Quick follow-up to my last post about Agentic Development. A couple people asked what it actually means, so here is the simple version: This is for anyone trying to use AI agents to build real applications, but the agent keeps: - changing things you did not ask it to change - burning tokens fixing its own mess - forgetting context halfway through - giving confident answers with weak proof - going down a logical rabbit hole - making you babysit every response If you understand Linux/DevOps, the idea should feel familiar. We do not trust systems just because they look busy. We want logs, tests, boundaries, rollbacks, and proof. Agentic Development is that same mindset applied to AI-assisted building. The course structure I am working on looks like this: 1. Modules 1-3: diagnose the problem Why agents drift, why project memory breaks, and why systems need boundaries before agents touch them. 2. Modules 4-6: install the starter kit Testing guardrails, workflow proof, and a Plan/Contract/Packet framework for giving agents controlled work. 3. The rest of the course: use the agent setup Queues, handoffs, production readiness, team rails, portfolio proof, and scaling the practice. I'm still working on course material, so I am not opening the whole thing today. But I am thinking about making the first 3 modules free as an AI Agent Drift Audit. Quick question for everyone: Which problem have you run into the most with AI agents? Comment one: 1. Drift - it changes things you did not ask for 2. Token burn - you spend too much correcting bad output 3. Context loss - it forgets what matters 4. Rabbit holes - it follows a logical path away from the goal 5. Weak proof - it says done but you do not trust it 6. Babysitting - you have to keep steering every response If you have a real example, post it. I may use some of these examples to shape the first free challenge. Would the first 3 modules as a free audit be useful here?