Enterprise AI Needs an Operating Model, Not More Agents

SAP's CEO Christian Klein said something at Sapphire 2026 that I think a lot of vendors don't want you to hear. He admitted the company changed direction on enterprise AI "around eight or nine months ago," pulling back from a strategy heavy on AI technology and refocusing on something more specific: business outcomes that add up to what they're calling the "Autonomous Enterprise."

That's a pretty significant course correction for the world's largest ERP vendor to announce publicly. And buried inside it is the same thing I keep running into with mid-market companies that are six to eighteen months into AI deployments and wondering why results haven't materialized: they built on the wrong foundation.

Enterprise AI Is an Operating Model Problem, Not a Technology Problem

Enterprise AI success in 2026 is not primarily a technology problem. Honestly, the models are capable and the tools are available. What most $50M to $500M companies lack isn't access to agents, so much as an operating model that gives agents real work to do.

SAP's Autonomous Enterprise vision rests on three principles: process knowledge (deep, industry-specific understanding of how the business actually runs), business data (enriched, connected, contextual data that AI can actually reason over), and governance (the backbone that keeps decisions traceable and within policy).

And IDC data backs this up. Its decision-velocity research notes that resolving a business decision manually can still take up to seven days—gathering the data, deploying the decision logic, weighing the options, documenting it, and getting approval. Frankly, autonomous systems are designed to close that kind of gap, but an AI agent running on top of disconnected systems, fragmented data, and unclear decision authority doesn't help... and more likely, adds more noise.

Why This Lands Even Harder for Mid-Market Operators

For the enterprise segment SAP serves, this reframing makes sense. But it lands even harder for mid-market operators who don't have the luxury of a multi-year transformation budget.

What I've generally landed on is this: the companies getting real ROI from AI right now are the ones who did the boring work first. They established clean data pipelines. They documented the actual decision logic inside their workflows (not just the steps, but the "who decides what with what information"). They built governance that's practical, not performative. And then they deployed agents on top of that foundation. That's not sexy stuff, but damn it works.

Gartner projects that 40% of enterprise applications will include task-specific AI agents by end of 2026, up from less than 5% today. That number is going to accelerate a lot of AI spending. Some of it will generate returns. A lot of it won't, because companies will keep deploying on top of the same operational gaps.

Three Places to Look Before You Expand Any AI Deployment

The first is data connectivity. Not whether data exists, but whether AI has access to enriched, contextual data in real time. Most companies have data in silos that require human translation to be useful. An agent working from a monthly export is not the same as an agent working from a live, connected data layer. The capability gap between these two scenarios is enormous, and most organizations are closer to the first than the second.

The second is decision authority. AI agents stall or hallucinate when the rules governing decisions are unclear or inconsistently enforced by humans. Before deploying an agent into any workflow, the right question is: who currently makes this decision, with what information, and under what constraints? If you can't answer that clearly, the agent can't either.

The third is outcome definition. What does success actually look like? Not the task the agent completes, but the business result it's supposed to move. Margin recovery. Cycle time reduction. Win rate improvement. You need a baseline, a measurement window, and a clear signal before deployment, or you'll spend a year automating activity with no idea whether it's creating value.

Frame AI Around Operating Model Maturity, Not Tool Count

If you're heading into a board meeting or an investor conversation where AI comes up (and it's going to come up), I'd think about framing it around operating model maturity rather than technology investment. The question isn't "how many AI tools do we have?" It's "do we have the data layer, decision framework, and governance structure that would actually let those tools produce results?" Companies that can ask AND answer this question are more than ready for the next wave of AI evolution coming our way.

The market is maturing past the deployment announcement and into the accountability phase... and we're all on deck to pay the bill.

The ones who built the operating model first are going to look very different on a due diligence call eighteen months from now than the ones who deployed agents and hoped the ROI would sort itself out.

Do the Boring Work First

If you're allocating budget or headcount toward AI in the second half of 2026, the highest-leverage thing you can do right now is pressure-test your operating model before you expand your agent footprint. I highly recommend mapping your decision logic, auditing your data connectivity, defining metrics (no brainer...), and ONLY THEN should you deploy.

That sequence isn't necessarily "slower". Instead, it's actually what gets you to durable ROI instead of 12 months of inconclusive automation.

If you want to work through what this looks like for your specific organization, I'm happy to dig into it directly. Reach out here.