Enterprise Agentic Computing—12 Rules
John Taschek is a market analyst at Salesforce who, inspired by Codd’s twelve rules for relational databases, has written twelve rules for agentic computing. They are a master class in what it’s going to take to govern and run agentic workflows at scale, and every CIO should read them and discuss them in depth with their teams.
For the rest of us, they are a wake-up call as to how heavy a lift pervasive agentic computing will be. Daunting as that might seem, it’s actually a call to action to start now with highly contained, high-value use cases where all twelve rules can be effectively addressed locally. Not only does this result in short-term ROI from releasing trapped value, it also lays the foundation for sizing future investments in infrastructure, personnel, tooling, and applications.
The agentic revolution is not for the faint of heart. Neither is it for the fly-by-the-seat-of-our-pants crowd. Time to buckle up and buckle down.
That’s what I think. What do you think?
Geoffrey, you might not remember, but years ago I contacted you after discovering something you wrote online. I was a salesperson suddenly asked to run marketing with no experience. You answered my questions via email and generously helped me plan the campaign. That work helped us grow 600% in the first year and then double each of the next three years. The company was eventually acquired, and the journey helped create the first PPM platform. I have never forgotten what you did for me. Reading this brought the lesson back. Your recommendation to start agentic transformation with a contained, high-value use case is not simply a deployment tactic. It is a way to build the complete conditions for adoption locally before asking the enterprise to scale them. The capability may be new, but the discipline is familiar: choose the right beachhead, solve the whole problem, establish trust, and expand from evidence rather than enthusiasm. Now, more than 20 years later, back out of retirement, I find myself trying to introduce another innovation into that same industry and working very hard to cross the chasm all over again. Thank you again for lessons that have lasted a career.
John Taschek Geoffrey Moore I love this 12 rules for Agentic AI transformation. Good read and best practice for all organizations going through Agentic enterprise journey. Salesforce
Love it. One dimension worth adding: information only has value when it reduces uncertainty. If the agent confirms what you already knew, the bits are wasted. Every rule adds signal. Every rule also adds cost: compute, latency, governance overhead, review cycles. 12 layers of validation for a $50 decision is noise pretending to be signal. The 13th rule: value per bit delivered. Trapped value is real. So is trapped cost.
🙏💐
Rule 9 (Human-in-the-Loop by Design) is the one I see broken most in practice. Teams say they have it, but what they actually have is rubber-stamp review — humans approving agent outputs they can't meaningfully verify at the speed agents produce them. In the agentic coding pipelines I've built, we had to rethink HITL entirely: instead of reviewing every output, we shifted to reviewing the agent's decision boundaries upfront and only escalating genuine edge cases. The result was fewer overrides but much more meaningful ones. Which of the 12 rules do you find enterprises struggle with most when moving from pilot to production?