Vanishing Gradients releases 'How to Build a Coding Agent' podcast with audio, video and repo
hugobowne · x · 2026-09-15
Hugo Bowne-Anderson publishes the full 'How to Build a Coding Agent' episode with Amp Code's Nicolay Gerold, with audio, video, transcript and repo links.
Highlights:
- An agent is 'an LLM, a loop, and enough tokens'; the episode builds up from a minimal agent in Go through tools, context management and harness design.
- Harness components are assumptions the model can't do it alone, and expire: Amp deleted plan mode once models reliably planned without editing; dedicated read tools gave way to Bash-based search.
- After model upgrades, ask which workarounds to remove and what more to delegate, including having models challenge sub-agent outputs.
- Practical advice: use AI speed to try three implementations in an hour, iterate 20 times, and build your own agent to see which instructions survive summarization.
Related event: Podcast Explains How to Build a Coding Agent from Scratch(2 posts)→
More from coding & agent
- Astra's Context Compaction Takes 3x Longer Than Sol, Dev Notes — banteg · 2026-09-16
- Simfinity.js example: expose selected GraphQL operations as MCP tools via stdio — Affectionate_Hawk971 · 2026-09-16
- Sol 5.6 orchestrating Terra and DeepSeek 4.1 workers delivered the user's best day yet — pvncher · 2026-09-16
- GPT-6 spent $367 and 6.1M tokens over 7h50m to build a rocket in The Universim on its own — Angaisb_ · 2026-09-16
- DataFast Adds MCP/API Support for Mention Tracking and Chart Notes — marclou · 2026-09-16
- Semantica Hits 13k Stars: Open-Source Graph-Native Infrastructure for Accountable AI — tom_doerr · 2026-09-16