AI agents - autonomy - 2026

The gap between what agents can technically do and what products safely let them do is still much wider than most people think. That gap feels more like product design than model capability.

Source: repeated usage across coding, research, and workflow automation tools.

software engineering - AI usage - 2026

Software remains the default proving ground for agentic systems because the feedback loops are short, the interfaces are textual, and success is easier to validate than in most knowledge work.

Source: practical observation from current AI developer tooling adoption.

learning - retrieval - research

People often confuse exposure with understanding. Retrieval under some amount of pressure still feels like the cleanest dividing line between familiarity and actual usable knowledge.

Source: synthesis from learning science and product design work.

writing - clarity - 2026

Most drafts improve less from adding one more paragraph than from deleting the sentence that was protecting the writer from saying the real thing directly.

Source: personal editorial pattern from writing in public.