IBM engineer maps 5 levels of AI coding, from manual to a 3-day agent software factory
TejasKumar_ · x · 2026-10-16
IBM AI engineer Tejas Kumar breaks down 5 levels of AI coding workflows he's tested:
- L1 manual coding: obsolete
- L2 single-threaded agentic coding: Claude Code/Codex conversations or spec coding — slower but controllable
- L3 parallel agents: 7-8 Claude Code sessions, each on its own branch + worktree, merged after independent review
- L4 full software factory: a Linear roadmap plus "PM", "maintainer", and multiple engineer agents coordinating via @-mentions; he let one run 3 days nonstop to clear an entire board, burning heavy tokens but working better than expected
Tradeoffs: L2 is slow; L3 is nice but worktree setup/teardown is annoying; L4 is hard to track — he often doesn't know what's happening in his own project and has to ask for updates, joking about imposter syndrome and FOMO.
Verdict: he'd pick a hybrid of L2/L3 depending on roadmap and timeline — L2 simulates real human collaboration and pair programming, "which is maybe what we all really want anyway."
More from coding & agent
- Open-Source Google Analytics MCP Server Lets LLMs Query GA Data via Natural Language — modelcontextprotocol · 2026-10-16
- a16z podcast: AI cut a 6-month feature to 10 days and widened engineer gaps 10x — a16z · 2026-10-16
- "Hold On, My Agents Are Freezing My Screen": The Daily Grind of the Agent Era — every · 2026-10-16
- Dev builds MCP feeding Claude 3,000 real launch videos to fix samey AI output — Designer_Landscape25 · 2026-10-16
- Satire: 19 Claude agents 'reverse-engineer' Blender, forgetting it's already open source — Dan_Jeffries1 · 2026-10-16
- Engineers Should Mine Gong Calls for Ideas That Never Became Tickets — curious_vii · 2026-10-16