Van Brabant: AI speeds execution, not broken inputs
Key insights
- AI accelerates execution but software bottlenecks typically sit upstream in vague requirements and missing domain expertise, not in code output speed.
- The essay draws on Goldratt's The Goal and The Toyota Way to frame AI as local optimization applied to a globally broken input process.
- 474 Hacker News points and 340+ comments suggest working engineers widely question whether the AI-as-speed-multiplier framing correctly diagnoses delivery constraints.
Why this matters
Practitioners building AI-assisted workflows are currently measuring ROI in lines of code written or tasks completed per hour, a metric blind to the upstream requirements failures Van Brabant identifies. If the bottleneck in enterprise software delivery is genuinely in specifications and domain translation rather than execution, the entire market for AI coding assistants is solving the second-order problem while the first-order problem remains unfunded. The Hacker News response of 474 points and 340+ comments signals this critique is already shaping how technical leaders frame budget conversations around AI tooling investments.
Summary
Frederick Van Brabant's enterprise architecture essay hit 474 points on Hacker News with a blunt claim: AI cannot speed up a slow process because the bottleneck is rarely in execution.
The real drag is upstream, in vague requirements, unclear specs, and missing domain expertise. Drawing on Goldratt's The Goal and The Toyota Way, Van Brabant argues that layering AI onto a broken input process just produces wrong output faster, not better delivery.
Essentially: (engineering teams, enterprise architects) are optimizing the wrong constraint.
- 340+ HN comments debated whether the AI-as-multiplier framing misdiagnoses where delivery time actually goes
- Van Brabant's core claim: code-generation speed is irrelevant when requirements are ambiguous
- The Toyota Way lens frames AI adoption as local optimization applied to a globally broken system
Orgs measuring AI ROI in execution speed are tracking a metric that doesn't touch the real bottleneck.
Potential risks and opportunities
Risks
- Enterprise AI coding tool vendors (GitHub Copilot, Cursor, Tabnine) face a credibility challenge if engineering leaders adopt Van Brabant's framing and reattribute slow delivery to requirements quality rather than tooling gaps.
- Organizations that have already purchased AI coding tools at scale may face internal pressure to justify spend in the next 12 to 18 months if delivery velocity does not improve alongside adoption rates.
- The framing could accelerate budget shifts away from AI execution tools toward requirements engineering and domain expertise, staffing categories that are harder to automate and more expensive to scale quickly.
Opportunities
- Requirements management and specification tooling vendors (Linear, Notion, emerging Jira alternatives) could capture redirected AI budgets if engineering leaders move investment upstream toward better spec quality.
- Consultants and firms specializing in enterprise architecture and process improvement (ThoughtWorks, McKinsey's engineering practice) gain sharper sales leverage for engagements framed around fixing upstream bottlenecks before AI deployment.
- AI tools specifically targeting requirements clarification and domain modeling rather than code generation have a stronger differentiated pitch if Van Brabant's framing gains traction among CTOs and engineering VPs through 2026.
What we don't know yet
- No data is cited on whether organizations that have improved upstream spec quality see AI tools deliver proportionally larger speed gains afterward.
- The essay uses software development as its sole case study; whether the argument holds in non-software enterprise processes such as procurement or compliance is unaddressed.
- Van Brabant does not quantify what share of delivery time goes to upstream requirements versus execution across different industry verticals or team sizes.
Originally reported by frederickvanbrabant.com
Read the original article →Original headline: HN: 'I Don't Think AI Will Make Your Processes Go Faster' — 474 Points as Essay Argues Bottlenecks Are Upstream, Not in Execution