Every two weeks, Constellation TV brings together analysts to break down what’s really happening in enterprise technology. Episode 134 covers what’s in store for the second half of 2026, how SAP is building its agent platform, the five moves companies need to make according to Esteban Kolsky’s monthly board report, and why Larry isn’t impressed with the latest AI separation open letter.
H2 2026 Discussion: Failure, FinOps, and the Rise of AI Project Managers
Liz, Larry, and Martin announced their predictions for the rest of the year at the beginning of the show, but the group didn’t completely agree. That created a better discussion.
- Larry’s thoughts: By the end of the year, we can expect a wave of agent AI failures severe enough to push laggards into the governance bandwagon. He expects a mix of headline-grabbing failures and the quieter, more familiar kind of failures: implementation failures where vendors, consultants, and customers all blame each other.
- Martin countered from a different angle. Instead of governance built around data and security, the real enforcement function will be financial. He argued that the “FinOps for AI” discipline will emerge to streamline the sprawling array of tools, agents, and corporate missions that currently cancel each other out, with companies operating AI the way they run their businesses using generators rather than power grids, which is highly inefficient. This means new models and new metrics, as well as faster versions of old financial ledger processes.
- Liz brought a project management lens to the conversation.n argued that enterprises already have built in capabilities to manage this kind of complexity, namely project managers, and that AI operations personnel will increasingly take on that role, bringing in CIOs and CDOs to create the cross-functional control layers that AI requires.
Asked to name the most “ridiculous” conversation likely to dominate the second half of the year, the group came up with an outcome-based pricing model. The consensus is that AI implementation is not yet mature enough to draw a straight line from spend to results, and those who sell results at this point are taking substantial risks. Larry added his own wildcard prediction. The idea is that hyperscalers have quietly reined in AI infrastructure spending, and then media coverage of an AI infrastructure “bubble” continues, even as companies with good governance begin to see real ROI. He also expects the NIMBY-style backlash against data centers to intensify ahead of the US midterm elections in November.
Holger Müller begins a three-part series on enterprise application platforms
Holger Mueller then introduced the first of a three-part series on Enterprise Application Platforms (EAPs), using SAP’s work with Joule Studio as a case study. He frames EAP around three common use cases: extend, integrate, and build, and argues that EAP has become a critical element of enterprise software in 2022 and beyond. That’s because there are no vendor off-the-shelf products that cover everything, and businesses need a platform to build the rest themselves.
The AI angle impacts all three use cases. Application enhancements can now be performed through natural language requests, integrations are increasingly powered by AI, and construction, including launching agents, is moving toward intent-based development where prompts generate code. Part 1 of the accompanying benchmark report compares Joule Studio and SAP’s BTP across three AI agent use cases: creating agents, adding skills, and building full-fledged backend applications. Parts 2 and 3 of the video series provide further details on the report’s findings.
Esteban Kolsky’s 5 Actions from This Month’s Board Report
From there, Esteban Kolsky’s two-minute board report distilled the state of enterprise AI into five actions.
- Technology spending has moved away from general economic caution, with companies prioritizing investments in data readiness, security, governance, private platforms, and infrastructure.
- The public frontier model alone can no longer create differentiated value. Context, privileged data, and homegrown models separate organizations that successfully leverage AI.
- With agent AI adoption reaching 74%, the discussion has shifted from adoption to authority. Agents require authority, cost models, oversight, and clear reversal paths, making cybersecurity an increasingly important foundation for execution governance.
- Along with storage, edge computing, and observability, enterprise AI also requires a balance between “brains” (CPUs) and “brains” (GPUs).
- Finally, human resources are becoming the biggest constraint. Experienced people who can navigate governance ambiguity and make decisions are what actually make AI more cost-effective.
Next month’s question: Will companies be able to link AI spending to real impact, or will governance and talent gaps continue to widen the gap between adoption and value?
Larry vs. “We Must Act Now” Open Letter
To conclude the episode, Larry highlighted this week’s open letter from economists and other signatories urging policymakers to prepare for job losses due to AI. His reading is that the letter is long on academic risk avoidance, with heavy use of words like “could” and “might,” and a lack of evidence that evacuations are actually still happening. He points out that while AI has been blamed for layoffs, many of those companies are already overemployed due to the pandemic and may be using AI as a convenient cover.
His conclusion is: While this topic is worth studying, it is too early to build incentives, guardrails, and institutions around unproven change. Especially since, as he points out, economists tend to be inherently backward-looking.
