LINKEDIN_POST2026-07
Why I still check the code, but stopped reading every line
LLMs rarely make syntactic mistakes anymore — when something breaks, it's architectural. Two things changed how I review: asking the agent what could break elsewhere before merging, and keeping a running spec of past decisions it checks against new requests.
read on linkedin →LINKEDIN_POST2026-07
57.5% of web traffic is bots now — is your site agent-ready?
Cloudflare's numbers show agentic traffic growing faster than any other automated category. If a site isn't readable by agents (markdown content, llms.txt, a proper sitemap), it risks being invisible to both agents and the humans using them.
read on linkedin →LINKEDIN_POST2026-06
I gave my AI agent a job I kept putting off
An agent now reads my Japan travel blog's RSS feed, tracks what's been covered, and drafts new posts in Markdown pushed straight to a GitHub repo. The real bottleneck was never the writing — it was picking the topic and deciding to start.
read on linkedin →LINKEDIN_POST2026-06
Cutting token costs when agents run all day
A few habits that actually moved the needle: a new chat per phase of work, compacting context before it fills up, cheap models for binary checks, and trimming long logs to just the last 20-30 lines.
read on linkedin →