AIs can now often do massive easy-to-verify SWE tasks and I've updated towards shorter timelines
I've recently updated towards substantially shorter AI timelines and much faster progress in some areas. [1] The largest updates I've made are (1) an almost 2x higher probability of full AI R&D automation by EOY 2028 (I'm now a bit below 30% [2] while I was previously expecting around 15% ; my guesses are pretty reflectively unstable) and (2) I expect much stronger short-term performance on massive and pretty difficult but easy-and-cheap-to-verify software engineering (SWE) tasks that don't require that much novel ideation [3] . For instance, I expect that by EOY 2026, AIs will have a 50%-reliability [4] time horizon of years to decades on reasonably difficult easy-and-cheap-to-verify SWE tasks that don't require much ideation (while the high reliability—for instance, 90%—time horizon will be much lower, more like hours or days than months, though this will be very sensitive to the task distribution). In this post, I'll explain why I've made these updates, what I now expect, and implications of this update. I'll refer to "Easy-and-cheap-to-verify SWE tasks" as ES tasks and to "ES tasks that don't require much ideation (as in, don't require 'new' ideas)" as ESNI tasks for brevity. Here are the main drivers of my update: Opus 4.5 and Codex 5.2 were both significantly above my expectations (on both benchmarks and other sources of information). This isn't that much of an update by itself, we should expect some variation and some models to be decently large jumps, but then Opus 4.6 (and probably Codex 5.3 and
In recent months, there has been a significant shift in the expectations surrounding artificial intelligence (AI) development timelines and capabilities. This update is driven by several factors, including advancements in AI models and their performance on specific tasks. The most notable changes involve a substantial increase in the probability of full AI research and development (R&D) automation by the end of 2028, as well as improved short-term performance on massive, difficult yet easy-to-verify software engineering (SWE) tasks that do not require much novel ideation.
The primary driver behind these updates is the performance of AI models such as Opus 4.5 and Codex 5.2, which exceeded initial expectations. While some variation in model performance is expected, the consistent above-expectation results from Opus 4.6 and the anticipated performance of Codex 5.3 and 5.4 have significantly influenced the revised timelines. In 2025, there was an observed doubling of capabilities every three and a half months on the METR 50%-reliability time horizon, with an additional jump at the beginning of 2026, albeit with some uncertainty.
These advancements have been particularly evident in the AI's ability to accomplish large and impressive software engineering tasks with only moderately sophisticated scaffolding. Tasks that would typically take humans months or even years to complete have been achieved by AI in a fraction of the time. This capability is not limited to simple or straightforward tasks but extends to more complex, difficult SWE tasks that are easy to verify.
The implications of these updates are significant. By the end of 2026, it is expected that AI will have a 50%-reliability time horizon of years to decades for reasonably difficult easy-to-verify SWE tasks that do not require much ideation. However, achieving high reliability, such as 90%, is projected to take significantly less time—hours or days rather than months. This progress will be highly sensitive to the distribution of tasks, with some areas seeing faster advancements than others.
The shift towards shorter timelines and faster progress in specific areas of AI development underscores the rapid pace of technological change. As AI models continue to improve and their capabilities expand, the potential for automation in research and development processes becomes increasingly realistic. This not only accelerates the pace of innovation but also has implications for industries reliant on software engineering, as AI becomes a more integral part of the development process.
In conclusion, the recent updates to AI timelines and capabilities are a result of impressive model performance and rapid advancements in software engineering tasks. With AI demonstrating the ability to handle complex, difficult tasks with ease, the future of automation in R&D and SWE is looking increasingly promising. As these capabilities continue to grow, the integration of AI into various industries is poised to transform the way we approach research, development, and problem-solving.
