Stop looking for AI use cases

The use-case workshop feels productive because it produces visible output. But a spreadsheet of AI ideas can become a substitute for understanding where the organization is actually paying for friction, uncertainty and bad decisions.

Most companies do not have an idea shortage. They have a problem-selection problem. In Episode 2, Berk Bayri takes apart the familiar “AI use-case” workshop and replaces it with a more demanding question: where is the organization repeatedly paying for friction, uncertainty or poor decisions? Through sales, content and productivity examples, the episode shows why the same AI tool can create very different outcomes depending on task, expertise and workflow. It also explains why specificity matters more as software gains more autonomy. Featuring a short excerpt from Andrew Ng on the advantage of concrete, falsifiable ideas.

References

Stanford HAI, 2026 AI Index Report — Economy. 88% organizational AI adoption; 70% of organizations using generative AI in at least one function; agent deployment remained in the single digits across nearly all business functions.

Deloitte AI Institute, State of AI in the Enterprise 2026. Survey of 3,235 business and IT leaders across 24 countries. 30% said they were redesigning key processes around AI; 37% reported relatively surface-level usage with little or no change to underlying processes; only 21% reported mature governance for agentic AI.

Erik Brynjolfsson, Danielle Li and Lindsey R. Raymond, Generative AI at Work, Quarterly Journal of Economics 140(2), 2025. AI assistance increased customer-service agent productivity by 15% in the studied firm, with heterogeneous effects across workers.

METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. Randomized controlled trial: in this specific setting, AI access increased task completion time by 19%.

METR, We are Changing our Developer Productivity Experiment Design, February 2026. METR says late-2025 tools likely provide greater speedups, but selection effects make the newer estimate unreliable.