AI Made Coding 10x Faster — So Why Aren't Products Shipping Any Sooner? A 25-Year Dev Leader's Diagnosis
AI coding tools have sped up writing code, but coding is only 20% of the job. A development leader with 25 years of experience identified the 5 real bottlenecks and their solutions, earning 112 upvotes on Hacker News.
People say adopting AI coding tools makes development faster. And indeed, the speed of writing code has improved dramatically. But why does the time it takes for a product to reach users remain unchanged? Andrew Murphy — who has written code for 25 years, led teams for 15 years, and mentored over 1,000 developers — laid out the reasons in a post that received 112 upvotes on Hacker News.
Andrew Murphy — A 25-year development leader who has worked at Australia Post, Xero, Linktree, and more
Writing Code Faster Isn't the Real Problem
Murphy borrows a management concept called the Theory of Constraints (a framework that says every system has one weakest link limiting its overall performance). The idea is simple: "Every system has exactly one bottleneck. No matter how much you optimize anything that isn't the bottleneck, the overall system won't get faster." In other words, just like water gets stuck at the narrowest point of a pipe, work speed is determined by the slowest stage in the process.
In most organizations, writing code accounts for only about 20% of total work time. Murphy says, "I've seen features that were finished in half an afternoon take two months to actually reach users — multiple times." Even if AI makes that 20% lightning fast, it's pointless if the other 80% stays the same.
The 5 Real Bottlenecks — They're All Outside of Coding
1. You're Building the Wrong Thing
Here's a real case Murphy witnessed. A team spent 6 weeks building a feature based on requirements passed along by a sales rep. The result? The prospect never bought, and the feature was used by a total of 11 people — 9 of whom were the internal QA team. Even if AI had built it in 2 weeks, the outcome would have been exactly the same.
2. Wait Times After the Code Is Done
Even after the code is complete, there's still a long pipeline ahead: code review → CI pipeline (automated build and test systems) → staging (pre-production testing environment) → QA → security review → deployment approval. When AI churns out code faster, this queue actually gets longer. Reviewers suffer from context switching (the mental cost of jumping between different tasks), and end up hitting the approve button without thoroughly reviewing.
3. A Culture Where Deployment Is Scary
Teams that fear deployment (pushing new code to the live product) tend to batch up changes and ship them all at once. The bigger the batch, the higher the chance something breaks — and when something breaks, deployment becomes even scarier. It's a vicious cycle. AI producing code faster just makes the batch grow even quicker.
4. No Feedback After Launch
Many teams ship a feature and never check whether users actually use it. When you decide what to build next based on guesswork instead of data, you circle right back to problem #1 — building the wrong thing.
5. Organizational Structure Holds You Back
A single person who must approve everything, plans that only change quarterly, calendars packed wall-to-wall with meetings — these are problems no AI tool can solve.
55 Hacker News Developers Weighed In — Opinions Were Split
The post sparked a heated debate with 112 upvotes and 55 comments on Hacker News.
Those Who Agreed
"The real bottleneck is the speed of understanding" — Multiple commenters agreed that "writing code fast and understanding a problem fast are completely different things." One commenter put it this way: "Why can't you understand faster? That's like saying talking faster makes a conversation deeper."
Those Who Disagreed
"Failing fast is learning in itself" — The counterarguments were equally compelling. "If you're going to build the wrong thing anyway, it's better to build it fast and realize it fast." The idea is that rapid prototypes (quick, rough versions built to test an idea) let you validate hypotheses sooner.
Those Who Pointed to Real-World Constraints
"Customers have no patience for experiments" — In B2B (business-to-business) environments, customers won't tolerate repeated failures. In regulated industries, rapid experimentation is often impossible. While speed equals survival for startups, the situation is very different at large enterprises.
The Hidden Side Effects of AI Coding Tools
There's a core issue that Murphy flagged and Hacker News commenters echoed repeatedly.
The code review bottleneck. When AI generates code rapidly, the volume of code humans need to review explodes. One commenter proposed a rule: "For every PR (pull request) you submit, you review one of someone else's."
The understanding gap. If a developer doesn't fully understand the code AI generated, responding to a service outage at 3 AM becomes much harder. Dashboards may show a 40% productivity boost, but the number of features actually reaching users shrinks — a cruel irony.
So What Should You Actually Do?
Murphy isn't saying to ditch AI tools. He's saying to change the order of priorities.
① Map your entire workflow — From the moment an idea is born to the moment a user actually uses it, write down how many days each stage takes.
② Find the slowest point — It's not writing code. Check where work is actually stuck. Most of the time it's waiting for reviews, waiting for approvals, or waiting for deployment.
③ Limit work in progress — Finishing 2 things is better for overall speed than juggling 5 things at once.
④ Measure outcome time — Don't measure "how many lines of code were written." Measure "how many days from commit to reaching users."
⑤ Ask your developers — When you ask what frustrates them most, the answer is almost never "coding is too slow." It's usually "decisions aren't being made" or "reviews are backed up."
Murphy's conclusion boils down to one sentence: "Fix the bottleneck. It's not the keyboard."
If you're using AI coding tools, it's worth pausing to look at your entire workflow. Once you find where things are truly slow, you'll also see how to use AI far more effectively.
Related Content — Get Started with AI Using EasyClco | Free Learning Guide | More AI News
Stay updated on AI news
Simple explanations of the latest AI developments