IBM AI engineer maps 5 levels of agentic coding — and why he stays at Level 2
TejasKumar_ · x · 2026-10-17
IBM AI engineer Tejas Kumar wrote up the agentic coding levels he tried while shipping 4 production side apps (Dewfall, PrayingPeople, Daiku, his site).
- Level 1: manual coding — nobody does this anymore
- Level 2: one agent in one conversation (Claude Code), his default sweet spot
- Level 3: several agents in parallel git worktrees, for tight deadlines
- Level 4: a software factory — a "pm" agent runs a Linear board and hands work to other agents; one has been building Daiku nonstop since Oct 13, working better than expected
He contrasts agentic coding with Karpathy's vibe coding (reviewing is the difference) and concludes Level 2–3 fits his roadmap and deadline needs best.
More from coding & agent
- Claude can now reverse-engineer and rebuild apps like TradingView and OBS without source code — PrajwalTomar_ · 2026-10-17
- One-prompt interactive pond website from Gemini shows single-file HTML chops — R3d-Gr33n-Blu3 · 2026-10-17
- 15 AI Tools Worth Knowing in 2026: From Firecrawl to Agent Zero — nikola_mr64990 · 2026-10-17
- Your agent eval gate passed, but the next served request still scored zero — papaparija25 · 2026-10-17
- clauden lets Claude Code run on NVIDIA's free hosted models, dev says nemotron nears Sonnet 5.5 — gmaggiotti · 2026-10-17
- Interactive 3D simulator walks you through the inside of an AI agent — SeveralSeat2176 · 2026-10-17