Agentforce was sold as autonomous. It turns out autonomous still needs someone who knows what they’re doing.
Salesforce recently added a deterministic scripting layer called Agent Script, essentially a structured control layer that governs how agents behave in production. The reason it exists is telling: enterprises were getting inconsistent results, with identical scenarios triggering different agent behavior depending on how the underlying model interpreted intent. Salesforce engineers reportedly called it a “doom-prompting” cycle.
The fix makes sense. But it also means Agentforce implementations now require workflow mapping, data modeling, and scripting discipline on top of the AI skills everyone was already scrambling to find. That’s not one profile. That’s two or three.
I’ve been talking with architects who are actually standing these implementations up, and the picture is humbling. Agent Script just came out of beta and isn’t fully production-ready. The platform is moving fast enough that the people implementing it are navigating issues in real time, not following a playbook. What goes into prod matters, and right now attention to detail is critical.
What I’m watching is companies posting Agentforce roles based on last year’s understanding of what the platform requires. The job description hasn’t caught up to what the platform actually demands.
If you’re building a team around Agentforce in 2026, the role you think you’re hiring for may not be the role you actually need. Happy to think it through with you.
https://oakcresttalent.com/bookme
Are you seeing implementation complexity catch companies off guard?
