2x, not 10x: coding with LLMs in 2026 (obryant.dev)

In a recent reflection on the state of large language models (LLMs) in software development, the author shares their insights on the productivity gains of using LLMs by 2026. They highlight that while LLMs have become significantly more reliable for coding tasks, allowing for iterative feedback and clearer goals, the anticipated productivity boost may be capped at around 2x rather than the previously hoped-for 10x. The author emphasizes that while LLMs can effectively execute specific coding tasks and meet defined acceptance criteria, they still struggle with broader questions regarding code maintainability and documentation quality. The significance of this observation lies in the understanding that future improvements in LLM capabilities may not directly translate into proportional boosts in developer productivity. Instead, the focus should shift toward refining workflows and tools that leverage the current abilities of LLMs, rather than expecting them to perform at a higher level. By advancing industry practices around LLMs, developers may find innovative ways to overcome current limitations, suggesting that substantial gains can be achieved through better integration and retooling rather than solely on model enhancements.