Boards are debating the wrong question. When ‘how will AGI affect us?’ makes it onto the agenda, the room predictably splits between two tired frames: existential threat or limitless opportunity. Both miss the point. The more useful question is operational: when an AI agent currently running inside the company makes a consequential error, who is accountable, through which channel, and within what timeframe? Any board that cannot answer this has not earned the right to discuss AGI. The governance infrastructure for today’s bounded agents is the prerequisite — not a side task — for managing whatever comes next. Discussing AGI without that foundation is like designing a penthouse before laying the ground floor.Consider a regional private insurance company in Ankara, employing 312 people, with a portfolio concentrated in SME workplace insurance. In the second half of 2025, the company deployed a conversational agent to handle policy renewal reminders, claims reporting guidance, and document checklist verification. Over the first quarter, measurable efficiency gains came through. The board saw the numbers and, in the following quarter, approved extending the agent’s role to assist with contract pricing recommendations. The decision moved quickly — but no committee discussed which data layer was feeding the pricing component, which employee’s name would appear on a disputed recommendation, or what the customer could do if the output led to a harmful outcome. That gap is a governance failure, not an AGI problem. It’s a problem with the agent running today.AGI debates matter to boards precisely because how an institution responds to AI now will shape its decision-making architecture for the next five years. This doesn’t require deep technical fluency — it requires one foundational governance question: ‘Who can challenge a decision made by this system, through what channel, and how quickly?’ Companies with a clear answer to that question approach the EU AI Act’s high-risk system categories with considerably less disruption. The regulation, adopted in April 2024 and entering phased enforcement from 2025, imposes mandatory human oversight, documentation standards, and accountability chains for high-risk applications. Turkish companies that export to EU markets or operate through European partners are not insulated from these requirements. The board’s real AI agenda should be built around this compliance reality, not around speculative AGI timelines.What does the board actually need to do? Not conceptual preparation — operational structure. The first step is a system inventory: a documented list of every active AI system, what decisions it influences, what data it consumes, and which employee’s role the output is ultimately tied to. A representative scenario: a medical device distributor in Izmir with 378 employees completed this inventory and discovered that its order forecasting model had been influencing logistics decisions for 14 months without a single update cycle. The model was assumed to be performing well because no one had established how performance would be measured. The second step is assigning a named accountability owner to each system — not a technical role, but an operational one: the person who links the system’s output to a business decision, notices when something goes wrong, and knows how to report it. The third step is measurement: tracking AI system performance not through success stories but through error rate, frequency of human override, and usage volume on the challenge channel. Until these three steps are in place, no board is ready to discuss AGI seriously.A legitimate objection deserves space here: ‘If AGI genuinely arrives, the current governance framework will be obsolete anyway — a fundamentally different structure will be needed.’ That’s partly true. If a general-purpose system becomes genuinely autonomous across an open domain and operates at a speed that renders human oversight meaningless, today’s accountability models won’t hold. But even in that scenario, the starting point is identical: an institution that cannot manage a narrowly scoped agent today will not manage an unconstrained system tomorrow. Governance capacity is a muscle built through repetition with simpler systems. And for most of Turkey’s business community, AGI remains speculative — the concrete challenge now is agentic chains, where one agent’s output feeds directly into another agent’s input, and errors cascade invisibly across a workflow before any human notices.Turkey’s unresolved data sovereignty discussion adds a specific dimension boards must own. Which cloud provider processes the data feeding your AI systems? In which jurisdiction does it sit? Is that jurisdiction subject to a regulator who can compel disclosure? These aren’t abstract questions. Turkish companies using EU-based service providers are already receiving compliance inquiries touching both the KVKK and EU AI Act requirements simultaneously. An insurance company routing customer claims data through a US-hosted LLM API, for instance, faces a risk map that requires legal, technical, and strategic input — the kind that can only be coordinated at board level. This is not a technical preference to be delegated to IT; it is a governance decision about where the company’s data boundary sits.AGI belongs on the board agenda, but only in its proper position. Starting with that Ankara insurance company and its pricing agent: when that agent quietly shapes a recommendation sent to a customer, and no accountability line has been drawn, the board is already facing a risk far closer than AGI. The productive board question is not ‘when will AGI arrive?’ but ‘how many of our current AI systems have documented oversight protocols that we review quarterly?’ A board that can answer that question confidently is not caught off guard by the next capability jump. The governance architecture built on today’s bounded agents is exactly what allows the institution to absorb tomorrow’s more powerful ones without losing control.
This article was originally published in Turkish by Gökhan MERCANOĞLU on July 15, 2026. The English edition has been reviewed and edited by the author.