Latest coding agent drift on goal seek mode issues……Got a really smart model to do plan spec etc. created a spec, task list, validation, test/fix loops against the spec. Then got a slight less really smart model to execute the plan, to save tokens….By morning and a lot of tokens among other things it had somehow created two PRs in the same repo for the same fix, that it then needed to work out how to consolidate the differences. Really quite funny, maybe my new smarter LLMs also have multiple personalities, that want to fix the problem slightly differently and if they stay on goal seek mode, they just keep thinking up new ways to solve the same problem, slightly differently.