Process optimization consultant Frederick Van Brabant published a blog post arguing that AI will not accelerate business processes like software development unless organizations first address upstream bottlenecks. Van Brabant, who re-read classics like The Toyota Way and The Goal, contends that AI-generated code may produce output quickly but fails to resolve the core issue: unclear or incomplete feature requirements that slow down development.
Van Brabant uses a Gantt chart to illustrate a typical software project timeline, where the development phase spans 70 days—far longer than scoping or deployment. He notes that organizations often respond to such delays by adding more developers or assuming AI will automate the work. However, he argues this approach ignores the root cause: software development is not about typing speed but translating vague requirements into precise, executable solutions. Without resolving ambiguities in feature requests, AI-generated code risks producing incorrect or incomplete outputs.
The blog post highlights a common misconception about AI in software development: that it can bypass the need for detailed scoping. Van Brabant presents a revised Gantt chart where AI reduces the development phase to just 3 days, but he warns this is unrealistic. AI-generated code still requires human oversight to ensure accuracy, particularly when interpreting ambiguous requirements like 'send mail to user once sale is completed.' Questions about error handling, content, and timing remain unanswered, demonstrating that AI cannot replace the need for clear upstream documentation.
Van Brabant references The Toyota Way and The Goal, two foundational texts on process optimization, to underscore his argument. He notes that these books emphasize identifying bottlenecks rather than applying quick fixes like adding resources or adopting new tools. In software development, the bottleneck is often the lack of clarity in feature requests, which AI cannot resolve independently. The post suggests that organizations must first improve their scoping processes before expecting AI to deliver meaningful speed improvements.
The author critiques the assumption that AI can act as a project manager for software development. He argues that AI-generated code requires significant 'handholding'—human intervention to refine inputs, validate outputs, and correct errors. This undermines the idea that AI can autonomously accelerate development timelines. Instead, Van Brabant suggests that organizations should focus on improving communication between domain experts and developers to reduce ambiguity in feature requests, which would naturally shorten development cycles.
Van Brabant’s post includes a hypothetical scenario where a feature request like 'send mail to user once sale is completed' lacks critical details. He asks: What should the email contain? Should it be sent if the sales process fails? When exactly is a sale considered complete? These questions highlight the limitations of AI in interpreting vague instructions. Without addressing such ambiguities, AI-generated code may introduce new inefficiencies, such as requiring additional rounds of review and revision to align with business needs.
The blog post also addresses the broader trend of organizations turning to AI for process optimization during market downturns. Van Brabant observes that many companies adopt AI as a silver bullet without first analyzing their existing workflows. He warns that this approach is akin to 'throwing people at the problem'—a tactic that rarely yields sustainable improvements. Instead, he advocates for a return to process fundamentals, such as those outlined in The Toyota Way, which prioritize identifying and resolving bottlenecks before introducing new tools.
Van Brabant concludes by emphasizing that AI is not a substitute for well-defined processes. He argues that while AI can generate code quickly, it cannot replace the human expertise required to scope and refine feature requests. Organizations that expect AI to accelerate development without addressing upstream bottlenecks are likely to face disappointment. The post serves as a reminder that process optimization requires a holistic approach, combining clear communication, iterative feedback, and a focus on root causes rather than superficial fixes.