H2 2026 Reckoning For Governance, FinOps, and Everyone's AI Budget | CRTV Episode 134

Every two weeks, Constellation TV brings together our analysts to break down what's actually happening in enterprise technology. Episode 134 covers what the second half of 2026 has in store, how SAP is building out its agent platform, the five moves enterprises need to make, according to Esteban Kolsky's monthly board report, and why Larry isn't impressed by the latest AI job-displacement open letter.

The H2 2026 debate: screw-ups, FinOps, and the rise of the AI project manager

Liz, Larry, and Martin opened the show with their predictions for the rest of the year, and the group didn't fully agree — which made for a better debate.

• Larry's take: expect a wave of agentic AI screw-ups serious enough to push laggard enterprises onto the governance bandwagon by year-end. He expects a mix of headline-grabbing failures and the quieter, more familiar kind — botched implementations where the vendor, the consultant, and the customer all point fingers at each other.
• Martin pushed back with a different angle: instead of governance framed around data and security, the real forcing function will be financial. He argued enterprises are running AI the way someone might run a business off a generator instead of the grid — wildly inefficient — and that a "FinOps for AI" discipline will emerge to rationalize the sprawl of tools, agents, and corporate mandates that are currently canceling each other out. That means new models and new metrics, not just faster versions of old financial ledger processes.
• Liz brought a project-management lens to the conversation, arguing that enterprises already have the function built to manage this kind of complexity — the project manager — and that AI ops will increasingly fall to that role, pulling in the CIO and CDO to create the cross-functional control layer AI needs.

Asked to name the most "ridiculous" conversation likely to dominate H2, the group landed on outcome-based pricing models. The consensus: AI deployment maturity isn't yet mature enough to draw a straight line from spend to outcome, and anyone selling on outcomes right now is taking on real risk. Larry added his own wildcard prediction — a hyperscaler quietly throttling back AI infrastructure spend, followed by a media narrative about an AI infrastructure "bubble," even as enterprises that get governance right start seeing real ROI. He also expects a NIMBY-style backlash against data centers to build toward a fever pitch ahead of the U.S. midterm elections in November.

Holger Mueller kicks off a 3-part series on Enterprise Application Platforms

Next, Holger Mueller introduced the first in a three-part series on Enterprise Application Platforms (EAPs), using his work with SAP's Joule Studio as the case study. He frames EAPs around three generic use cases — extend, integrate, and build — and argues they've become table stakes for enterprise software since 2022, because no vendor's out-of-the-box product covers everything and enterprises need a platform to build the rest themselves.

The AI angle touches all three use cases: extending an application can now happen through natural-language requests, integration is increasingly AI-assisted, and building, including standing up agents, is moving toward intent-based development, where a prompt generates the code. Part one of the accompanying benchmark report compares Joule Studio against SAP's BTP across three AI agent use cases: creating an agent, adding a skill, and building a full-scale backend application. Parts two and three of the video series will go deeper into the report's findings.

Esteban Kolsky's five actions from this month's board report

From there, Esteban Kolsky's two-minute board report distilled the current state of enterprise AI into five actions:

1. Technology spending is separating from general economic caution, with enterprises prioritizing investment in data readiness, security, governance, private platforms, and infrastructure.
2. Public frontier models alone can no longer create differentiated value — context, privileged data, and homegrown models are what separate the organizations doing AI well.
3. With agentic AI adoption now at 74%, the conversation has shifted from adoption to authority: agents need permissions, cost models, monitoring, and a clear reversal path, with cybersecurity increasingly the foundation for execution governance.
4. Enterprise AI also requires a balance between "brains" (CPUs) and "brawn" (GPUs), alongside storage, edge computing, and observability.
5. And finally, talent is becoming the hardest constraint — experienced people who can navigate governance ambiguity and apply judgment are what make AI's cost efficiencies actually pay off.

The question for next month: will enterprises manage to connect AI spending to real impact, or will governance and talent gaps keep widening the space between adoption and value?

Larry vs. the "We Must Act Now" open letter

Closing out the episode, Larry took on this week's open letter from economists and other signatories urging policymakers to prepare for AI-driven job displacement. His read: the letter is long on academic hedging — heavy use of words like "could" and "may" — and short on evidence that displacement is actually happening yet. He points out that where AI has been blamed for layoffs, a lot of those companies were already over-hired coming out of the pandemic and may be using AI as convenient cover.

His bottom line: the topic is worth studying, but it's too early to build incentives, guardrails, and institutions around a transformation that hasn't yet proven out — especially since, as he notes, economists tend to be backward-looking by nature.