Turn dictated specs into tickets and code with LLM agents
A developer dictates a spec, has LLMs densify it, then plan and scaffold the project, and an agent files implementation tickets per module. ·
Done with a custom-built or unnamed agent
- What they did
- The developer dictates a rambling spec, Whisper transcribes it, and an LLM condenses it into a structured markdown spec filed as issue 1. After human edits and another LLM densifying pass, an LLM writes a plan and scaffolding. An agent files per-module tickets, parallel agents write commit-level tasks and implement them as PRs, and an adversarial reviewer plus rework agents iterate.
- How it went
- The author reports quite good results, with most PRs landing as-is and occasional corrections when the agent took the wrong approach. No numbers are given; the author says it is far cheaper than human developers despite many parallel agents and CI runs.
- Worth knowing
- Works best when code is highly modular so agents rarely cross modules; the author keeps shared workflow, style, and lint policies in one repo vendored into each project.
Try it yourself with Claude Code
I'm going to paste a rough, dictated spec for [project idea]. First rewrite it into a tight, complete spec and list any open questions for me. Then propose a module-by-module plan, scaffold the project in [language/framework], and write one implementation ticket per module as a markdown file in a tickets folder, each with scope, acceptance criteria, and dependencies. Stop when the scaffold runs and every module has a ticket, then show me the list before doing any more work.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: HN · Oct 31, 2026
Related use cases
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Configure a local AI lab and build a knowledge graph by chat
Users had Hermes Agent set up a local AI stack (liteLLM, Postgres, Prometheus, Grafana) and install Gbrain as an MCP server to build a personal knowledge graph.
Build a Pong game and website from scratch
A tester asked Devin to create a basic Pong-style game and a website, and it finished both in under 20 minutes.
Migrate a password reset flow to modern libraries
A developer had Claude Code migrate a legacy web app's password reset flow; it needed close supervision, drifted off task, and ended up slower than doing it by hand.
Point a computer-using bot at any repo
A commenter says Grok Bot can use a computer and browser and be told to work on any given repo, unlike a repo-specific cloud agent.
Migrate a website from a CMS to raw code and Markdown
Lee Robinson used Cursor agents to move cursor.com off its CMS in three days for $260 in tokens, with subagents handling a broad refactor.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.