USA, 12th May 2026 — Generative AI has moved faster than the enterprise has been able to govern it. What began as a wave of productivity tooling has quietly evolved into something more consequential: autonomous agentic systems capable of generating code at a speed that outpaces any organisation’s ability to validate it. The productivity gains are real, but so is the structural risk accumulating beneath the surface.
It was a question that demanded the right room. Scarpetta’s Library Room in the heart of NoMad; warm-lit, intimate, and tucked away from the noise of the city, offered precisely that. There is something fitting about discussing the architecture of trust and technical governance in a space that is itself defined by quiet authority. The setting did not merely host the conversation, it shaped it.
THE BRIEF
When productivity becomes a liability: governing AI at speed
In partnership with JetBrains, The Ortus Club convened a closed roundtable dinner in New York City to bring this conversation into the open. Entitled “Stopping Shadow Tech Debt: Responsible AI Adoption at the Leadership Level”, the evening gathered senior technical and operational leaders to discuss one of the most pressing and underacknowledged challenges in enterprise technology today: how to embrace AI-driven development without creating the agility-killing debt that will define the next decade of remediation work.
The aim was to move beyond the headlines. With AI adoption accelerating faster than governance frameworks can follow, the goal was to create space for the people actually accountable for these decisions to speak plainly, to surface what is working, what is not, and what the enterprise needs to get right before the window to act closes.
Candour candlelight: an evening in NoMad
Guests arrived at Scarpetta, tucked into the NoMad district at 88 Madison Avenue, as the city settled into its Tuesday evening pace. The choice of venue was deliberate in spirit; the restaurant’s name is drawn from the Italian expression “fare la scarpetta”, meaning to savour a meal down to its very last bite: a philosophy that suited an evening designed around depth of conversation rather than breadth of attendance.
The evening was held in Scarpetta’s Library Room: a lower-level private dining space with a capacity that keeps the gathering genuinely intimate, warm lighting that encourages ease, and the kind of unhurried atmosphere that allows a room full of senior leaders to stop performing and start talking. Scarpetta has always been at once informal, comfortable, and elegant, grounded in a passion for old-world Italian hospitality.
The people who have already made the decision
The value of any Ortus forum is inseparable from the people at the table, and this evening set a high bar. The guest list brought together engineering, transformation, and technology leadership from some of the most consequential organisations in global finance, technology, and enterprise services. Decision-makers navigating it from within, these individuals were not observers of the AI adoption wave. The table was filled with leaders from:
Collectively, they represented institutions managing some of the most complex and highest-stakes codebases in the world, specifically the organisations where the cost of unvalidated AI-generated code is not hypothetical.
What made this particular combination of leaders valuable was not simply their seniority, but the range of pressure points they each brought to the table. From financial services firms grappling with regulatory exposure on AI-generated code, to technology organisations managing the pace of developer tooling adoption across distributed engineering teams, every seat represented a different vantage point on the same structural challenge. That breadth of perspective across verticals, functions, and organisational scales is what transforms a dinner conversation into genuine knowledge exchange.
Structure that serves the conversation
The Library Room’s warm, enclosed atmosphere reinforced the Chatham House spirit: a space where titles faded into the background and frank exchange took precedence. Dimmed lighting, close seating, and the unhurried pace of a well-crafted Italian dinner all contributed to a setting that felt less like a corporate event and more like the kind of conversation executives rarely get to have. That structure was not incidental.
The evening concluded with Scarpetta’s Valrhona Chocolate Cake: rich, considered, and quietly memorable. Much like the evening itself.
Gartner’s prediction that the shift toward “prompt-to-app” development will produce a 2,500% increase in software defects by 2028 is not a forecast to be dismissed. It is a signal that the verification debt accumulating inside enterprise codebases today is not a peripheral concern, it is a structural one.
For the leaders in the room, the question was not whether to adopt AI. That decision has already been made, at scale. The question was how to govern it. The discussion was anchored by four lines of inquiry:
Where speed meets structural risks
The conversation surfaced a consistent tension at the heart of enterprise AI adoption: speed versus soundness. AI tooling delivers immediate, visible productivity gains. And those gains are politically difficult to slow down, even when engineering leaders can see the risks accumulating downstream.
On tool sprawl and shadow AI, the group observed that much of the debt being created today is not the result of deliberate decisions, but of decisions not being made. Individual teams adopting point solutions without central visibility creates a fragmented codebase that no single leader fully understands. The consensus was that governance frameworks need to move faster than the adoption they are trying to shape, and that waiting for an incident to trigger the conversation is a strategy that has already failed for several organisations in the room.
On metrics and codebase visibility, leaders acknowledged a genuine gap. Traditional engineering metrics were not designed to surface the structural risks introduced by AI-generated code at volume. New approaches, including architectural integrity scoring and AI-attribution tagging within codebases were discussed as early-stage responses to a problem that does not yet have a standard playbook.
On scaling beyond the pilot, the room was pragmatic. Many organisations represented had run successful AI pilots. Far fewer had successfully expanded those pilots into production-grade programmes with the governance infrastructure to match. The difference, in most cases, came down to whether engineering leadership was involved early enough or whether AI adoption had been driven by business urgency and handed to engineering to operationalise after the fact.
The broader conclusion was clear: responsible AI adoption is an architectural discipline, not a policy document. The organisations best positioned to benefit from AI-driven development are those that have invested in making their systems legible to themselves and to the tools operating within them.
1. Governance must precede scale, not follow it
The organisations that will avoid the worst of shadow tech debt are those building architectural controls before adoption reaches critical mass, not responding to the consequences of adoption that has already happened.
2. New metrics for a new problem
Traditional engineering health metrics are not calibrated for AI-generated code at volume. Technical leaders need new frameworks for assessing structural integrity, and the conversation about what those frameworks should look like is still in its early stages.
3. Shadow AI is a leadership visibility problem
Much of the risk accumulating in enterprise codebases today is not the result of reckless decisions; it is the result of decisions being made below the threshold of leadership awareness. Closing that visibility gap is as much an organisational challenge as a technical one.
A conversation that earns its place in the calendar
The roundtable at Scarpetta brought together a room of senior technical and transformation leaders who are managing AI adoption challenges at an institutional scale that few organisations have navigated before. The quality of the dialogue reflected that — candid, specific, and oriented toward the practical realities of governance at speed.
For JetBrains, the evening reinforced their position at the intersection of developer productivity and enterprise code quality — a space that is only growing in strategic relevance. For The Ortus Club, it was further evidence that the most important conversations in enterprise technology are not happening at conferences. They are happening in rooms like this one.
The challenge of shadow tech debt is not going away. But the leaders who will manage it best are already asking the right questions, and, on evenings like this, asking them together.
Our Feasibility Studies show exactly who we can reach and how we can engage them. Using your brief, we analyse roles, regions, industries, and company size to deliver a clear, data-driven report that highlights your audience potential.
