Dopamine-fueled IT and the Gordian Knot

When to build, when to adopt, and how to keep complexity from closing back in.

At a recent gathering of CIOs from peer institutions, a twice-yearly routine where we compare notes and reflect on shared challenges, a colleague shared how their team was using GenAI. To move fast and respond to emerging customer requests, they’d established a new group within IT, separate from their ERP and data and decision-support teams, focused entirely on delivering AI-generated solutions, using techniques sometimes referred to as “vibe-coding.”

This team was using models like Claude, guided by prompts developed in ChatGPT, to generate custom programming to deliver ERP automations, reports, and analytical tools. The strategy, my colleague explained, allowed them to create highly tailored solutions quickly, without relying on vendor-delivered tools. It was efficient, cost-conscious, and, in their view, an ingenious use of code-writing that GenAI now makes possible.

From my facial expression, though, my friend could tell I had strong reservations. I’ve seen this movie before, and I know where it all leads, and it’s nowhere good.

In today’s Dispatch, I explore why the rush to build, even with the best of intentions and tools, often takes us back to a world we thought we’d outgrown. A world where speed and precision come at the cost of sustainability and resilience, and where every new line of custom code quietly ties another strand into Andy Kyte’s Gordian knot.

The big picture

There’s a thrill in writing code, by hand or with GenAI. It scratches the problem-solving itch, delivers instant feedback, and rewards us with the sense of progress. For technologists, it’s creative. For early adopters, it signals capability. For stakeholders, it feels like momentum.

But that feeling is misleading. When we equate code with value, we risk repeating a pattern we should have outgrown: building bespoke solutions that feel empowering in the moment but erode resilience over time. Locally maintained platforms, one-off scripts, and custom automations layered across ERP systems promise agility, but deliver fragility. They create maintenance headaches, security exposures, and brittle systems that no one wants to inherit.

GenAI supercharges this dynamic. The tools are fast, the results immediate: solutions appear in minutes, tailored to our needs. But that speed enables vibe coding: a dopamine-fueled rush to build without guardrails, testing, or alignment. It’s software by instinct, not architecture. And when the platform evolves, the process changes, or the author walks away, the whole thing likely collapses under its own weight.

From a resilience standpoint, vibe coding is a false economy. It creates local wins and institutional debt. Every line of ungoverned custom code adds to Andy Kyte’s Gordian knot. None of it cuts through.

The False Confidence of Custom-coded Solutions

The first era of enterprise information systems was defined by a “build-it” mindset. Meeting users, coding, and deploying before dinner was thrilling, fast, personal, and deeply satisfying. Custom applications mirrored local business processes, delivering unmatched functional fit.

But that speed and precision came at a cost, one we only fully grasped later. Over time, the trade-offs of custom systems became inescapable, and ultimately led us toward vendor-supported platforms. Here’s what is sacrificed.

• Maintainability: These systems often depend on one or two individuals. When they leave, knowledge disappears and technical debt compounds.

• Compatibility: Custom solutions rarely align with evolving platforms. What fits today becomes friction tomorrow.

• Efficiency: Rebuilding what vendors can provide drains resources. The control feels good, but it hides duplication and unnecessary complexity.

• Portability: Code tied to a specific environment locks you in. Modernizing or migrating becomes costly and difficult to justify.

• Usability: Developer-built solutions reflect internal logic, not user needs. They lack the polish and intuitiveness of mature commercial products.

• Reliability: Unlike vendor platforms tested at scale, custom systems are fragile. Breakdowns tend to happen where it hurts most: live, exposed, and unsupported.

GenAI tempts us to return to the build-it era, offering quick wins with less friction. We left that world not for lack of excitement, but for lack of sustainability. GenAI removes the barriers to building, but we need discipline to keep from adding even more complexity to the Gordian Knot that Andy Kyte rightly railed against.

The Role of Custom Code in an AI-Infused University

Becoming an AI-infused University means responsible AI is woven throughout the tapestry of the institution’s academic and administrative processes, best understood as four distinct use cases: game-changing AI and everyday AI, each applied either to internal capabilities or to customer-facing experiences. This framework clarifies when custom code is helpful and when it becomes an expensive detour.

• For game-changing AI focused inward, custom coding remains essential. This is the domain of faculty research, scientific discovery, and breakthrough innovation. It’s experimental by nature, and often risky. That’s the point. In these environments, coding, iteration, and bespoke AI development aren’t just tolerated, they’re foundational to the mission of advancing knowledge.

• For game-changing AI focused on customers, the priority is process automation and work simplification. Modern ERP platforms like Oracle, Workday, and Ellucian offer embedded tools and low-code options, while a growing number of mature third-party platforms, many with AI capabilities, can layer across these systems to streamline workflows and improve student services. Custom code should be rare, limited to system integration via vendor APIs. Looking ahead, Agentic AI will unlock even greater potential to optimize simplified processes and reduce administrative costs, if we keep complexity in check.

• When it comes to everyday AI, whether enhancing faculty and staff productivity or powering student-facing chatbots, customization is both possible and appropriate through the low-code capabilities built into platforms like Microsoft Copilot, Google Gemini, and ChatGPT. The institutional focus should be on responsible enablement and thoughtful adoption, not ground-up coding.

This is what maturity looks like in an AI-infused university: the ability to thoughtfully conceptualize where AI belongs, reducing risk and maximizing return on investment. It means knowing when custom solutions are warranted and when to rely on low-code, vendor supplied solutions. Innovation isn’t about building everything yourself, it’s about applying judgment, context, and clarity to what’s already possible.

The bottom line

Leadership in the AI era is not about how fast we can build, but how clearly we can see. In an environment where GenAI makes creation frictionless, the real challenge is exercising restraint. The institutions that thrive won’t be the ones that code the most, but the ones that think the clearest about where code belongs and where it doesn’t.

The AI-infused university will succeed not by abandoning custom code entirely, but by sharply defining its role. When used to power research breakthroughs or strategically integrate systems, custom code can serve the mission. But when used indiscriminately, layered on top of ERP systems, duplicated across departments, or written to recreate what platforms already offer, it only tightens the Gordian Knot Andy Kyte warned us about: a tangle of technical debt, shadow systems, and fragility that holds our institutions back.

Here’s the hard truth: Business leaders who chase quick, local wins without alignment to enterprise strategy aren’t accelerating progress; they’re engineering failure on a delay. What feels like success is an illusion. Sooner or later, the whole house of cards will collapse. By then, the executives who demanded it and the heroes who built it typically have moved on. The fallout will land on the IT organization, the very team bypassed in the rush to build, but left holding the bag when it fails.

This is playtime masquerading as innovation, and CIOs should shut it down.

Restraint isn’t hesitation. It is strategy. In this moment of possibility, the most important leadership quality is being willing to speak hard truths by deciding when not to build. That choice may be the most transformative decision CIOs can make.