On-Prem or Cloud for AI? Outcome-Based Pricing and Spotting Agentic AI Washing

Enterprise AI strategy keeps circling back to the same hard questions: where should workloads actually run, how should vendors get paid for outcomes, and how can buyers tell real AI capability from a marketing label. Episode 132 of ConstellationTV takes on all three.

On-Prem or Cloud? The Debate Continues

Host Holger Mueller opened the episode with the Big Debate, joined by Michael Ni and Esteban Kolsky, on whether AI has to run in the cloud or can live on-premises. Michael argued that AI follows its own kind of gravity, pulling compute toward wherever inference demand and real-time decisions actually happen, which points toward a distributed model spanning cloud, edge, and on-prem layers. Esteban pushed back with a data-gravity and elasticity argument, comparing cloud economics to the historical shift away from companies generating their own electricity. Holger argued for the edge on cost and efficiency grounds, citing Apple's on-device AI push as evidence that enterprises don't need to run everything through the server. The debate landed where these things usually do: it depends on the layer, and the next few years will be about getting that placement right rather than picking a single side.

Personalization Is a Means, Not the Outcome

Liz Miller used her Marketing Minute to share takeaways from Pega World, Pegasystems' annual event. Her core point: personalization is a tactic, not the business outcome itself. The real goal is a more durable, profitable customer relationship, and Pega's customer decision hub has been quietly doing that work for years, well before "agentic" became the term of the moment. Liz also referenced comments from Pegasystems CEO Alan Trefler on deliberately testing and deploying agentic AI, without losing sight of why the work matters in the first place.

Outcome-Based Pricing: Great on the Whiteboard, Hard in Practice

Larry Dignan covered the renewed push toward outcome-based pricing, with Oracle, Pegasystems, HubSpot, and UiPath all referencing it recently. His take is skeptical. Outcome-based pricing runs into the same problems revenue-share models always have: nobody agrees on how to measure the outcome, nobody wants to give up their share once there's something to share, and it's unclear who bears the cost if an outcome takes more time or tokens than expected. Larry expects the topic to come up heavily on Q2 earnings calls, but his advice for now is to tread carefully.

Five Ways to Spot Agentic AI Washing

Holger closed the episode with a preview of his new Best Practice report on agentic AI washing, which extends the same skepticism that applied to cloud washing a decade ago. Of the 18 questions in the full report, he highlighted five: whether the capability existed before generative AI took off in 2023, whether it's just a chatbot with a new label, whether it actually runs in the cloud and can scale elastically, whether it can access third-party data rather than just the vendor's own, and whether the underlying capabilities are extensible rather than fixed. Vendors that fail these tests, in Holger's view, are most likely repackaging something that already existed.