AWS Q with Claude was the first truly eye opening results and that was only earlier this year. Cline with Claude and Qwen3 coder (weeks old) is truly remarkable. MCP servers such as Context7 dramatically reduce lying and guessing as the model has access to accurate information vs quantized outdated memory.

Marketing set expectations too high, and most people try something and if it doesn’t work, they are been there, done that, not trying again.
Copilot sucks, and GPT doesn’t impress me much.

There is something here, but it’s not fully baked and still missing some ingredients. I laugh at the AGI people and I hope they enjoy their hallucinations, but naysayers are a dime a dozen and I don’t hear them saying anything insightful.