Automate Now and Keep the Door Open for AI Agents

You can automate the work that slows you down today and design it so AI agents can take on more of it tomorrow. Speed now and a governed route to autonomy later are one design decision, and you can make it with the processes you already run.

Picture a vendor onboarding that crosses marketing, legal, finance and IT, with approvals in three different systems. The work gets done, often through a side message or a quick call that sits outside the approved model. Those workarounds are useful. They show you where the process needs a better home and which hand-offs are worth designing first.

This episode separates four things that rigid automation tends to fuse together: process design, executable work, decision rights and runtime evidence. Keep them as distinct, connected parts and you can raise the level of autonomy step by step without rebuilding. It also explains why runtime evidence for an agent needs to carry the context behind a decision, the way an itemised receipt does, and not only a log that a step finished. The discussion draws on Deloitte’s work on agentic AI and orchestrated workflows and on the use-case and risk-based approach in the NIST AI Risk Management Framework.

Five questions give you a practical start: which workflow creates immediate operational value, what autonomy level it should reach over time, which hand-off or approval creates the longest avoidable wait, what runtime evidence would show the approved process is being followed, and which human decisions and controls stay with people. IGX360 Insights shows which parts of your process knowledge you can stand behind, with provenance, confidence, freshness and ownership beside each answer. Which workflow would you automate first if you knew it could carry an agent later?