About Work Writing

Loading the 3D scene.

tag

#ai

Four Lenses on Explanation→An eval can show that an AI agent failed. Four practical lenses can help explain why it failed, what the score missed, and what to fix next.2026 · 09 · 025 minaiagentsevalsexplanationagentic engineering Being instrumentalistic→If an oracle can give us every correct answer, is there still a reason to understand anything? AI makes an old question about knowledge feel newly personal.2026 · 08 · 315 minphilosophyaiknowledgelearningreflection Agent Loop Engineering with Claude Code & Codex→AI made writing code fast. The harder problem now is everything around it — deciding what to build, keeping agents on track, reviewing the result, and earning enough confidence to ship.2026 · 08 · 046 minaiagentsclaude codecodexagentic engineeringdeveloper tools Take a Step Back→Flow gets all the press. But your brain has a second mode, and the hardest problems I've solved were never solved inside the tunnel — they were solved the moment I climbed out of it.2026 · 07 · 316 minfocusflowproblem-solvingaiengineeringlearningphilosophy Make Claude Speak→I stare at Claude Code ten hours a day. First I stopped typing to it. Now I've stopped reading it too — voicemode makes Claude talk back, fully local, and this is the short version of how.2026 · 07 · 302 minclaude codevoicedeveloper toolsaiworkflow What and Why of explain-it→An agent skill that turns a review, a PR, or a subsystem into a small static HTML explainer site — a primer that builds the mental model first, then one page per topic with runtime stories, trimmed diffs, and quantified stakes.2026 · 07 · 253 minagent skillsclaude codecode reviewdeveloper toolsai The Making of Tars 3.0→How we are rebuilding Tars from first principles — combining ten years of CX experience with a secure AI-native conversation engine, native live chat, multichannel infrastructure, and evaluations.2026 · 07 · 2112 mintarsaiconversational aicustomer experienceengineeringsecurity Principles of Agentic Engineering→Everything I've learned about building AI agents in the past few years, distilled into nine principles — the loop, tools, context, memory, sub-agents, evals, observability, and guardrails.2026 · 07 · 1611 minaiagentsagentic engineeringllmsoftware engineering Why manual reviews are still relevant in the age of AI→We are generating more code than ever and reviewing less of it than ever. In this article, I share why offloading code reviews entirely to AI is a trap, what it silently costs us, and the workflow I believe in.2026 · 07 · 156 minaicode reviewsoftware engineeringdeveloper productivity

i write to think — on simplicity, deep work, and the craft of building software. if a piece lands, the door’s open.

P.S. subscribe via rss, reach me at [email protected], or head back home.

Ankit Vashisht · 2026アンキット