The AI followed the SOP perfectly.
Unfortunately, the SOP had the structural integrity of the cockpit in Airplane!
- The source file was outdated.
- The client name was ambiguous.
- Step 7 said, “Send the file.”
- Otto the Autopilot (now apparently alive and well as a chatbot) approved the client copy with three fake stats and the overconfidence of a rubber pilot who has never once worried about consequences.
Surely, someone reviewed it…..right??
Someone did not.
And don’t call me Shirley.
Otto Is Flying the Workflow
“A checklist is a specialized tool. It is for those who realize that even the most expert and experienced among us are fallible.” — Atul Gawande
AI can help turn a process living inside Brenda’s head into an actual SOP.
Want the shortcut here? Damien made a video: How to Create SOPs with AI Chatbots
Please Define “Check the AI’s Work”
“Have a human review it” sounds responsible. So does “land the plane safely.” Both instructions could use a few more details.
The reviewer needs to know:
- Which facts require a source?
- Which calculations need to be redone?
- Which privacy, legal, or client rules apply?
- What should stop the workflow immediately?
- Who owns the final decision?
Otherwise, the review becomes a quick scroll, a little nod, and APPROVED typed with the confidence of Ted Striker during a deeply inconvenient flashback.
“Leaders must own every aspect of their mission and make adjustments to ensure success.” — Jocko Willink
Build the Emergency Exit Before Takeoff
Every AI-assisted SOP should answer five questions:
- What can enter the tool? Approved data only. A great prompt does not turn confidential information into carry-on luggage.
- What is the source of truth? Name the file, system, or database.
- What must a person verify? Include facts, math, tone, context, risk, and anything headed toward a client.
- What stops the workflow? Missing sources, conflicting information, sensitive data, or three statistics that materialized from the same mysterious dimension as the in-flight fish dinner.
- Who is flying the plane? Assign one person the final review. “Someone in Operations” is how everyone ends up staring at Ted Striker for answers.
These skills need practice.
Our video on the human advantages that matter more in an AI-heavy workplace covers several worth strengthening….before Otto starts making personnel decisions.
This Is Work We Help With
We help businesses turn “we should probably have a process for that” into workflows people can actually follow.
- Next-Level AI Workshops help teams use AI responsibly, improve their prompts, and practice the human judgment required around the output.
- Fractional COO/CTO support helps build the process around the tool: ownership, approvals, system boundaries, escalation points, and the occasional emergency exit.
Otto may stay in the cockpit. A qualified human still gets the final button.
PONDER THIS
- Where does your AI workflow currently rely on “someone” checking the result?
Books
- The Checklist Manifesto — Atul Gawande
- Extreme Ownership — Jocko Willink
Build Efficient Systems
Next-Level AI Workshops show you how to turn AI into repeatable leverage.

Comments