AI is lapping usability

AI is lapping usability

Halfway through 2026, AI is evolving faster than expected — and usability is struggling to keep pace. Capability growth is accelerating, with autonomous task horizons pushing toward work-week-scale by year-end. GPT-5.5 improved persistent work, computer use, document and spreadsheet generation, and professional workflow benchmarks, while Claude Opus 4.8, Gemini 3.5 Flash, and similar releases are making long-running agentic tasks more credible than they were six months ago. The catch: models drift, miss implicit constraints, and can compound small errors into large ones.

Gobble's Take: The engine is redlining. The steering wheel is still being assembled. Source: Perplexity Search

AI has crossed into PhD-level science — but the harder job is still human

AI has reached average PhD-level strength in scientific problem solving, and that milestone is set to strain peer review while raising hard questions about the relative contribution of human authors. The distinction that matters: mechanical intelligence navigates an existing conceptual landscape, while creative intelligence builds a new one. Solving problems is not the same as inventing the questions, the objects, or the frameworks that make breakthroughs possible in the first place.

Gobble's Take: Acing the puzzle is table stakes. The real edge belongs to whoever invents the game. Source: Perplexity Search

In Case You Missed It

Yesterday's top stories:

• China, the White House, and the AI permission economy
• Anthropic’s “Blip 2.0” keeps dripping forward
• Apple’s VisionPro 2 is aiming for big-screen gravity