AI continues to be hard. Our research (and that of others) shows clearly that deploying even a single, substantive generative AI application or agent is exceptional.
There are two antipatterns that complicate adoption: first, when engineers go off and build solutions without significant business collaboration; second, when parties do collaborate and teams contribute their own special dysfunction to the inevitable bottlenecks and meeting hell.
To be fair, everyone’s trying their best. AI apps are a legitimate frontier, and being a pioneer is hard. In this light, we share new data: In Forrester’s Developer Survey, 2025, 89% of development executives indicated that their firm is either currently implementing or actively planning a citizen developer strategy.
Low-code platforms, having long proved their value, got us here. Now, trends in AI-assisted software development such as prompt-based vibe coding and emerging application generation platforms make the long-term case for citizen development even more compelling.
But making software development easier is only part of the AI-plus-citizen development story. The movement now has another raison d’être: Citizen development is arguably the most practical strategy for discovering and scaling AI’s business value in the real world.
Democratizing Development Is A Pragmatic Path To Unlocking AI’s Business Value
Large language models (LLMs) and their “applied” forms, such as AI agents, hold significant, unexploited value to digitize and automate many of the “squishy” judgment calls and garden-variety creative tasks that humans do imperfectly — and that traditional software cannot. For most firms, however, unlocking this value in a reasonable time frame requires that three conditions be met:
AI experimentation is scaled to many (hundreds or thousands) of use cases in a given company in the context of its specific processes and opportunities. Many or most will fail, but some will yield significant returns.
These experiments take the form of applications — for example, narrow “agents” to perform one or more actions as part of an orchestrated process — and not just isolated personal productivity tricks such as content generation.
These experiments are led by business domain experts who can imagine what a solution might look like, have the domain knowledge both to direct LLMs (e.g., through prompts or lightweight context engineering) and judge the output in the context of their applications, and can monitor and adjust these applications to ensure their continued effectiveness beyond the janky POC stage.
In this light, serious and scaled citizen development — where businesspeople are systematically empowered (with pragmatic governance) to deliver applications themselves — provides both precedent and an obvious strategic framework for AI-enhanced apps.
Real-World Examples — And Data
Our research shows that empowered citizen developers are indeed successful when experimenting with delivering AI apps and agents. Some examples:
A strategist at a global law firm delivered a database and workflow application that used AI to perform complex legal reasoning required in private-equity contract reviews.
A marketing manager at a Fortune 10 firm delivered an app for managing the process of marketing content production. An LLM now generates copy as part of this process instead of third-party agencies.
A mechanic at a national railroad wrote a mobile railcar inspection app, incorporating AI to analyze railcar photos for maintenance and safety needs and then kick off and manage remedial work orders. Now, the railroad’s data scientists are refining and scaling the mechanic’s work by having AI analyze video feeds from its railyards and kick off remedial actions proactively.
It is only the smallest extrapolation from examples like these to envision hundreds or thousands of ideas for AI applications put into action by systematically empowered domain experts — i.e., citizen developers. Our data supports this vision: In Forrester’s Developer Survey, 2025, when development executives were asked what types of low-code apps their citizen developers are (or will be) allowed to deliver, AI-infused applications topped the list.
Remarks From A Real-World Practitioner
Let’s close with the (lightly censored) remarks of an unusually perceptive citizen developer we interviewed:
“What we need, and what the business world needs, is an easy way to deploy capabilities against specific problems. That’s one of the key ways we’re using low-code … This never would have happened if I had to direct an engineer. It would have been like trying to direct a movie through the big end of the telescope … Citizen development is a compression of the development process. I view it as allowing the expert to get ever closer to the result. That compression makes way better products because the expert is able to create the feature themselves without explaining it to five different people … Why does that matter? Because velocity is the ****ing strategy.”
