On September 16, Forrester published The AI Enterprise Architect, followed two days later by a public essay from Charles Betz and Joseph Schiavone titled “Will AI Eliminate Enterprise Architects?” Their answer is direct: enterprise architecture is becoming more important, not less, but the basis of that importance is changing.
Forrester’s point is economic as much as technical. AI can already generate diagrams, draft standards, analyze portfolios, and answer questions about large technology estates. Quality is still uneven, but work that once took weeks can increasingly take hours. When artifact production stops being scarce, value moves to enterprise context, judgment, and accountability. Forrester’s phrase for the new job is “the control plane for bounded autonomy.” Architects become curators of context, designers of governance, and advisors on high-consequence decisions, helping build a reusable intelligence layer of policies, standards, decision records, and business context that both people and machines can consume.
Practitioners who responded in Architecture & Governance Magazine on September 23 largely agreed, with a useful caution: AI-generated documentation still needs human verification, and periodic review boards will be too slow once agents are acting in delivery and operations.
Gartner is pointing in the same direction. Its October 7 note on 2026 priorities for heads of EA, Leading AI-Driven Transformation, says they must define an ambitious AI future, govern use cases and technology choices, enable repeatable pilots, and design safe, scalable agentic systems. A public Gartner briefing is more blunt: unstructured AI adoption generates chaos, not strategic value, and architecture leaders should replace slow committee approvals with adaptive guardrails. Gartner reports that 80% of CEOs expect AI implementation to lead to significant change in their organization’s business capabilities by the end of 2026, and that 66% of heads of EA say scaling AI use is a critical priority this year. They also expect 30% of EA activities to be AI-augmented by 2027.
A week later, on October 15, Gartner published Build the Data and Guardrails Now for the Autonomous Operating Model EA Leaders Expect. The abstract is a warning: EA leaders anticipate autonomous operations by 2032, but that future depends on agent-ready data, encoded guardrails, and human accountability for each agent-run process — which most organizations lack.
This is landing on EA because agents do not stay in a lab. In an August 30 essay, Forrester analysts Mike Gualtieri, David Mooter, and Rowan Curran argued that agentic AI is adding a new layer to the enterprise stack, and that today’s architectures were built for request/response integration and human-paced work. Autonomy sits on a spectrum, agents will arrive from many vendors, core systems often lack the interfaces agents need, and governance buried inside individual agents will not scale. Their proposed response is Forrester’s Agentic Runtime Architecture, with shared context, governed tools, orchestration, observability, and policy enforcement at execution time. “No action executes ungoverned” is the design rule.
That point showed up again this month. On November 4, Kong expanded AI management in Kong Konnect for discovery, identity, credentials, cost, and observability as architectures move from model calls to agent-driven workflows. Three days later, Jeff Smith, writing from DevRev’s Office of the CTO, argued that enterprise memory, context, and an agent’s authority should stay independent of whichever model is executing the task, because the model will change. If agents are going to act across the estate, we need a governed connectivity and policy layer. If models are going to turn over, that layer cannot live inside one vendor’s model or one application’s prompt library.
The inventory problem is already real. In a May 13 post, SAP LeanIX wrote that more than 180 organizations had already discovered over 180,000 agents in production through SAP AI Agent Hub, which covers agents, MCP servers, and LLMs across vendors. Treat that as a vendor-reported figure, but it explains why “we’ll catalog agents later” is not a plan.
In an October 19 commentary on the 2026 Gartner Magic Quadrant for Enterprise Architecture Tools, published October 13, Ardoq quoted Gartner as saying the global rush to deploy AI has “rocketed AI governance to the top of the executive agenda,” and that practitioners often struggle with data quality and the opaque nature of AI insight when they try to rely on those outputs for critical decisions. If AI is going to draft models and answers about the landscape, the repository has to be current and the answers traceable.
On September 24, Ardoq announced eight specialist Architecture Agents for work such as blast-radius analysis, data quality checks, stakeholder maps, rationalization, and contract extraction. They run against Ardoq’s knowledge graph and hand architects a first draft, not a final decision. SAP LeanIX has been on a parallel path. Its Enterprise Architecture Assistant, generally available as of June 23, is an in-workspace chat interface grounded in fact sheets, the decision log, and connected sources. LeanIX also offers an MCP server so external assistants can reach the same inventory through a governed protocol. The pattern is consistent: make architecture knowledge machine-readable, keep permissions in place, and use AI to remove the data-gathering tax in front of a decision.
