Frederick Van Brabant, CTO of Belgian software consultancy Oswald, argues in a 15 May 2026 post on frederickvanbrabant.com that generative AI will not accelerate corporate processes and warns that current optimisation projects are misreading both lean theory and the technology’s limits.
Van Brabant revisited two canonical texts—*The Toyota Way* and *The Goal*—while auditing client workflows that had recently added large-language-model copilots. He found that teams treated AI as a faster typist rather than addressing systemic bottlenecks such as sequential legal reviews and budget approvals. A Gantt chart in the post shows a typical 53-day feature cycle where exploration and documentation tasks still wait on upstream sign-offs; inserting AI code generation merely compresses the 25-day development slice without shortening the critical path.
The critique lands as enterprises pour capital into AI productivity suites. McKinsey’s 2025 survey cited by Van Brabant estimates $18 billion in annual spend on “AI-driven process acceleration”, yet median cycle time reductions remain below 8 %. Comparable roll-outs at Siemens and ING, referenced from their respective 2025 annual reports, produced similar plateaus once human checkpoints were reinstated for compliance. Van Brabant contends that ignoring constraint theory turns AI into an expensive parallel lane that still feeds the same bottleneck.
Oswald will publish a follow-up white paper on 30 June 2026 benchmarking ten client teams that removed at least one non-value-adding approval stage before deploying AI coding assistants. Early data, previewed in the post, suggest cumulative lead-time cuts of 22 % when the constraint is addressed first. Van Brabant plans to open-source the anonymised dataset under a Creative Commons licence on GitHub the same day.