I Read Dynatrace's 2026 Predictions. Two Are Right, One Is Wrong.

Dynatrace just published their 2026 Observability Predictions. Six predictions, glossy PDF, executive tone. I read it on a flight this week and wrote down what I actually think, not the LinkedIn-carousel version.

Two predictions are exactly right. One is dressed up to sound new and isn’t. Here’s my reaction.

Prediction #2 is the one most executives will ignore

The prediction: “The path to autonomy begins with proven operational maturity.” In plainer English: you cannot hand an AI agent the keys to production if your humans still can’t get a clean deploy out the door. AI inherits whatever dysfunction already lives in your pipelines, your on-call rotations, and your data quality.

This is correct. It is also the prediction every board-level AI presentation skips over, because “we need to fix our release process before agentic AI pays off” does not clear the budget committee.

I have seen this pattern twice already this year. A retailer with a €200M tech budget, pitching me on “agentic operations” while their staging environment drifts three weeks behind prod on a good month. A bank that wants autonomous customer-service agents with four different ticketing systems and no unified customer record. The AI part is the last 20% of the problem. The first 80% is infrastructure the CFO would rather not pay to fix.

Dynatrace frames the sequence as Guided Automation → Preventive Operations → Full Autonomous. That maps cleanly to how I’ve been advising clients this year, and I wrote up a version of it for the AI maturity model at wetheflywheel.com. Short version: if you cannot honestly say you’re operating at stage two, stage three is a PowerPoint, not a roadmap.

Prediction #3 deserves a louder megaphone

“Resilience becomes the new benchmark for operational excellence.” The eBook cites independent research pegging annual UK revenue at risk from payment outages at £1.6 billion, and France at €1.9 billion. Customer patience drops inside fifteen minutes. One in three customers will walk after a single incident.

Those are stats worth printing and taping to a wall. What the eBook understates is how asymmetric the exposure is. A payment outage does not distribute its damage evenly. It concentrates on your highest-intent customers, the ones mid-transaction. You lose the revenue and the cohort of customers most likely to have converted on repeat visits. The compounding is brutal.

Resilience is not a tech-stack line item. It is a P&L line item that has been hiding inside the infrastructure budget. The executives who grasp this in 2026 will treat reliability engineering the way they treat fraud prevention: a revenue-protection function, not an operational cost.

Prediction #1 is recycled thinking

“Agentic AI triggers a new era of system complexity.” No, it doesn’t. It triggers the same era we have been living in since 2014.

I wrote about this in more detail, but the short version: I have watched engineering organisations slide into the same complexity trap three times. Microservices sprawl at Huge around 2014. API gateway sprawl at adidas a few years later. Now agent sprawl. The technology label changes. The failure mode is identical: individual components work, the coordination layer is missing, and the pain is always felt two quarters after the architecture decision that caused it.

What Dynatrace calls “a new era of system complexity” is the same pattern at a higher layer. Agents coordinating with agents is not conceptually different from services coordinating with services. It is distributed systems with a fancier hat. The observability work required is largely the work you should have done five years ago, applied to a new surface area.

I am not dunking on the prediction for sport. I am dunking on it because framing it as novel lets leadership teams write a new budget line instead of asking the uncomfortable question: why didn’t we solve this the last two times?

The one I’m still thinking about

Prediction #5, that human + machine collaboration becomes the growth engine with AI as a “high-speed intern,” is the one I don’t yet have a confident take on. The framing is clean. The practice is messier. More on that one once I have run it through a few more client conversations.

Until then: fix your release process before you hire the intern.