# Coding Faster at a Standstill *February 2026* I've been rereading Charlie Munger's talk on [Elementary Worldly Wisdom](https://fs.blog/great-talks/a-lesson-on-worldly-wisdom/). Since coding agents hit an inflection point in late 2025, his framework maps onto what's happening in the software industry with uncomfortable precision. What really matters when code becomes a commodity? What differentiates those that prosper from those that merely survive? >[!quote] [Charlie Munger](https://fs.blog/great-talks/a-lesson-on-worldly-wisdom/) >*The great lesson in microeconomics is to discriminate between when technology is going to help you and when it's going to kill you.* ## Thesis Every software engineer I know works wider across the stack and ships more code than they did a year ago. AI is [intensifying the work itself](https://hbr.org/2026/02/ai-doesnt-reduce-work-it-intensifies-it) even as it makes each task faster. This is the paradox: you can code faster and still not be better off. Creating value is not the same as capturing it. - Value creation is adopting technology that increases your output. Shipping more code, reaching more users, building faster - Value capture is converting that output into something durable and exclusive. Assets, leverage, position. Firms do this by [capturing the productivity of employees](https://every.to/p/are-you-playing-a-career-game-you-actually-want-to-win) into company output. Individuals do it by converting work into career leverage When everyone gets faster at producing the same thing, the gains don't accrue to producers. They flow on to users or accumulate to the owners of platforms and complements. ## Why creation does not imply capture > [!quote] [Charlie Munger](https://fs.blog/great-talks/a-lesson-on-worldly-wisdom/) > *When we were in the textile business, which is a terrible commodity business, we were making low-end textiles—which are a real commodity product. And one day, the people came to Warren and said, "They've invented a new loom that we think will do twice as much work as our old ones."* > > *And Warren said, "Gee, I hope this doesn't work because if it does, I'm going to close the mill." And he meant it.* > > *What was he thinking? He was thinking, "It's a lousy business. We're earning substandard returns and keeping it open just to be nice to the elderly workers. But we're not going to put huge amounts of new capital into a lousy business."* > > *And he knew that the huge productivity increases that would come from a better machine introduced into the production of a commodity product would all go to the benefit of the buyers of the textiles. Nothing was going to stick to our ribs as owners.* > > *That's such an obvious concept—that there are all kinds of wonderful new inventions that give you nothing as owners except the opportunity to spend a lot more money in a business that's still going to be lousy. The money still won't come to you. All of the advantages from great improvements are going to flow through to the customers.* Equating value creation with value capture is one of the most dangerous mistakes a firm can make. Technology can increase a firm's productivity and outputs while having little to no effect on the bottom line. From the outside: prices fall, input costs rise, complements extract rent, and imitators compress away margins. This is because of a few primary mechanisms: - **Demand substitutability** - If buyers can easily switch to alternatives, firms face stronger competitive pressure to pass cost reductions through - **Diffusion and imitation** - If rivals can copy or imitate improvements quickly, even dramatic technical progress will diffuse into an industry-wide productivity shift and the firm will struggle to convert the innovation into a rent stream - **Complements** - Technology isn't used in isolation. Even when technology is hard to copy, value capture can be redirected to whoever controls bottlenecks in complementary assets The textile mill dramatises all three mechanisms at once. Textiles are a commodity (substitutability). The new loom's gains would diffuse industry-wide (imitation). And the surplus flows to buyers or is extracted by complements. [Lucking, Bloom, & Van Reenen (2019)](https://www.nber.org/papers/w24622) found the same pattern in R&D broadly: social returns exceed private returns by a factor of four. When improvements are widely adopted, the value created by innovation flows overwhelmingly to everyone except the innovator. But some businesses thrive despite industry-wide technology adoption. The ones that capture value hold persistent advantages in areas complementary to the technology itself. The classic examples are network effects, switching costs, and economies of scale. Apple can afford to not adopt all the latest technology in their products because they have such positional advantages in key vertical integrations and ecosystem bottlenecks. Coca-Cola didn't invent refrigeration or modern distribution, but when those technologies made cold drinks ubiquitous, their brand and franchise bottling network captured a disproportionate share of the value. Value capture, at its core, comes from owning complementary assets that are bottlenecked in rivalrous ways. But even a strong position can erode if the bottleneck is controlled by someone else. ## What this means for software engineers For software engineers who define themselves by writing code, all three mechanisms are in full flight. Code production has near-perfect substitutability. AI will diffuse across the industry faster than any prior technology. And the labs have every incentive to accelerate this - more demand for software means more demand for tokens. The complements that matter (platforms, distribution, AI infrastructure) are controlled by others. ![[matrix.png]] Most software engineers today sit in the bottom-right of that matrix: code is easy to imitate and the complements are controlled by someone else. If your productivity doubled with AI but your salary didn't, that's value creation without capture. The gains went somewhere, just not to you. Cheaper code will create greater demand for software, but that growing pie only benefits those who can capture their share. Demand growth without capture just means coding faster but with $200 a month in tokens as the participation cost. ## This is still a great time to do something about it Diffusion is still uneven. Not every engineer has adapted, not every firm has reorganised, and the complementary skills that matter are nowhere near commoditised. >[!quote] [Charlie Munger](https://fs.blog/great-talks/a-lesson-on-worldly-wisdom/) >*... when you're an early bird, there's a model that I call "surfing" - when a surfer gets up and catches the wave and just stays there, he can go a long, long time. But if he gets off the wave, he becomes mired in shallows ...* During the lag of adoption, early adopters can earn temporary rent. But staying on the wave requires converting that temporary advantage into a durable position, and that conversion isn't free. [Brynjolfsson et al. (2021)](https://www.aeaweb.org/articles?id=10.1257/mac.20180386) showed that general-purpose technologies produce a productivity J-curve: large upfront investments in complementary intangibles precede measurable gains. The J-curve is itself a kind of standstill - you invest and see no returns, which is precisely why most people under-invest and fall off the wave. The habits, networks, domain depth, and taste that compound over time all require sustained investment during a period where the payoff is invisible. ![[j-curve.png]] ## Three durable areas So what are these complements and areas? Some people call it taste, talk, architecture. I think the better frame is that the entire value chain from code to solving a human problem is littered with bottlenecks, technical and nontechnical, that limit the efficacy of all generated code, any of which become immensely more valuable as code gets cheaper. These are some of the most durable: - **Whether you can bridge what's human and what's digital** - Software solves human problems, and bridging the gap between a human need and a well-specified technical solution requires empathy with users, staying grounded in the problem, and the ability to get people to take action. The engineer who can empathise with a frustrated non-expert user and reframe the problem captures value that no amount of generated code can substitute - **Your efficiency in converting tokens into software** - It's bewildering to think that we now have a mechanism to convert sand to silicon, silicon to intelligence, and intelligence to software. But the last piece is by no means a solved problem. It is especially [jagged](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4573321), meaning knowing where AI excels and how to combat failure modes requires tacit judgment that compounds with experience. The engineer who knows how to create novel feedback loops and verification ships better software from the same tools - **And finally reputation** - Code is losing its use as a signal, which makes reputation, demonstrated judgment, and track record even more important. Credibility is inherently rivalrous: yours is yours, it compounds over time, and it can't be imitated. The person with a track record of solving hard problems in visible ways has an advantage that widens with every cycle These are the bottlenecks that become more valuable as AI diffuses, which is precisely Munger's test for when technology is going to help you rather than kill you. The engineers who step back from the [LLM slot machine](https://arxiv.org/abs/2509.22818) to invest in these things won't immediately look as productive. But that is precisely the investment that stops you from coding faster at a standstill.