The research and the product releases suggest a practical sequence.
First, treat the architecture repository as a system of record for machines as well as people. Invest in freshness, ownership, and source attribution before you invest in more impressive prompts.
Second, inventory agents the way we inventory applications: owner, systems and data, models and tools, the process they sit in, and whether a human must approve high-risk actions. If you cannot produce that list, you do not have a governance model.
Third, move guardrails out of slide decks and into runtime — policy, identity, entitlements, and observability that intercept tool calls and outputs, with a clear split between what an agent may do alone and what requires a person.
Fourth, keep context and authority independent of the model, so a model change does not force you to rebuild the business process or the decision rights.
Fifth, let AI take first drafts of diagrams, extracts, and portfolio summaries, then spend the hours you get back on decision rights, risk trade-offs, and advice that requires enterprise understanding. Verify the drafts. Uneven quality is why the architect is still in the loop.
That is the through-line I will keep returning to in this series: AI is changing the economics of architecture work, and it is changing the systems we have to design. The practitioners who come through this well will be the ones who stop defending the artifact and start running the control plane.
Charles Betz, Joseph Schiavone, and JT Thykattil, “The AI Enterprise Architect,” Forrester, September 16, 2026. https://www.forrester.com/report/the-ai-enterprise-architect/RES201854
Charles Betz and Joseph Schiavone, “Will AI Eliminate Enterprise Architects?,” Forrester, September 18, 2026. https://www.forrester.com/blogs/will-ai-eliminate-enterprise-architects/
Holt Hackney, “Forrester Posits, ‘Will AI Eliminate Enterprise Architects?’ Experts Chime In,” Architecture & Governance Magazine, September 23, 2026. https://www.architectureandgovernance.com/uncategorized/forrester-posits-will-ai-eliminate-enterprise-architects-experts-chime-in/
Lucas Kobat and Gilbert van der Heiden, “2026 Heads of Enterprise Architecture Priority: Leading AI-Driven Transformation,” Gartner, last updated October 7, 2026. https://www.gartner.com/en/documents/8131929
Gartner, “Start Leading: Scale Enterprise AI Value Now” (public briefing for heads of enterprise architecture), 2026. https://emt.gartnerweb.com/ngw/globalassets/en/enterprise-architecture/documents/lead-ai-driven-business-tech-transformation.pdf
Alana Nolan, “Build the Data and Guardrails Now for the Autonomous Operating Model EA Leaders Expect,” Gartner, October 15, 2026. https://www.gartner.com/en/documents/8515077
Mike Gualtieri, David Mooter, and Rowan Curran, “Enterprise Architects Must Choose: Agentic Runtime Architecture Or Agent Sprawl,” Forrester, August 30, 2026. https://www.forrester.com/blogs/enterprise-architects-must-choose-agentic-runtime-architecture-or-agent-sprawl/
Kong Inc., “Kong Expands AI Management to Help Enterprises Govern and Operate Agentic AI,” November 4, 2026. https://konghq.com/company/press-room/press-release/kong-expands-ai-management-to-help-enterprises-govern-and-operate-agentic-ai
Jeff Smith, “Your AI model will change. Will your enterprise architecture survive it?,” Data Centre Solutions, November 7, 2026. https://datacentre.solutions/blogs/24730-your-ai-model-will-change-will-your-enterprise-architecture-survive-it
Dominik Rose, “How SAP LeanIX is shaping the future of AI-native enterprise architecture,” SAP LeanIX, May 13, 2026. https://www.leanix.net/en/blog/leanix-at-sapphire
Austin Steinmetz, Rowan Wakeford, and Rhyan Light, “Gartner Magic Quadrant for Enterprise Architecture Tools,” Gartner, October 13, 2026. https://www.gartner.com/en/documents/8502677
Deborah Theseira, “Ardoq Named a Leader in the 2026 Gartner Magic Quadrant for EA Tools,” Ardoq, October 19, 2026. https://www.ardoq.com/blog/leader-gartner-magic-quadrant-ea-2026
Ashima Bhatt, “Hand Off Hours of Work to Ardoq’s 8 New Architecture Agents,” Ardoq, September 24, 2026. https://www.ardoq.com/blog/8-new-architecture-agents
SAP LeanIX, “AI Enterprise Architecture Assistant: now included in SAP LeanIX,” June 23, 2026. https://www.leanix.net/en/blog/ea-ai-assistant