Erich Gazaui, Chief Information Officer at Papa, says AI makes it look as if typing and writing are the only skills you need, and when you leave a gap in a prompt or a design, AI is going to fill that gap very quickly. He explains the intent engine he wrote about, why systems do not automatically know the finish line, and how his team built AI-driven playbooks into almost every department.
Erich Gazaui's first computer was an Apple II, and his mom put him in a computer class to learn BASIC. He was pre-med until organic chemistry, graduated in economics, and learned DOS batch scripts in the dorms so he could play Privateer. In college he and two friends started a company whose first customer was the university: a survey tool for a political science professor, then a computer lab to run it. That company later became a security firm he ran for 15 years. He went on to be a senior director at TrueCar with "a really big hand in the security side, the governance and compliance", was recruited to Beachbody by its CIO, and was head of IT at ZipRecruiter, where vetting job posters made marketplace safety a competitive advantage and procurement ran "janitorial all the way through AWS". Today security, IT compliance and enterprise apps sit under him at Papa.
His warning about AI: "When you leave a gap in a prompt or in a design, AI is going to fill that gap very quickly." The output looks so professional that you won't know what was filled, and the miss may only show up when it is very expensive to change. He compares it to firing a laser pointer at the moon, where being off by a fraction of a millimeter means you miss it. His answer is what he calls an intent engine: a central source of truth for what the company believes. One request at ZipRecruiter to "make sure that marketing is included" in an event shows why: even the head of marketing came back with a question.
Then the other end of the work: the success criteria, the definition of done. Using mean time to resolution as the example, he explains why "systems do not automatically know the finish line" and keep refining past the state you wanted. He talks about the political capital he spent with his CEO and president to build AI properly, how restrictive his team was about which AI features to enable while they built a central knowledge store, and the playbooks that now run in almost every department, including the day the CEO asked to take his old laptop for personal use and the system handled it with no human interaction. His rule for his own meetings: "Everybody has a voice and it's equal. You leave the title at the door."
Key takeaways: Give every AI system and automated playbook its finish line: write the success criteria and the definition of done, or it will keep refining past the state you wanted; Instruct the models to identify gaps rather than fill them, and decide upfront which gaps are intentional; Before you switch AI features on, build one central knowledge store with the goals, processes, teams and names; disable or downgrade the tier on tools that can't point at it yet; Test your definitions on something small. "Make sure marketing is included" turns into who goes on the list, who maintains it and who writes the rules; Put a gate on every action an AI-driven playbook can take, and keep it supervisory until you trust it; Use the visibility IT has. Procurement and governance let you show the board and the audit committee where security and compliance drive revenue
Chapters:
00:00 Cold open
00:42 Meet Erich Gazaui: an Apple II, a 286 and a computer camp
03:15 Pre-med, Privateer and a first company in college
07:21 TrueCar, Beachbody and a workout at orientation
09:52 Beyond the cost center: governance at Papa
14:28 ZipRecruiter: marketplace safety and company-wide procurement
21:22 The missing skills: AI will fill the gap
26:12 Make sure marketing is included: the intent engine
31:43 Start and finish: systems don't know the finish line
39:21 A beautifully wrong sys...
