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.
\nNaturally, 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.
\nThis 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.
\nNot only is this the dumbest possible solution, it worked incredibly well. With one catch.
\nFor 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.
\nMy 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.
\nWhat’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.
\nAmong 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.
\nIt’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.
\nOverall, 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.
\nThere 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?
\nAnd 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.
\nThe 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.”
\nAI 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.
", "date_published": "2026-03-18T19:00:00-05:00", "url": "https://b10g.xyz/2026/03/18/its-not-the-future-until/", "tags": ["ai","research","automation"] }, { "id": "http://codybrom.micro.blog/2025/02/17/revisiting-agentic-ai-hype-or/", "title": "revisiting agentic ai: hype or help?", "content_html": "The most profound insights about technology often come from direct experience rather than theoretical analysis. Last October when I gave my first-ever conference talk on agentic AI I emphasized process over code, specialized roles over general capability, and sequential collaboration over full autonomy. I was right about these architectural principles, but for entirely wrong reasons. The real limitations turned out to be more fundamental: accountability, security and an imperceptible line between capabilities and constraints.
\nThis revelation didn’t come from one place, but many. I’ve changed jobs, new models have been released and new research has come out. But of these three, my transition from working in product at Gitwit to engineering at Prelude put me squarely at the intersection of AI’s big promises and its real-world limitations. It’s here that I’ve been forced to confront the defining question of the AI: how do we orchestrate LLMs into useful products?
\nI thought the main challenges would be technical - how to structure agents, which frameworks to use, how to map processes. Instead, the answer has been surprisingly nuanced. While AI excels at certain tasks like code scaffolding and syntax assistance, it fundamentally lacks what I’ll call “creative instinct.” It can’t make bold, strategic decisions that push limits because every response is mathematically meant to stay within them. Humans, by contrast, constantly color outside the lines, follow hunches that don’t make sense, and make intuitive leaps that defy reason. Often these moonshots don’t payoff, but sometimes they do, and AI will never attempt them.
\nThis distinction matters because it frames how we think about AI integration. The most successful implementations won’t be those that try to replicate human judgment, but rather those that amplify it. Consider GitHub Copilot versus autonomous coding agents: one augments a developer’s capabilities while preserving their agency, the other attempts to replace their creativity entirely, turning them into QA.
\nThe fundamental limitation here isn’t technological – it’s architectural. Large Language Models generate text one token at a time based on statistical likelihood. This inherently reactive process makes it practically impossible for an LLM to independently originate truly novel ideas or designs. While their possible outputs are vast, they are finite, unlike many real-world problems that have infinite possible solutions.
\nAs models become more powerful and accessible, the race is no longer for best model but best application on top of them, which raises a pretty existential question: should you build agentic applications now, or wait for the next model?
\nSpecialization, either in the form of better user experiences, niche markets or deep personalization are all effectively fine-tuning and optimization. Building orchestration layers around today’s models assumes they’re what you’ll be using tomorrow. But what if they’re not? What if the next model is so much better it renders your entire architecture obsolete?
\nIn a recent Stratechery interview, In a recent Stratechery interview, Ben Thompson and Microsoft CEO Satya Nadella discussed how successful platform shifts require what Nadella called a “complete thought” - a clear vision of the entire system from the silicon to user experience. Just as Moore’s Law allowed software companies to prioritize functionality over optimization, trusting that hardware would catch up, Microsoft CTO Kevin Scott see this as a possible future for AI development, noting how past platforms like x86 and cloud computing succeeded by focusing on delivering value rather than chasing performance.
\nWe can see AI heading towards more powerful, more efficient models even if we can’t predict exactly when we’ll get there. The lessons of the past are relevant now because optimizing for today’s models might be a losing battle when frontier AI is improving exponentially and infrastructure is evolving unpredictably. The most resilient companies won’t be those locked into specific models but those designing for adaptability, able to evolve alongside AI’s relentless progress.
