Your AI Agent Will Inherit Your Team’s Workarounds

Is your AI agent inheriting your team’s workarounds?

I’ve completed operator interviews, captured baselines, and supported future state redesigns across dozens of companies.

They had two things in common:

Workarounds everywhere, and documentation as an afterthought.

This was true when I was deploying supply chain solutions using SAP ECC, SQL Server, and Crystal Reports in the mid-2000s, and it’s true today.

Teams run the same plays long enough that the process goes on autopilot and the cheat codes never leave people's heads.

Introducing a new hire means transferring all of it: how teams collaborate, who owns decisions, what data represents truth, and governing policies. I've rarely seen an onboarding plan that actually pulls that off.

Historically, this was tolerated because human effort could absorb it. With AI agents these problems are harder to observe and easier to amplify.

People notice ambiguity and ask questions. An agent won't reliably flag what it wasn't built to flag. Which means it can execute incomplete logic repeatedly, quickly, and across far more transactions than a person. Auto-routing POs based on incorrect lead times can trigger a cascade nobody catches until it's too late.

Technology does not compensate for missing process discipline. So fix the discipline first.

Start with a high-volume, high-consequence exception workflow. Document what triggers it, what data is used, who decides, when it escalates, and what a good resolution looks like. That's the input your agent actually needs.

VentureBeat recently covered this from an enterprise-wide lens: https://venturebeat.com/orchestration/enterprise-agentic-ai-requires-a-process-layer-most-companies-havent-built