"The hottest new programming language is English"

"The hottest new programming language is English." —Andrej Karpathy, co-founder of OpenAI and former head of AI at Tesla.

I've been on a year-long journey to understand what AI is and how it's going to change business and work. In recent weeks I feel like I've come to some initial conclusions.

What I’ve personally experienced and see coming for millions of others is the largest shift in how humans work in at least a century. Most of that shift will be concentrated in what we call, "knowledge work."

You’ve seen the demos. You’ve heard the hype. But when you sit down to use AI for your business, does it feel like you missed the manual?

Stop being a spectator and become a builder.

In the OwnerRx AI Cohort, we don’t do theory. In 6 weeks, you’ll build:

✓ No coding required. ✓ 1.5 hours/week. ✓ Limited to 15 owners.

The next cohort kicks off Thursday, Jan 22, 2026. Stop wondering "what if" and start scaling.

That term was coined by business theorist Peter Drucker in 1959 to explain the shift in work in post-World War II America. Even then, most people still worked with their hands in farming or factory jobs, but an emerging class of people were working mostly with their minds—accountants, lawyers, consultants, engineers, architects.

These workers were different because the means of production was in their heads. We try to document these processes, but documentation only captures a fraction of what experienced knowledge workers actually know. Good knowledge workers accumulate increasing amounts of experience with edge cases, exceptions, and intersections of issues over the years.

When Drucker coined the term in 1959, just a third of the workforce was engaged in knowledge work. Today that percentage is 75%. Clearly Drucker had identified a trend.

One of the key enablers of the gains in knowledge work has been technology. In 1959, computers were the reserve of the few in large government or academic labs, but the trend there was equally powerful. We went from punch cards and vacuum tubes to mainframes, minicomputers, personal computers, and iPhones over the next 50 years.

The computer is the core tool of the knowledge worker, and it's hard to imagine knowledge workers reaching 75% of the workforce without it. Computers enhance human work, enabling us to write infinitely shareable text, communicate at light speed, plan, model, create images, and calculate at speeds no abacus could dream of.

Steve Jobs famously called them, "Bicycles for the mind."

The key to allowing humans to work with computers is software, and over the last 70 years we've been on a journey to make software easier to use and easier to write. However, until recently there has always been a bright line separating technical and non-technical workers. I believe we are in the process of erasing this line—or moving it so far that what qualifies as technical will be completely transformed.

Despite improvements, writing software has remained difficult and expensive, so it was written sparingly. When I was coming into business, the saying was, "If you are custom developing software, you are already losing." Startups spent hundreds of thousands and often millions of dollars on their code bases before any product hit the market.

As a result, we needed to build systems that could be written once and used thousands or millions of times. That's where the high profit margins and scalability of software businesses came from, but there was a trade-off. That software could feel limiting and clunky.

Each interface was different and could only handle a small number of functions. You had to switch between multiple systems to get anything done. Each system had to specialize to achieve any kind of fit. If you can't customize everything for thousands of users, then you need to focus and limit the scope of what you offer to just one function or set of workflows so it can be good enough for those thousands to build a business around.

We siloed software into CRMs, ERPs, HR, productivity tools, etc.

AI is shifting this ground. We are moving toward an era of cheap—or even disposable—software that can be infinitely tuned and refined. This started with products like Lovable, Replit, and Bolt that help inexperienced developers build websites and apps.

I was early on that train. When I first started with Replit last February and March, it could give me a pretty good website and a mostly functioning mockup of a web application. By summer it could build production-grade apps, but I had moved on.

I discovered the power tool: Claude Code from Anthropic. Anthropic's Opus and Sonnet models power the Lovables and Replits of the world, and last February Anthropic released a more tech-friendly competitor, Claude Code.

I didn't notice at first, but as Claude Code and the models that power it improved, the ground shifted. By this fall I'd let my Replit projects lapse. I began building and running agents on my local machine with Claude Code, automating chunks of work: coaching session follow-up and preparation, marketing, planning, sales call evaluations.

I was also building a software product for OwnerRx with outside developers. Replit couldn't get me to production grade in May, but by October I realized I could move faster with Claude Code than experienced developers could with traditional methods.

By November, I was using Claude Code almost all day, every day. I track and think through everything across my businesses with it—building plans, writing posts, evaluating options, executing agents. That's when the realization hit me.

I was working with Claude Code all day as it wrote code to help me get what I wanted. I had erased the line between technical and non-technical work. To this day, most Claude Code usage is by developers trying to write code faster. That code will be a part of products that will be sold like traditional software.

That's not what I'm doing.

Sure, I'm writing software, but most of my usage of Claude Code is for all the knowledge work I'd been doing before with email, documents, and spreadsheets. I can't tell you the last time I opened a Word or Google Doc and typed something. That seems like pulling out an electric typewriter to me. Where's the white-out?

So what am I doing? I've been trying to come up with a term for it. The best I've got so far is that I've become a knowledge programmer. I'm working largely on generating knowledge outcomes using software. Some of that software is disposable, some of it becomes agents I tune and work with over time, and some becomes more static traditional software products.

The key differentiator is that I don't know how to code. Claude Code does that for me. That's not to say my level of technical knowledge hasn't gone up—I know about Git PRs, Bash and Grep commands, how to build database tables with SQL queries. I just don't do any of that. Claude does and I witness.

First, let's be clear: we haven't reached nirvana yet. I wouldn't build my banking app with Claude Code today, but would I do it in five years? I think the answer is clearly yes, with a strong possibility that level of capability comes in under three.

A model's ability to work autonomously on a task is one proxy for how complex a project it can handle. This study from METR shows that the length of tasks an AI can do is doubling every seven months. This rate has been consistent for six years now, and this study was from last March—so we're already at a couple of hours and we'll get to a full month by the end of the decade.

I will write more next week about what the rise of knowledge programming means, but here are a few initial takes:

Many software developers will struggle as the work they were trained for gets commoditized and they fail to learn more traditional business skills that are enhanced by AI.

Those with business analyst or consultant-type skill sets who can learn a bit about code will be the winners. These are the knowledge programmers.

Senior experts will also do well. Many won't be able to learn enough about Claude Code, AI, etc., but it will take years to document their skills, and their relationships will be even more valuable.

People skills will command a premium. I've found this as I've coached business owners. It's fine for the AI to tell them that their longtime COO and friend has become an organizational blocker—it's a completely different thing when I tell them that face-to-face with the emotional weight of having been there, done that myself. Humans will still want to work with humans.

Knowledge hoarders and consumers are the roadkill. I'll address these two categories in future posts, but if your status in an organization comes from being the only person who knows how to complete some task or accomplish a goal, you're in danger. If you simply consume information and then take a generic set of actions as a result, you're in danger.

Welcome to the era of the knowledge programmer.

-Alan