2026 in LLMs (so far)

In November 2025, two significant models were released: Claude Opus 4.5 and GPT-5.1. These models represented incremental improvements over their predecessors, but they introduced a notable development – their coding agents started working reliably. Previously, coding agents had been inconsistent, often making mistakes. However, these new models improved the coding agents to the point where they could be used effectively on a daily basis.

Throughout the year, various projects and discussions centered around coding agents and their security. One notable project was an obscure GitHub repository called "Warelay," which gained significant attention. Another project involved taking on numerous new projects in an effort to explore the capabilities of coding agents.

The year also saw debates about the limits of LLMs and their potential for writing good code. Some predictions included the belief that LLMs were already writing good code, and the goal of solving agent security issues. Predictions about agent security included the possibility of a "Challenger disaster," but no major incidents had occurred by this point.

Additionally, a term coined during the year referred to the feeling of "AI-induced ennui," where software engineers felt listless due to the vast capabilities of AI. This phenomenon was discussed by several speakers at the conference.

In January, the author experienced "AI mania," a state where they felt compelled to constantly push their agents to build more projects, often at the expense of their well-being. This led to the creation of a JavaScript interpreter entirely in Python and a WebAssembly runtime in Python. These projects, while somewhat exaggerated, helped the author understand the capabilities and limitations of AI in software development.