AI-Wiki

Addy Osmani

3 min read

Addy Osmani

Confidence 0.85 · 4 sources · last confirmed 2026-07-09

Software engineer at Google working on the Chrome team and a prolific web-developer advocate. Blog at addyosmani.com; books on web performance and JavaScript patterns. From 2025 onward, his blog and O’Reilly Radar contributions have shifted substantially toward AI-assisted software engineering and agentic engineering as a discipline.

Osmani’s wiki anchor is harness engineering as a named discipline — his Agent Harness Engineering article on O’Reilly Radar (15 May 2026) is the wiki’s first article-altitude treatment of harness engineering as a deliberate practice, covering: the Ralph Loop pattern, [[AGENTS.md]] convention, hooks, sandboxes, compaction, and harness-as-a-service. The article positions harness engineering as the scaffolding-around-the-model-as-real-artifact-that-tightens-every-time-the-agent-slips discipline — convergent with Chatterjee’s Anatomy of an Agent Harness (Osmani credits Viv as the originator of the harness engineering term).

Paul Everitt at AI Dev 26 SF (one week later) cites Osmani as an influence: “Addy Osmani wrote a blog post about Agentic Engineering. And as I tell you these blah-blah wrote a blah-blah in your mind you should be thinking it’s entering the hype cycle. I should put this on my LinkedIn as a global recognized expert.” The self-deprecating framing notwithstanding, Everitt’s nine-element agentic-engineering practice taxonomy includes harness engineering as a discrete element — Osmani’s article is a direct conceptual predecessor.

Appears in this wiki via

  • 2026-03-26-osmani-code-agent-orchestra-multi-agent-coding — The Code Agent Orchestra, write-up of his O’Reilly AI CodeCon talk (26 Mar 2026) — chronologically his earliest wiki source, predating the Agent Harness Engineering article below by seven weeks. Explicitly builds Steve Yegge’s eight-levels framework into concrete patterns: Subagents vs. Agent Teams vs. a 3-tier orchestration-tool landscape, the factory model (Plan/Spawn/Monitor/Verify/Integrate/Retro), the Ralph Loop, and a citable empirical claim (Gloaguen et al., ETH Zurich) that LLM-written AGENTS.md files underperform human-curated ones.
  • 2026-05-15-osmani-agent-harness-engineering — first-party article: Agent Harness Engineering on O’Reilly Radar; the wiki’s first article-altitude treatment of harness engineering as a named practitioner discipline.
  • 2026-05-22-everitt-jetbrains-deeplearningai-ai-dev-26-sf-shift-to-agentic-engineering — cited by Everitt as influence on the nine-element agentic-engineering practice taxonomy.
  • 2026-06-17-vo-how-i-ai-ai-agent-loops-claude-code-codex — Claire Vo cites Osmani’s loop engineering article as the best breakdown of what an effective loop needs, reproducing its five-element checklist (work trees, skills, plugins/connectors, subagents, state). Confirms Osmani’s loop/harness engineering writing as a reference practitioners reach for at the product-leader altitude, not only the engineer altitude.

Open questions

  • Osmani’s pre-2025 web-performance + JavaScript-patterns writing — substantial body of work but pre-AI-agent context. Worth flagging if a wiki concern surfaces around web-performance + AI-coding-agent intersection (the AI-generated front-end performance regression pattern).
  • Osmani’s specific Ralph Loop origin — Everitt mentions Osmani citing Viv as the originator of harness engineering; the precise attribution chain (Ralph Loop → AGENTS.md → harness engineering) is partly traced in the wiki via 2026-05-07-chatterjee-anatomy-of-agent-harness but worth tracking further.

Graph View

Backlinks