Vjeux says coding loops still fail outside training data unless a human babysits them
Vjeux · x · 2026-07-26
Vjeux says software engineering is not “solved” yet, despite claims that loops are working.
He argues that on tasks outside the training distribution, the latest models still produce frustration after frustration unless a human is babysitting the process.
His point is not that progress is absent, but that current agentic coding workflows still depend heavily on close supervision for anything nontrivial.
More from coding & agent
- Agent workflows need deadline propagation, not just timeouts — blaizedsouza · 2026-07-26
- MCP release candidate adds stateless scaling and enterprise auth for agents — davemccollough · 2026-07-26
- Shopify support agent splits low-risk address edits from approval-gated refunds — decentBab · 2026-07-26
- A developer says Devin won them over after spending $534 in two days — NERDDISCO · 2026-07-26
- Grok Build’s Plan mode and worktrees are the real leverage points — tetsuoai · 2026-07-26
- Codex now handles massive parallel QA better and catches complex bugs — steipete · 2026-07-26