What Gartner AIBS London 2026 Told Us About the State of Enterprise Architecture

Two days on the show floor at Gartner’s Application Innovation & Business Solutions Summit in London gave us plenty to notice. A few things came up again and again, in the keynote, in session after session, and in conversation after conversation, and those are worth writing down.

Here’s what stood out to us, and why we think it matters.

1. The AI governance gap: ambition is outpacing oversight

The opening keynote spelled out a shift that’s been building for a while: AI, agentic AI, and generative AI now sit at the top of CIOs’ list of concerns for 2026, ahead of cybersecurity and IT strategy, which held those top spots for years. Boards and CEOs are pushing in the same direction, survey data shared on stage pointed to a large majority of CEOs increasing AI funding and expecting it to reshape their industries, with boards actively encouraging more technology risk-taking, not less.

Our own read on this: the ambition at the top is real, but it’s running well ahead of most organisations’ ability to actually govern what they’re building. Several sessions circled the same idea from different angles without quite saying it outright, so we will: the organisations getting real value from AI aren’t the ones moving fastest. They’re the ones who can see what’s connected to what before they let an agent touch it.

2. Agent sprawl is the new shadow IT

More than one session this year focused on what happens when AI agents multiply faster than anyone’s tracking them, department by department, tool by tool, each one deciding on its own with no central view of what exists or what it’s allowed to touch. It’s a familiar shape to anyone who’s been in this industry a while, the same ungoverned growth that turned shadow IT into a problem a decade ago, just faster, and harder to audit after the fact than a rogue spreadsheet ever was.

The proposed fix, across multiple sessions, was consistent: agents need to be modelled and governed the same way everything else in the enterprise is, tied to real ownership, real risk context, and a live picture of what they can reach. We’d go a step further than most of what we heard on stage: that’s not really an AI problem at all. It’s an architecture problem wearing an AI costume, which happens to be the exact problem we spend our working lives on.

3. AI strategy should start with your business, not your platform

One of the sharper sessions of the conference took direct aim at a habit a lot of enterprises have fallen into, asking “what can vendor X’s AI do for us” instead of starting from their own strategy and working outward. Survey data shared in the session found that over half of enterprise application leaders now run 60% or more of their application portfolio through a single suite provider, a concentration that leaves plenty of room for a platform’s own roadmap to quietly become the strategy, rather than the other way around.

What struck us about that session wasn’t the warning itself, it’s a familiar one, but how openly it was said out loud, on a Gartner stage, to a room full of the people behind exactly those decisions. Strategy that starts with “what does the platform already do” tends to end up shaped by that platform, whichever one it happens to be. Starting from your own architecture first is the only way we know of to keep that from happening quietly.

4. AI cost management: from cost-per-token to cost-per-outcome

One of the best sessions we sat in wasn’t run by an analyst, it was a customer, Haleon, walking through what they’ve learned about the economics of running AI at scale. Their core point: as inference gets cheaper per unit, the total bill for agentic workflows can still climb, sometimes by several multiples, because a single task can consume wildly different amounts of compute depending on how it’s run. Cheaper intelligence doesn’t automatically mean a cheaper outcome.

Their recommendation was to stop measuring cost per token or per model call and start asking a different question: how much did it actually cost to resolve one customer problem, end to end. We think that’s the right question, and not just for finance teams. It’s a governance and architecture question first, the finance team just happens to be the one that notices when nobody’s asked it.

5. Why enterprise architecture matters more, not less, in the AI era

Maybe the most encouraging thing we saw all week: EA was not treated as a back-office function anywhere on the agenda. Multiple sessions, from multiple vendors and practitioners, converged on the same idea, that the shift to agentic AI doesn’t sideline enterprise architecture, it turns it into the precondition for doing AI responsibly at all. Modernisation was framed less as a one-off leap and more as a continuous, iterative discipline: understand where you are, prioritise, plan, execute, and go round again.

We’ve watched that cycle play out directly with organisations we work with. HMRC’s own path from an unmanaged capability to a governed architecture practice serving 1,000+ users is one recent example, and the shape of it matched almost exactly what multiple sessions at AIBS described in the abstract.

We heard versions of that thesis from analysts, from other vendors on the show floor, and from the practitioners we spoke with directly, independently of each other, in language that clearly wasn’t borrowed from one another. When the same idea surfaces that consistently, without anyone comparing notes first, it’s usually not a coincidence. It’s a signal.

The throughline

Put together, the story of AIBS London 2026 was less “AI changes everything” and more “AI raises the cost of not having your architecture in order.” Agent sprawl, platform-led strategy, unpredictable AI economics, the renewed case for EA itself, every theme above comes back to the same starting point: a connected, honest, governed picture of your business, data, applications, and technology.

That’s not a new idea. It’s just a more urgent one than it was a year ago, and this conference left that pretty hard to miss. It’s also, unsurprisingly, the thing we spend most of our time thinking about.