Maybe Sgt. Curtis Culin is the prototype for citizen developers? He innovated in theater and jerry-rigged a way for tanks to break through the hedgerows that bottled Allied forces up in the Normandy beachhead. However, he was a complement to, not a replacement for, the labs, factories and headquarters that developed and orchestrated the technologies and forces required for victory.
Likewise corporate success in the AI age will require citizen development and traditional, at-scale enterprise technology to coexist. Yes, CIOs must build platforms citizen developers can use. But they also should inspire citizen developers to think, at least a little bit, like software engineers. That makes it more likely that citizen developers will build applications that the technology organizations can co-opt and scale.
Citizen development at Normandy
More than 900 officers and other ranks worked for General Eisenhower in COSSAC (Chief of Staff to the Supreme Allied Commander) coordinating thousands of other military and civilian personnel in planning Operation Overlord. Staff studied Normandy from aerial photographs and pre-war maps and yet, as John Keegan recounts in Six Armies in Normandy, they failed to plan for the 15-foot-tall earthen hedgerows that would stymie the advance of invasion forces. Staff knew about the hedgerows, but they so focused on getting onto and then past the beaches, they didn’t think enough about what would happen next.
Culin attached metal prongs to a tank so that it could breach the hedgerows. A demonstration of the Culin Rhino impressed Gen. Omar Bradley, who ordered the First Army Ordnance Section to produce it at scale. Decades later in Iraq, US soldiers innovated in similar fashion. When Humvees proved vulnerable to improvised explosive devices they found ways to augment them with improvised “hillbilly armor” in theater.
You can see the skeptics of big IT smiling at each example. Innovation happens at the front lines, they will say. Nobody in an enabling function will understand the business well enough to automate it. I’ve heard plenty of people say that citizen development will cause the IT shop to wither away, no less than Engels believed that the state would wither away under Communism. Neither has happened yet.
Citizen development and at-scale technology will coexist in large institutions
Sgt. Culin did not and could not have coordinated the 160,000 troops, 7,000 ships and landing craft and 12,000 airplanes deployed on just the first day of the invasion. Nor could troopers in a maintenance depot in Iraq have realized the Second Offset, which brought stealth platforms and precision-guided munitions into reality.
Citizen development and at-scale technology will coexist in large institutions. Citizen development will push and reshape enterprise technology no less than the desktop revolution did in the 1980s and 1990s. We graybeards remember that ambitious users -- often with equipment funded in departmental budgets -- pushed glass house IT to adopt new technologies and consider new types of applications. Sometimes companies deployed applications initially developed for a few users widely -- when I worked at Deloitte Consulting in the 1990s, everyone entered time and expenses into a database an associate had thrown together in Clipper and created charts using an Excel plug-in some BA had built.
No, the analyst on the front-lines will not vibe-code your next ERP platform, but he or she has local knowledge that many requirements processes will struggle to capture. How can enterprise technology unleash creative talent close to the front-line -- in a safe, scalable manner? Last week I talked about how vibe coding will require platforms to enforce architectural standards. But that’s not enough -- it will also require skills and, especially, mindsets on the part of vibe coders (or citizen developers).
Inspire citizen developers to understand the power they have
Yes, citizen developers will need to learn the protocols for the platforms they use, but skills have boundaries. Mindsets compound. Every developer learns syntax. Good ones combine curiosity with intellectual precision -- they have a desire to move past the fuzziness that dominates corporate life, understand how things work and how to make them better.
We want citizen developers to ask not only “How do I do X?” but also “What do I mean by X? Why can’t I do X now? And: What is the best way I might do X?” We want them to learn to think a little like software engineers.
Maybe not a small lift, but no important one is. Can CIOs and CTOs inspire citizen developers not only to do, but also to ask and to learn? If I were to write a letter to a vibe coder, here’s what it might say:
Automate what you understand. Years ago, someone asked, a great user of yellow pads, “What can a computer do?” I stand by my reply: only what you can do on a yellow legal pad, only much more quickly. If you can’t sketch it out on a whiteboard, graph paper, in a spiral notebook or on legal pad, you have no business writing a program about it.
Models aren’t magic. They just perform a statistical prediction of what tokens to use in response to the tokens you provide. A model is only as good as the context it receives. You have to figure out not only what type of context you can provide, but also how accurate an answer that context will yield.
Agents need to write things down, just as people do. We’ve all stared into a chat window and wondered why the machine can’t remember something. Context windows are limited and attention degrades. Almost every compute architecture faces memory bottlenecks. LLMs handle large inputs imperfectly, suffering from attention degradation over long context windows. Like a human colleague who forgets verbal directives, force your coding agents to write state and requirements down into explicit artifacts like Product Requirements Documents (PRDs).
Many things must be deterministic. The uncertainty and variety inherent in probabilistic execution make for great brainstorming, but can create unacceptable risk in business logic. Use models to write deterministic code (Python, SQL) that you can isolate, test, and run reliably, rather than relying on an LLM to interpret business logic on the fly.
Sometimes models are just as superficial as humans. Training creates incentives for a model to deliver an answer, not the best answer. Just as with a new joiner, you have to probe a coding agent: Did you consider all possible causes for a bug? Is this the most robust and flexible possible solution?
Don’t be afraid to co-opt what citizen developers build
For CIOs and CTOs watching an impending onslaught of citizen development: be ready to co-opt the best of it. Gen. Bradley didn’t penalize Culin for modifying armor; he ordered the Rhino cutter into scaled production. CIOs in the 1990s cherry-picked the best user-developed desktop databases and hardened them for the enterprise. Treat citizen development as a finely-tuned sensor array for organizational friction—and an opportunity to scale what works.
