest will soon. My project's bottleneck has never been a lack of material, but getting access to it in the digital vaults where it is imprisoned. There are many paid-only, private databases that hold incredible, sprawling troves of public domain content but are barely indexed and have interfaces that seem almost designed to punish efficiency. Using them feels like the company is daring you to reverse engineer a better option.
Naturally, I tried. I've spent more time than I care to admit trying to make headless fetch scripts work and I could never get enough of it functioning long enough to be useful. What little of the backend architecture I managed to expose was entirely defensive, explicitly designed to honeypot bots and track their behavior before banning them outright. Cumbersome access beats no access, so I dropped it. The only viable solution is doing the work directly in the browser. Anything else triggers a constant, infuriating war of attrition over cookies and session tokens that standard scripts simply cannot win.
This wasn't really a coding problem, so instead of Claude Code I gave Cowork another spin. Turns out the best way to beat web infrastructure that violently hates automation isn't to write better code. It's to point an AI at a regular browser window and let it do the clicking for you.
Not only is this the dumbest possible solution, it worked incredibly well. With one catch.
Worth every token
For my money tokens, the computer-use version of Claude in Chrome is easily two or three times more capable than handing it the keys to Playwright or other browser automation tools. The catch is it burns through tokens faster than a laserdisc arcade game.
A job I started before bed hit Claude's rolling session limit by morning, on the $200 Max plan, during Anthropic's off-peak 2x promotion. I run Claude Code nearly all day at work and almost never hit that limit but one overnight research Opus thread blew past it. The compute cost is real, but measured against what it replaced, it's probably worth it.
My full project isn't quite ready to launch yet, but one portion of it involves tracking down newspaper columns I believed existed but had never appeared in any of Will Rogers's collected works. I'd already found a few myself, published sporadically in whatever order each newspaper felt like running them, and thought there might be around 25 in total. In one night, Claude found about 50, and it probably hasn't even found all of them.
What's staggering is that Rogers has been studied and written about in dozens of books, and for about a quarter century had a state-funded academic commission established specifically to collect and preserve his works. So how was there a pile of his writing left to gather dust for a century? Mostly, I think, because nobody had the time or the tools to find it. I own and have pored over almost every book published about Will's life and writings, including the entire set published by that commission for the last 50 years. None of these columns I've collected were ever officially listed much less republished. There could have been some complications over copyright, and if this was the case it was never mentioned. For the most part, I believe it's legitimate lost history.
Among what I've gathered are three short columns on the topic of evolution, including Rogers's contemporary take on the famous Scopes monkey trial as that spectacle was unfolding. Rogers was a prolific writer, occasionally to a fault, but I've read everything else he ever wrote and this is genuinely uncharted territory. Whether it holds up is almost beside the point. That it's been sitting there for a century, uncataloged and uncollected is remarkable.
It's not actually surprising that a bunch of 100-year-old newspaper columns were missed, or even potentially excluded by researchers. When the memorial commission started its work 50 years ago these would have been hard to find unless someone had been intentionally collecting them. But for me to go from 0 to 50 overnight is nuts. The human researchers of that era would have needed the time and money to visit dozens of cities and pore over miles of microfilm. They'd have gone blind staring at the glow of the reader, hand-cranking through years of newsprint just to find Will Rogers making wisecracks in tobacco ads. A lot of intentional work with unclear outcomes and high costs. Instead, I typed about 300 words into a text box and a computer in another state spent compute tokens while I was unconscious.
Overall, what Claude did last night has technically been possible for a while, but mostly if you were willing to set up a fragile Rube Goldberg machine of scripts and proxies to make it happen. With Cowork and Claude in Chrome, I just wrote a prompt and let it run. Even three years ago, the idea that you could single-prompt an AI, go to sleep, and wake up to find it had done something that would have taken you months would have seemed insane.
The miracle adjustment period
There are still things about AI that make people shrug and throw their hands up. It hallucinates. It makes things up. If it can't be trusted to be 100 percent right, what's the point?
And then there's stuff like this, where the only honest answer is that you never would have or could have done it without the machine. When an AI agent hands something back that's imperfect but real, the right response isn't "this isn't good enough." It's: holy shit, it did something I never would have done. The output doesn't have to be flawless when the alternative was having nothing at all. Maybe it cleared the real hurdle and you handle the last 20 percent. Maybe it just gets you close enough to learn something or see something you couldn't before. Either way, something exists now that didn't before.
The most interesting thing, though, is the psychological trap this opens up. We adapt to the miraculous incredibly fast. Compressing fifty years of geographical logistics and manual labor into a few hours of compute time will feel like a profound breakthrough for a little while, but a year from now it's just "technology," and the expectation will be that if it takes longer than an hour it's "slow."
AI is not without its very real problems. But it is also turning magic into plumbing, one miracle at a time. Which is, when you think about it, exactly what progress has always done. I'd still like my flying car, but even that would become boring at some point. The future is boring.