Have you had to help your developers learn to orchestrate AI agents during their coding workflows, or was this something they were able to do without much guidance?

I would say more developers need guidance at first on how to design, orchestrate and govern AI agents within their workflows, once they understand the pattern and guardrails they continue further.

I’m not sure I fully understood the intent, but yes, we contributed as a team overall. Orchestration isn’t something static—it requires design decisions, and every team member should be part of those discussions.

Yes, there's been a significant education curve. The agentic paradigm requires a different mental model than traditional development, and that's not always intuitive. The biggest areas where I've needed to provide guidance:

Agent orchestration fundamentals - How agents interact, establishing rules of engagement between agents in a network, and understanding when to let the LLM determine workflow vs. prescribing it upfront.

Tool architecture decisions - Distinguishing between developing discrete tools/skills that agents can invoke versus relying on system prompting to enable tool use. Developers may default to one approach when the other would be more effective.

Identity and observability - Traditional developers don't always think about traceability and consistency in the context of non-deterministic systems. Establishing proper identity management across agent interactions and building observability into agentic workflows requires deliberate design that doesn't come naturally from standard software engineering practices.

The shift from "I control the execution path" to "I establish guardrails and let the agent reason through the path" is the hardest adjustment. Most of my developers needed hands-on examples and pattern guidance before it clicked.