In 2026, Programmers Don’t Write Code Anymore—Then What?
Matt Pocock posted an AI coding timeline on X covering November 2025 to November 2026. After reading it, one feeling stands out: writing code no longer seems to be the problem—but the new problems are harder to deal with.
Matt Pocock wrote a “brief history of AI coding” on X, spanning November 2025 to November 2026. The timeline uses one node per month to record the landmark events and mood shifts in the AI coding world over the year. The core conclusion is distilled in the final sentence: no one is writing code anymore, and we’re still trying to figure out what that means.
Chronologically:
November 2025: Opus 4.5 was released, and OpenClaw was built. Model updates and tooling infrastructure moved forward at the same time.
December 2025: The community largely reached a consensus that models handling 95% of coding tasks was good enough. That number was later cited repeatedly and became the starting point for all subsequent discussion.
January 2026: Ralph loops started to become popular—an early attempt at loops and orchestration. People began thinking about how to get models to run multiple rounds rather than doing single-turn Q&A.
February: Collective fatigue began to emerge. There were too many strategic decisions, and concerns about cognitive debt started to appear. Tools got stronger, but the upfront judgment people had to make became heavier instead.
March: Skills exploded. The community coalesced around easily shareable workflows, packaging experience into reusable modules. Projects like superpowers had already positioned themselves ahead of the curve.
April: Most SOTA developers hadn’t written a single line of code for months. A few months earlier, this state would have been hard to imagine.
May: The “unreasonable effectiveness” of HTML was being discussed, and more people began focusing on richer review interfaces. Code was no longer read line by line; interfaces and visualizations became the new entry point.
June: Loop Engineering took the title of the year’s vaguest buzzword, and hype at the process level became more obvious.
July: Every conference talk turned into “this is my software factory.” Concepts started being wholesaled.
August: Opus 5’s output was hard to understand, and attention to ASD-STE100 (Simplified Technical English standard) skyrocketed. Model capability went up, but output readability became a new problem.
September: Grok Bot took off. Persistent “manager” threads started to become popular; these threads retained context and were responsible for assigning work.
October: The promised speed improvements didn’t materialize, and the vision of the “dark factory”—a fully automated software factory—still felt far away.
November: The anti-slop manifesto was published, and reading code still seemed essential.
After reading this timeline, what’s most interesting isn’t actually which tools or concepts it predicts. What it captures is a turning point in the collective mood. From “the models are good enough” to “no one writes code anymore,” there are months of fatigue and fear in between, and in the end it loops back to “you still have to read code.” After going in a circle, the problem hasn’t disappeared—it has just changed shape.
Of course, this is a fictional timeline, posted in November 2026, with obvious satire. It isn’t writing a technology roadmap; it’s writing about what people doing AI coding were turning over in their minds this year.
Original link: https://x.com/kloss_xyz/status/2122372793513951581
发布时间: 2026-11-17 02:29