Do AI coding tools actually make experienced engineers faster, or only increase code volume?

Amjad Masad · Mar 10, 2023

We don't need to speculate about whether the AI coding jump is real — measured results already show it is the start of something, even if not yet a 10x

Replit sees 30–50% of Ghostwriter users' code written by the AI with tasks sometimes cut in half, and a study of GitHub Copilot users showed programmers were 55% more productive

Scope: explicitly not a 10x improvement yet

24:09 20VC: Why AI Will Lead to Thousands of Billionaires and Elon Musk's, Will TikTok Be Banned and How Facebook Should Be Investing in AI & Why Startups Have Become Too Soft; We Need a Spiritual Reform with Amjad Masad, Founder & CEO @ Replit

Cem Kansu · Jun 20, 2025 · hedged

Today's AI coding tools mostly lift average engineers rather than turning 10x engineers into 100x, and the gain is only about 10-20% more productivity — far less than the hype suggests

That is his read of actual productivity today; some companies are making overblown statements like no longer hiring engineers

Scope: 'my read right now'

22:23 20Product: How Duolingo Build Product 10x Faster with AI | Duolingo's Biggest Lessons on Paywalls, Push Notifications and In-App Purchases | Why Small Teams are the Future of Product | Why PMs Will Become Extinct with Cem Kansu, CPO @ Duolingo

Martin Casado · Jul 28, 2025

AI coding tools make 10x engineers roughly 2x, not 100x

Every company he works with uses Cursor, yet he doesn't see product velocity increase much

38:15 20VC: a16z's Martin Casado on Anthropic vs OpenAI: Where Value Accrues | Cursor vs Replit vs Lovable: Who Wins and Who Loses | The One Sin in AI Investing | Why Open Source is a National Security Risk with China

Paul Erlanger · Jun 27, 2026

AI coding tools make engineering genuinely faster, not just higher-volume — including for the best engineers

Reviewing AI-written code is far faster than writing it, and having a generated framework accelerates the learning curve for unfamiliar components; the team shipped a full product in three weeks and a web app in one month

Scope: engineers still restructure and rewrite most of the generated code

25:42 20VC: How We Got Fred Wilson, Benchmark and Index to Invest $94M | Why Robinhood's Strategy is Wrong | Why 1-1s are BS and What Every Founder Gets Wrong About Equity | Why Taste Beats AI But How AI Kills Org Charts with Paul Erlanger, CEO @ fomo

Arvind Jain · Jul 11, 2026

Paying the cost of human code review is still net faster than the pre-AI workflow

The writing part is dramatically faster now, and the person who writes the code performs the first review

Scope: AI-generated code is not perfect right now

24:31 20VC: Why OpenAI and Anthropic Won't Win the App Layer | Why Teams Will Get Bigger Not Smaller in a World of AI | Why AI Removes Incumbents Advantage of Bundling | China vs America: Who Wins the AI War with Arvind Jain, Co-Founder @ Glean

Andrew Ng · Nov 17, 2025

AI-assisted coding has delivered an enormous productivity step-change: projects that once took six engineers half a year can now be built by one person in a weekend

First-hand experience at AI Fund plus personal examples like generating printable flashcards for his daughter instead of buying them

Scope: software engineering specifically

12:38 20VC: Andrew NG on The Biggest Bottlenecks in AI | How LLMs Can Be Used as a Geopolitical Weapon | Do Margins Matter in a World of AI? | Is Defensibility Dead in a World of AI? | Will AI Deliver Masa Son's Predictions of 5% GDP Growth?

Andrew Ng · Nov 17, 2025

AI-assisted coding is already delivering real productivity gains and real returns, and is changing how software is written.

Observed productivity returns and his own and friends' experience of coding being far more effective and enjoyable with AI help.

44:43 20VC: Andrew NG on The Biggest Bottlenecks in AI | How LLMs Can Be Used as a Geopolitical Weapon | Do Margins Matter in a World of AI? | Is Defensibility Dead in a World of AI? | Will AI Deliver Masa Son's Predictions of 5% GDP Growth?

Zach Lloyd · Oct 17, 2025

In professional development environments, nobody actually knows whether AI coding tools produce productivity gains, and several studies suggest they don't

The noise created by people applying vibe-coding techniques to production code bases can outweigh the gains and slow you down; companies rush to deploy agentic coding without knowing if it works

15:56 Unclear productivity gains studies suggest no real benefit

Zach Lloyd · Oct 17, 2025

AI coding tools do work well for professional developers, but only if you tell the agent how to build the thing, which requires understanding the code

Naive 'build me this thing that looks like X' prompting doesn't tell the agent how it should work engineering-wise, so on a large complex code base it fails, loops for hours, or produces something you'd never release

20:53 Gains require explicit engineering guidance not naive prompts

Martin Casado · Jul 28, 2025

AI coding tools have made programming pleasant again by offloading the non-foundational framework, packaging and hosting overhead that had come to consume roughly 90% of a developer's time

By the mid-2010s most of the work was downloading packages, running dev servers, hosting and reconciling library incompatibilities rather than writing code; now the AI handles hosting and package choices and the developer can focus strictly on logic

35:18 Ai removes tedious overhead making coding enjoyable again

Martin Casado · Jul 28, 2025 · hedged

The main impact of AI coding tools will be more robust, maintainable code bases with fewer bugs rather than feature velocity

Genuinely hard work like experimenting on frontier models can't be offloaded, but tests, visualization and documentation can

38:41 Gains materialize as code quality not feature velocity

Severin Hacker · May 19, 2025

AI coding tools excel at zero-to-80% on simple apps and at isolated single-file transformations, but degrade sharply on large codebases and create tech debt they cannot resolve

The last 10% of a feature takes as long as the first 80%, and the tools generate their own tech debt they can't then fix

12:22 Strong on simple greenfield tasks degrades on large codebases creating tech debt

Daniel Dines · Dec 18, 2024

AI coding tools will deliver only modest, not gigantic, productivity improvements for companies building genuinely hard software spanning many technologies

UiPath's technology is much harder to build than Salesforce's and spans many different technologies, which limits how much AI tooling can accelerate it

43:35 Hard multi technology software limits ai coding tool productivity gains to modest levels

Eran Zinman · Mar 2, 2026

AI coding tools raise individual engineer productivity, but each gain surfaces new bottlenecks unrelated to writing code, so team-level output gains lag

Every time you increase one productivity dimension you find a new constraint

24:34 Individual gains are real but new bottlenecks cap team output