Interview: Dexter Horthy (Founder, HumanLayer) on Research‑Plan‑Implement and AI Code Quality

Description

HumanLayer founder Dexter Horthy shares tools and practices for making coding agents effective on hard problems. He outlines a research‑plan‑implement workflow, why senior engineers must set AI coding standards, and how code review should emphasize mental alignment. Dexter also discusses Claude Skills and the AI Tinkerers community. CHAPTERS: 0:03 - Intro: HumanLayer and coding agents for hard problems 0:18 - Tools: OSS prompts and an IDE for many agent sessions 0:29 - Parallel coding agents; research‑plan‑implement 0:43 - Beyond small tasks: toward 99% AI‑assisted code 1:34 - Adoption rift: senior skeptics vs junior acceleration 2:41 - Need top‑down standards to avoid “slop” 3:04 - Standard setting in the AI era 3:49 - Code review hierarchy; style is least important 4:25 - Code review for mental alignment, not box‑checking 5:34 - Real example: huge Golang PRs require plans 6:18 - Research phase: objective codebase understanding 6:41 - Implementation plan: steps, tests, phasing 6:59 - Check deviations and tests 7:22 - The human layer: natural‑language docs and strategy 8:10 - Why big companies may adopt AI coding faster 8:31 - Expert knowledge of models and instructions 8:43 - Proliferate best practices via prompts/workflows 9:15 - Claude Skills vs slash commands/subagents 10:05 - Skills flexibility; avoid context forking 11:31 - AI Tinkerers: builders learning from builders 12:14 - Join: ai tinkerers.org (SF chapter and events)