While many talks at the London edition of Gartner's Application Innovation & Business Solutions Summit focused on how AI will change everything, the focus has shifted from AI capabilities to the urgent need for transparency, trust, and governance to achieve real value with AI.
We recap the big themes from the floor and what they mean for Enterprise Architects and their teams going into 2027.
Gartner's own data says EA is already expected to lead here. Sixty-six percent of heads of EA call scaling AI a critical priority for 2026. Gartner's framing was direct: heads of EA are on the AI frontline whether they have claimed that role or not.
Yet in conversation after conversation with EAs walking the floor, AI governance came up less often than we expected, even though it was the subject of nearly every other vendor's pitch in the hall. Our hypothesis is that, in practice, EA is not yet part of the AI conversation in many organizations and that’s a critical opportunity EAs need to seize. EAs are the crucial bridge across functional silos and business units, helping decision-makers understand AI's potential impact and real risk on initiatives and processes.
Gartner's own forecast raises the stakes further: AI agent-driven work is expected to grow 2.5x over the next three years, making this a question of when, not if. The advice from Gartner for closing the gap comes down to three moves:
This isn’t to say that EA has to govern AI alone. One of the more useful reframes we heard was ditching the old "department of no" posture in favor of co-managing AI governance as an alliance, documenting guidance by domain, and stepping in company-wide only where it's genuinely needed.
The practical starting point offered was to find out who is actually leading AI in your organization and own the fact that it could well be the EA team. Build a community of practice around AI governance, not a gatekeeping function. Publish your AI initiatives so they're out in the open, not buried in a backlog. And underpin the whole case for EA driving AI direction with one simple discipline: encourage questions, and stay transparent.
If the AI governance seat has not been claimed inside your organization yet, EAs need to get into that conversation sooner rather than later, before a policy gets written without EAs' input and the vicious cycle of being brought into important tech decisions is perpetuated.
Learn how EA becomes the natural control plane for AI governance. See how AI systems can be identified, assessed, and governed as part of your existing EA practice.
When it came to challenges EAs face today, governance topped the 2026 list outright: 17% ranked it their number one challenge, and 62% ranked it in their top three, ahead of budget, talent, and even AI itself.
The powerful stat worth noting, though, is that high performers are 7.2 times more likely to report a successful return on their governance investment. The gap between governance as a bottleneck and governance as a differentiator is almost entirely about execution, not appetite.
Perhaps easier said than done, so what solutions did they propose? Gartner’s answer is to augment existing governance in three moves:
Gartner stated that business value from AI comes from model capability multiplied by workflow fit, trust, and governance, not added to them. If any one of those factors hits zero, the outcome does too, no matter how strong the underlying model is. Gartner calls this the Zero Problem. Alongside it, they shared survey data from 117 software engineering leaders: 90% of AI technologies are in the pilot phase, and only 5% have reached production.
Gartner's forecast is even blunter: by 2029, more than 70% of enterprise agentic AI initiatives will fail, and the cause will be solution misapplication, not a weak model.
In Europe specifically, 74% of 360 IT leaders surveyed already cite governance and compliance as the biggest blocker to scaling AI. The fastest way to move AI out of pilot is not more experimentation. It is fixing whichever factor sits closest to zero, and for most organizations, that factor is trust or governance, not the model itself.
This only emphasizes the critical role EA can potentially play in AI enablement in their organizations. Getting a handle on governance is crucial for unlocking value from AI.
Bad data in the EA repository and data leakage were named the top risks to AI-enabled EA, and Gartner's recommended first step is blunt: audit the architecture data your tools already produce, clean it, then scale anything else.
This connects directly back to the need and value of governance. Getting repository hygiene sorted also lays the groundwork for being the obvious choice when the AI governance conversation finally reaches EA's desk.
There is significant urgency behind repository hygiene, making it much more than an idealistic nice-to-have. Gartner predicts that by 2028, 65% of heads of EA will have rebuilt their operating model around AI, and in a live poll during the session, roughly a third said they're already using AI inside their own EA practice today. Gartner’s journey to AI-enabled EA has seven stages, and auditing EA tools for AI readiness is one of them, sitting before the stages that mitigate risk and prove value. Skipping it risks the later stages inheriting the same bad data.
Getting repository hygiene right isn't just about avoiding risk. Gartner points to five AI capabilities worth evaluating in any EA platform: conversational interfaces, AI interoperability through standard APIs and the Model Context Protocol, generation and discovery, content enrichment, and decision support. None of these will work if people don't trust the repository.
The same gap is evident in how EA handles change. Only 25% of heads of EA say they're very confident adapting their architecture to rapid business change, and Gartner's own framing is blunt: EA can see change coming but struggles to act on it, becoming bystanders in its own organization while the rest of the business moves faster. The fix is the same shift described above, applied to architecture itself: move from static, plan-once documentation to a dynamic model that senses and reroutes continuously, closer to a GPS than an atlas.
Gartner used its own stage to push back on the "SaaSpocalypse" narrative that agentic AI and vibe-coding will gut the enterprise software market. Their counter-prediction: no more than 30% of enterprise application functionality will be replaced by custom-built AI by 2030, and 90% of actively used AI agents built before 2028 will need to be replatformed or rebuilt entirely.
One session went further, reframing the classic buy-versus-build decision as "buy and build, then blend": commodity systems at the base, marketplace add-ons above that, custom build above that, and a blend layer combining all of it into differentiated applications. Composability is the operating model going forward, not elimination of the enterprise app.
For EA, this changes the job itself. The work is no longer to document what you have. It's to model the impact of what you're about to compose. A blended stack only works if someone can see across the whole estate, how technology, process, people, and business capabilities connect, otherwise what should be composable becomes chaotic, and teams bolt on tools without understanding the dependencies.
Learn more about how and why composable architecture is replacing the binary.
Mark Cooper's session at Gartner Apps made a case that's easy to miss in the rush to talk about AI: plenty of what threatens EA's standing has nothing to do with EA's own performance. His "Threat Context Model" nests EA inside a series of rings, EA itself, the IT organization, the wider business, the broader business environment of competitors, vendors, and regulators, and macro trends like politics, economics, and regulation. Threats start in the outer rings and often land on EA's desk with little warning.
Three threats get traced from the outside in. Geopolitics and sovereignty: hyperscaler exits and AI-vendor restrictions push sovereignty onto the CEO and CFO agenda, and EA ends up managing whatever platform strategy leadership decides. IT chargeback: cost-recovery models designed by finance turn architects into billable-hours workers, straining relationships when business units don't see the value connection. Unplanned work: a fragmenting regulatory landscape forces portfolio exceptions to meet deadlines, and when those exceptions collide with shared constraints, "EA is the constraint" becomes the story, even when a rebound cycle of loosened governance later draws audit scrutiny back onto EA.
Cooper's proposed response framework is Control-Influence-Concern: separate what you can directly control, your own words, actions, and choices, from what you can influence without controlling the outcome, and what's simply outside your reach and not worth fighting.
He shared a low-lift recommendation: share the model with your team this week, build a 90-day threat-management roadmap, and mature it into a shareable capability over the following year. Cooper closed the session with a worthy reminder that organizations are inherently chaotic, and that chaos is exactly the argument for EA, not against it.
Two days in London made one thing clear: The organizations that succeed will not be the ones with the most AI, but the ones that can adapt to change the fastest. In our read of the two days, that adaptability starts with governance, clean data, and a seat in the AI conversation.