Recently on Hacker News, an article titled “I Don’t Think AI Will Make Your Process Faster” earned 557 points and 385 comments. That level of engagement alone signals something important: this topic struck a nerve for many.
Author Frederick Van Brabant reread two classics of process optimization—The Toyota Way and The Goal—and arrived at a counterintuitive conclusion: AI won’t make your project deliveries faster, because the true bottleneck is never in development itself—it’s upstream: in requirement definition, scope confirmation, legal review, and documentation writing.
His logic is clear: suppose building a feature takes 70 days, and upstream requirements work takes 23 days. If AI compresses development from 70 to 3 days, total cycle time drops from 93 to just 26 days. Sounds great—but only if those 23 days of requirements work are high enough quality that AI doesn’t need to repeatedly ask, “What exactly does this feature mean?”
The problem? That’s nearly impossible.
AI Magnifies the Cost of “Fuzzy Requirements”
Every programmer has endured this: a product manager says, “Build a user recommendation feature,” and you start coding—only to realize halfway through that you don’t know how the recommendation algorithm should work, how the UI should be laid out, or where the data comes from. You pause, ask for clarification, wait for a reply, resume coding, ask again, wait again.
AI amplifies this pain tenfold.
Because AI isn’t human. When you give a human developer a fuzzy requirement, they can fill in gaps using experience and intuition. AI has no such intuition. It either follows your instructions literally (producing completely wrong output) or begins wildly guessing (producing even worse output).
Frederick illustrates this with a Gantt chart in his article—strikingly intuitive: under human development, fuzzy requirements merely extend the development timeline; under AI-driven development, fuzzy requirements extend both the development timeline and the requirements documentation timeline—because you must invest significantly more time refining requirements until they’re precise enough for AI to understand.
Some object: “Then just have domain experts write more detailed requirements.” That sounds easy—until you consider what it really means. Asking a business expert to describe “a user recommendation feature” at the level of detail needed for AI to generate production-ready code may take more time and effort than simply having a human developer build it.
An Old Lesson from Bottleneck Theory—Newly Validated in the AI Era
The Goal contains a core principle: the bottleneck must receive predictable, high-quality input.
If your process stalls at legal review, the solution isn’t hiring more lawyers—it’s ensuring submissions to Legal are complete, correctly formatted, and require zero back-and-forth clarification.
AI sharpens this lesson dramatically. As an executor, AI demands higher input quality than humans do. Humans compensate for missing information with common sense; AI cannot.
So AI won’t speed up your process—but it will reveal exactly where—and why—it’s slow.
And that’s actually good news.
AI’s Real Value: Forcing Process Transparency
In the past, an inefficient process could hide behind the vague excuse: “Development is just slow.” Unclear requirements? No problem—the dev team figures it out. Slow legal sign-off? Fine—development takes forever anyway.
Now AI arrives, compressing development time—and suddenly, every upstream inefficiency is exposed. Like receding tide revealing who’s been swimming naked.
That may feel uncomfortable for managers—but for the organization, it’s an evolutionary opportunity. You finally have real motivation to fix long-ignored problems: the quality of requirements documentation, cross-departmental collaboration efficiency, and the length of decision-making chains.
AI isn’t a process accelerator. It’s a process X-ray machine.
A More Pragmatic Recommendation
So if you want to introduce AI into your organization and expect it to improve efficiency, don’t start with “Let AI write code.” Start with “Standardize inputs and outputs at every stage of the process.”
Ensure requirements documents follow a fixed template with mandatory fields. Ensure legal reviews come with explicit checklists. Ensure every deliverable has unambiguous acceptance criteria.
These tasks are tedious—but they’re the essential prerequisites for AI to function effectively.
Frederick closes his article quoting The Goal: “The bottleneck should receive predictable, high-quality input.” In the AI era of 2026, that statement is truer than it was in 1984—when the book first published.
AI won’t speed up your process. But it will force you to fix it—that’s its real value.