Executives have been bracing for AI disruption while struggling through AI pilots. Meanwhile, the P&L has hardly moved. That changes in 2027. Our new report, Predictions 2027: Artificial Intelligence, identifies five fault lines where pressure has been building and is about to give. These run through AI spending, governance, model economics, platform architecture, and the enterprise context that agentic AI needs to scale.

The technology is moving faster than most organizations can absorb, so running governed, secure AI in production in 2027 will give you an advantage over competitors still experimenting and trying to get the right strategy. Here are details on three of our five 2027 AI predictions.

AI Will Take One Dollar In Five Of Business Tech Spending

Global tech spending will reach over $6 trillion in 2027, according to Forrester’s global tech market forecast, and one dollar in five of business technology spending will go to AI. Most of that money buys servers and equipment, but that will translate to AI tokens and credits for enterprises, whose budgets won’t stretch to cover it. Most tech leaders do expect a bigger budget next year. Only 53% expect it to beat inflation, however, so nearly half will lose real purchasing power right when their business is asking for more AI and paying more for it.

Third-party services will take the hit. Leaders already plan their deepest cuts in services, and AI keeps pushing down what consulting and managed services cost to deliver. Open your services contracts and your seat-based software renewals six months early. Claw back what you can and put those funds into planning for AI consumption costs and context engineering.

Governance Gaps Will Derail Flagship AI At Three Big Enterprises

As we’ve recently seen, a set of principles won’t control an agent. Operational agentic governance takes skills, tools, and controls that enforce policy. In 2027, these will still be scattered across technology domains, out of reach for the business governance professionals setting policy. Over half of genAI decision-makers already name agentic sprawl as a challenge, and traditional governance was never built for software that picks up tools and acts on its own. In 2027, unregistered agents will continue to spread, and as models improve, what they do may get your firm in trouble.

Meta gave everyone a preview in June. It paused an internal program that recorded employees’ keystrokes and screens to train AI agents after that data turned out to be visible far more widely inside the company than anyone intended. It was a big “oops” moment and just one of many. In 2027, at least three well-known enterprises will reverse a flagship AI deployment or push it back a year, each time because business units carry accountability for AI with no power to enforce policy in the software their agents run. Stand up a dedicated AI governance team that works alongside security and risk, IT, development, legal, and data governance, then give it the means to enforce policy in code at scale.

Harnesses, Not Models, Will Drive Three-Quarters Of AI Platform Decisions

Enterprises will buy the harness around the model and care much less about which model sits inside it. The harness is the runtime that routes work, pulls context together, manages tool access, and checks the output. Accuracy and running cost both get settled in that layer. Teams running long-lived coding and research agents keep running into integration problems well before they reach the limits of model reasoning, and in 2027, that gap will widen. Vendors have already started turning model choice into a drop-down in the platform; in 2027, enterprise energy will go into selecting, building, or hardening the harnesses.

Score the vendors on your AI platform shortlist on runtime and context control, and let model benchmarks count for less. Before you sign anything, pin down which layer of the stack the vendor owns.

Two More Fault Lines Are In The Full Report

The report makes two more calls. One is about open-weight models doing a large share of enterprise AI work on a sliver of the spend. The other says that only one in four enterprise context-layer efforts will be able to support agentic AI at production scale.

Forrester clients can read the full Predictions 2027: Artificial Intelligence report and book a guidance session with me and my coauthors to plan for 2027.

Not a client? Download our complimentary Predictions 2027: Technology & Security guide, which covers all of our top predictions for the year ahead. And sign up for our public webinar on December 10, 2026, where we’ll dig into those predictions in greater detail. Additional resources, including blog posts, guides, and webinars, are available on the Predictions 2027 hub.

Share