Key Takeaways

  • While AI shows clear productivity gains in areas like customer support, developer coding speed boosted by AI often fails to accelerate product shipping.
  • The reason for this bottleneck isn't the AI itself, but downstream processes like stringent code reviews and the complexity of maintaining AI-generated code.
  • Arind, founder of Glean, argues that “AI ROI is really a throughput problem,” meaning the real value comes from optimizing the entire workflow, not just isolated tasks.
  • Instead of solely focusing on job replacement, founders should "invest around" AI by providing it the right context to enhance overall team performance and reduce costs.

The Productivity Mirage: Faster Coding, Same Ship Date

It's a common pitch: AI will make your developers write code at lightning speed. And in some metrics, it does. But according to Arind, founder of Glean, this doesn't automatically translate to faster product delivery. “But are they shipping the products faster or not and that's where we hear most of the companies saying that no that the actual shipping speed of products has not increased,” Arind told Harry Stebbings on 20VC.

The problem isn't the AI's ability to generate code. It's what happens after the code is written. Harry pinpointed one of the biggest blockers: “Well if you have to have a stringent code review process it almost removes the point of having a fastened code development process.” Teams often grapple with the quality and maintainability of AI-generated code, necessitating rigorous human oversight. This means the time saved in writing the initial lines gets eaten up in reviews, debugging, and integration headaches. What looks like a win for individual developer productivity becomes a wash for the product team's shipping cadence. This stark reality challenges the prevalent narrative that simply giving developers AI tools will magically accelerate your entire release cycle.

From Output to Throughput: The Real AI ROI

So, if raw coding speed isn't the answer, where does the real value of AI lie for founders? Arind reframes the entire discussion, declaring, "AI ROI is really a throughput problem." He highlights that while “pockets of value realization” exist today—like in customer support, where measuring productivity is straightforward—the larger picture requires a different lens.

The core insight is that AI's impact isn't about replacing individual tasks or roles, but about enhancing the overall flow of work through your system. Arind suggests that instead of focusing on headcount reduction, companies need to “invest around it. You have to make sure that you provide it the right context uh so that it can actually work faster at at a lower cost.” This means setting up AI agents with the right data, clear objectives, and seamless integration into existing workflows to remove bottlenecks and accelerate the entire pipeline. It’s about optimizing the system, not just one part of it. Think less about how many lines of code AI can write, and more about how much faster a fully tested, deployed feature can go from idea to customer.

What to Do With This

This week, audit your last two completed product features. Map out every step from concept to customer delivery, noting the actual time spent on each. Specifically, identify any steps where AI might have sped up an individual task (like coding or design iterations) but the overall project timeline remained unchanged. Pinpoint the exact bottlenecks that absorbed those gains (e.g., code review cycles, QA, cross-team communication, internal approvals). Re-evaluate your current or planned AI investments based on how they would directly impact these identified bottlenecks, not just isolated task speeds.