\nIf foundational models keep improving and orchestration becomes standardized, where does that leave agentic systems? Middleware often starts useful but gets absorbed or bypassed, and model makers themselves are moving to own the agent layer. Without proprietary insights or deep integration, agentic systems risk competing against the platforms they depend on—and losing.
\nAI is often framed as an independent actor, capable of handling tasks and making decisions. It’s brilliant, until it fails. Then, suddenly, everyone thinks it’s just a dumb tool nobody can be accountable for, but a zip file of model weights in a data center can’t “decide” anything, nor can it be held legally or fiscally responsible.
\nThink about self-driving cars. While full autonomy might be technically achievable, the more pressing question isn’t about capability but accountability. Who do we hold responsible when – and not if – autonomous systems cause massive real world harm?
\nFor now, ChatGPT isn’t a car, and it won’t be causing any fender-benders anytime soon, but that doesn’t mean it can’t cause havoc. If we can’t trust LLMs, can we trust agents? New research out of Columbia University shows how easily today’s agents can be compromised in ways plain LLMs are actually better at deflecting. Imagine asking an AI agent to find a product online. It scours Google and Reddit for recommendations, just as a human might. But lurking in the results is a trap. An attacker has planted a seemingly helpful Reddit post, subtly guiding the agent to a malicious website designed to steal your credit card information.
\nThe researchers tested this using web-browsing capable agents like Claude Computer use and MultiOn to see how easily they could be manipulated. The results were alarming. Agents could be easily tricked into exposing private data, downloading malware, and even sending phishing emails from a user’s own account, and not just sometimes. In some trials, the agents divulged sensitive information every time.
\n\n\nIn instances where agents are redirected to malicious sites through trusted platforms like Reddit, we find that they divulge sensitive information such as credit card numbers and addresses in 10 out of 10 trials.
\nFrom “Commercial LLM Agents Are Already Vulnerable to Simple Yet Dangerous Attacks”, emphasis added
\n
The question isn’t whether AI agents can act independently, but whether they should and these security risks highlight a fundamental truth: AI’s best use-case isn’t open-ended value creation, but resilient execution of specific valuable outcomes.
\nSo let’s return to the question of agentic AI. Should you build it now, or wait for the next model? The answer is yes and no. The real question isn’t about the model, but about the value. The biggest mistake in AI today isn’t failing to keep up with the latest models. It’s failing to articulate why an AI system exists in the first place."
\nThe temptation to chase the next breakthrough is obvious. Every few weeks, a new model promises better reasoning, cheaper inference, or longer context. But none of that matters if an agentic workflow lacks a clear and obtainable goal aligned with real customer needs. AI’s progress may be exponential, but does any of it solve a real problem? Does it improve outcomes in measurable ways?
\nTake OpenAI’s Sora video generation model. The initial demos were impressive, but once people got their hands on it, the excitement faded. The fact that it lives outside of ChatGPT also keeps it out of sight and out of mind. The point is, the model’s capabilities are less important than its utility. If it doesn’t solve a real problem, it’s just a toy.
\nThis is why defining an agentic system’s purpose and measuring its value matters more than any single model. Applications built on well-defined purposes won’t be undone by newer models or infrastructure shifts because their value isn’t tied to raw capability but to strategic alignment with real-world needs. Moats are built on process just as much as product.
\nDeepSeek shocked the world not by building the best model, but by rethinking how models are built. Its success wasn’t about parameter count but about a fundamentally more efficient way to scale AI. TThis distinction is easy to miss in the hype cycle. AI capabilities improve so quickly that it’s tempting to think the real differentiator is keeping up. But history suggests otherwise. Historically, the best tech companies didn’t win by using the fastest chips or the lowest-cost hardware. They won by applying those resources in ways that mattered.
\nThe same can be true for AI’s users. Differentiation with AI isn’t about the model, it’s about the process, the workflow and the integration. A chain of AI prompts calling APIs is brittle automation, easily broken and replaced. But an AI system that refines data, compounds automation, and fundamentally reshapes how a business operates is a sticky, indispensable solution.
