Theo claims new model hallucinates 10x worse in real codebases; bugs traced to ACP integration
cgarciae88 · x · 2026-11-01
Theo (t3.gg) slammed a new model as "incapable of working in a real codebase without hallucinating in egregious ways — not 5% worse, 10x worse." Per the attached screenshot, Antigravity's Patrick Toulme replied that two bugs in antigravity-acp — an ACP bug in the T3 integration, not a Gemini model issue — were already fixed internally and shipping in the next release, thanking Theo for the repro. A classic case of model-blaming vs toolchain bugs in coding-agent land.
More from coding & agent
- Skills hold know-how, MCP connects tools: a layered Claude workflow you can copy — cfregly · 2026-11-01
- Travelers share how personal agents proactively manage flights, packing and trip logistics — ColleenMBrady · 2026-11-01
- Deletion is the harder call in agent memory: Reddit devs debate when to drop vs update — MemStrata · 2026-11-01
- Don't default to summarize: simple /bookmark /apply-to skills for agent article handling — gethackteam · 2026-11-01
- Dev Open-Sources Sentinel to Block Indirect Prompt Injection in MCP Tool Outputs — Infamous_You7977 · 2026-11-01
- Developer implements the RSA algorithm from scratch — cneuralnetwork · 2026-11-01