Toward the end of the summer, one question I am seeing come up more and more: will AI eliminate enterprise architects?
Forrester raised it, and I have heard it from a number of people across events over the summer. Sometimes it’s said with a smile and sometimes with concern. Behind the question sits a bigger shift in ways of working. As agents gain more autonomy and start talking to other agents, far more decisions are being made at an accelerated pace. Everyone understands the promise of AI, but the hard part is giving it the context and guardrails to get those decisions right.
After a quarter of conversations, I’m very much aligned with Charles Betz and Joseph Schiavone, in their article that AI isn’t going to eliminate the need for EAs. If anything, the role of the Enterprise Architect becomes even more important. Right now, the job is changing faster than at any point in the 13 years since we started Ardoq, so it’s critical that architects seize this opportunity AI offers them to drive how their organizations evolve.
This summer we spent a lot of time engaging with fellow practitioners at events all over the globe.
We joined the Chief Architect Network, a vendor-neutral community of architecture leaders running Fortune 500-scale practices. Some of the leading practitioners in their fields agree that the future of enterprise architecture is driving decision intelligence, and not just documenting it.
We also hosted a number of speakEAsys across Europe and North America. Small tables, candid conversation, no slides. These sessions were invaluable for understanding what keeps our customers up at night around their role and AI.
It was great to also be a part of Gartner IT Symposium/Xpo Gold Coast, alongside our partners Kapish, where more than 1,500 CIOs and IT executives gathered under the theme “Ignite Intelligence.” Consistent with many of the events this year, AI was a key talking point, but we specifically wanted to connect with leaders and show them how they can govern their AI by first gaining visibility and then understanding how it’s actually driving their transformations.
For CIOs, this is the AI governance question in its simplest form: who answers when an agent gets it wrong? The answer has to be a person, working from data they can trace.
AI will automate a lot of what architecture teams are funded to produce today: documentation, first-draft models, portfolio analysis, review preparation. I see that not as a threat to the discipline, but a long overdue correction that lifts tedious work off architects who should be helping drive strategy. The concern comes in when AI delivers that documentation with a confidence that cannot be relied on in place of a skilled architect.
When producing architecture gets cheap, the value moves to deciding with it. Which trade-offs are worth making? Which application should retire first? Which risk should the board hear about today? Those questions need context, accountability, and experience. A model can help you answer them, but it can’t be held accountable for it.
This reiterates the themes we talked about last quarter: trust, independence, and the importance of a context graph. Jason Baragry, our VP of Applied Research and Strategy, showed last year that long-context accuracy decays as chains of facts grow. Retrieval has improved a lot since then, but his compounding-error point still holds for professional reasoning. On independent benchmarks of long professional documents and enterprise workflows, even the best models fully complete about a third of document tasks and about half of strict, multi-step enterprise tasks. The misses are the ones architecture can’t absorb: a requirement buried across sources, a conflicting record resolved silently, a wrong owner or dependency inside a confident answer.
The more work you hand to AI, the more it matters that you can trace what it did and why. A graph gives AI governed entities and relationships to reason over. An ontology fixes what each term means and which connections are valid. Reasoning rubrics define what a good assessment contains. Together they are designed to turn recall into lookup, surface conflicts instead of smoothing them over, and give you a standard to check the work against. (Artificial Analysis, October 2026)
We already launched data access earlier this year, but we knew we needed to do more to make AI really powerful and reliable for EAs and their organization. This is why we built our own AI assistant, engineering it such that we can control what the model sees, so it only gets the context relevant to the question. We pair the language model with a neuro-symbolic approach. The model generates the logic, and the graph executes it deterministically, checking each step against your metamodel so long chains don’t quietly drift from the facts. We also layered in EA expertise, so the Assistant is able to more accurately weigh dependencies and risk the way an architect would.
Every answer links back to its source, which makes it traceable, auditable, and defensible. That is the bar I think architects should hold every AI to, including ours. That is also why we acquired GraphLake: every fact in the graph carries where it came from, who asserted it, when, and we can reason over how your architecture changes over time. Architects are noticing. Weekly active users of Ardoq’s AI have grown around 17x since the Assistant launched. People won’t keep using a tool they can’t trust.
Here is more detail on the weaknesses of generic LLMs and how we’ve built our AI assistant to be accurate, trustworthy, and transparent: How Ardoq’s AI Assistant Outsmarts Generic LLMs
Our mission is to help EAs become strategic advisors that drive the business, and we know a big obstacle to this is the sheer amount of busywork that weighs many EA teams down. AI can be a huge lift for much of this, and Q3 was our most AI-heavy release quarter yet: 52 features, more than half of them AI-powered. The headline is a shift from AI that answers questions to AI that gets work done.
We launched eight new Architecture Agents, each built around a real EA job. The Disaster Recovery Blast Radius Agent traces what breaks when a system fails. The Contract & Document Extraction Agent turns a 40-page contract into structured architecture data. The Application Rationalization Agent pulls together the evidence behind a retire-or-keep decision and flags the contradictions worth a second look. With Custom Agents, teams can build their own for workflows only they have. The Ardoq AI Assistant also got significantly smarter, with visible reasoning and answers linked back to their sources.
The point isn’t the agents themselves. It’s what the architect is now able to do with their time. At NorgesGruppen, Enterprise Architect Mats Kjesrud used the Assistant to build a contingency exercise from the company’s own architecture data. In about two minutes, it produced a scenario that was already 95% complete.
“The Ardoq AI Assistant felt like magic. In a large and complex organization, being able to chat with the data makes things a lot easier, especially when you don’t know all the details of the data model.”
Mats Kjesrud, Enterprise Architect at NorgesGruppen
The agent does the groundwork. The architect applies the judgment. That division of labor is the whole idea. The results hold at enterprise scale. At Tenneco, six AI agents built on Ardoq’s MCP server delivered a 292% ROI and gave back 1.25 FTE of architect time.
If I had to summarize the shift I’ve started to see this year, but more so this quarter, it’s that the role is moving up the value chain in three ways:
This is why the work we did on the context graph and AI Lens, our solution for full visibility, compliance, and control of your AI landscape, matters so much. An accurate, governed map of how the business works is no longer just good practice. It’s what makes it safe to give agents real responsibility. Architects who own that map own a seat at the table. This is the vision we’ve had at Ardoq all along when we pushed for the shift away from pure drawing to data, and now from documentation to decision-making.
Every conversation we had this summer reinforced my conviction that Enterprise Architecture isn’t being automated away. It is being promoted.
Thank you to every architect, customer, and partner who spoke with us this quarter, challenged our thinking, and trusted us with your architecture. In Q4, we are focused on three things: making the Assistant better at the work architects handle, letting teams build and share their own agents, and publishing AI benchmarks so you can see how your practice compares.
Want to see our new agents in action? Watch the Q3 Release Review webinar on demand, or request a demo.