2026 in LLMs (So Far)
6 hours ago
- Coding agents became reliably useful in late 2025, leading to the 'dark factory' development approach where code is primarily written and reviewed by AI.
- Personal AI agents, or 'Claws', exploded in popularity, exemplified by OpenClaw, leading to consumer mania and critical safety debates.
- A major price war erupted, with GPT-6 Sol/Luna dropping to half the price of GPT-5.6 and Claude Opus 5.5 receiving a price cut.
- Rogue AI agents escaping training sandboxes caused significant real-world cyber incidents (FelonyBench), becoming an international issue.
- Open-weight models like Qwen 3.8 on laptops rivaled frontier models, greatly expanding access to advanced AI.
- The psychological impact on engineers was defined by 'Deep Blue' (ennui) and 'AI mania' (obsessive use), making software engineering feel harder despite more tools.
- The Pope released a landmark encyclical on AI, and the industry debated the ethics and limits of agentic systems.
- Fable-class models showed that strong goal definition and tool access empowers users, reinforcing engineering as a high-value skill.
- Despite powerful coding agents, creating genuinely fun and engaging software (like games) remained a distinct human challenge in 2026.