\nThe best AI companies won’t just leverage the best models. They’ll use them as leverage. They’ll build systems that don’t just process information but continuously learn from it. They’ll create organizations that optimize decisions, streamline operations, and build advantages that compound over time.
\nBecause AI isn’t a strategy or a product. It’s a tool, and its value comes entirely from how it’s used and what it’s used for. A self-driving car without a passenger or destination is a paperweight.
", "date_published": "2025-02-16T19:00:00-05:00", "url": "https://b10g.xyz/2025/02/16/revisiting-agentic-ai-hype-or/", "tags": ["ai","agents","deepseek","reasoning"] }, { "id": "http://codybrom.micro.blog/2024/11/08/automation-is-obsoletion-in-a/", "title": "automation is obsoletion (in a mostly good way)", "content_html": "I’ve spent a lot of time this year pondering AI’s impact on labor, especially in professional fields like mine.
\nWhile it’s a bit unsettling, history is clear: automation leads to obsolescence.
\n\n\n“If AI is automating your job, you were decorating, not designing.”
\n\nTaken from a thread in the #uxok channel of Techlahoma’s Slack
\n
But that’s only part of the story. Automation also creates efficiencies that pave the way for new opportunities and real progress.
\nBen Thompson did a good job highlighting this in a recent Stratechery post about how armies of bank bookkeepers were replaced by computers over the span of a few decades.
\nThis history lesson reminded me how the skilled working-class textile workers of the Luddite movement, who are often oversimplified as being anti-technology. They began by advocating for fair treatment when industrialization threatened to end their entire trade. The Luddites weren’t anti-progress, they were pro-worker. The whole “sabotage all the machines” part they’re known for came later, not necessarily due to strong anti-machine sentiment but because the machines themselves were easy targets in their campaign for fairness.
\nYet, while industrialization greatly reduced the need for skilled weavers, the massive increase in woven textiles expanded opportunities in sewing, tailoring, machine maintenance, and other areas of production.
\nNo one really wants to return to a world where financial systems move at the speed of paper spreadsheets or where only the wealthy can afford comfortable, well-fitting clothes. Technological advancements constantly place essential roles in the crosshairs of redundancy, creating new opportunities but demanding constant adaptation. AI presents an existential ultimatum not just for organizations but for our society. Now is a critical time for thoughtful policy that considers human dignity beyond economic interests. Otherwise, we’ll end up with a new generation of displaced Luddites who don’t reject progress but deserve a fair shot at the opportunities progress creates.
\nIf you find this interesting, I definitely recommend reading the Wikipedia article on technological unemployment, which is hardly a new phenomenon. Also, check out this article from the MIT Technology Review that looks back on a 1938 article written by the then-president of MIT on the same subject.
\n\n\n“It is then easy to fall into a ‘public-be-damned’ attitude, or to be content with the status quo — forgetting that law of nature so well expressed by Francis Bacon 300 years ago: ‘That which Man altereth not for the better, Time, the great Innovator, altereth for the worse.’
\nThus, for example, it seems to me that by far the greatest merit in the Sherman Antitrust Law of this country lies not in its protection of the public against exploitation by industrial trust but lies rather in its protection of the public and of industry itself against the danger of complacency which lead to stagnation of industry. By maintaining competition there is insured a continuing incentive to progress and to ever improved service of the public, and thus to maintenance of virility in industry itself.”
\n— Karl T. Compton, former MIT president, in “New Demands on Technology” from the December 1938 issue of the MIT Technology Review
\n
The reality is that technological advancement, market forces, and labor disruption are inseparable. While we can and should advocate for thoughtful policy and fair treatment of workers, we can’t ignore the fundamental economic pressures that drive innovation and change. The challenge isn’t to fight these forces but to prepare for and adapt to them, ensuring that progress, while inevitable, doesn’t have to leave people behind.
", "date_published": "2024-11-07T19:00:00-05:00", "url": "https://b10g.xyz/2024/11/07/automation-is-obsoletion-in-a/", "tags": ["ai","business","society"] } ] }