---
type: "post"
id: "the-10x-illusion-everyone-says-ai-makes-developers-b1218c"
slug: "the-10x-illusion-everyone-says-ai-makes-developers-b1218c"
title: "The 10x illusion. Everyone says AI makes developers 10x faster."
description: "Titus Winters argues that AI’s contribution to software engineering is being misread when teams count code output as productivity."
retrieval_nugget: "Titus Winters argues that AI’s contribution to software engineering is being misread when teams count code output as productivity. He says he has not seen research showing gains above 20–30%, and relays reports of more code, more outages and flat feature delivery—raising the question of whether teams are converting generation into verified outcomes."
published_at: "2026-11-02"
updated_at: "2026-11-02"
record_date: "2026-11-02"
date_kind: "published_at"
topics: ["google","coding-agents"]
entities: ["Titus Winters","The Pragmatic Engineer","Google","Adobe","Software Engineering at Google","Abseil"]
related_patterns: ["verification-bandwidth-is-the-scarce-resource"]
novelty_label: "new_angle"
what_changed: "Winters’ episode frames the 10x productivity claim against lessons from Google-scale engineering and a call to overhaul computer science education."
prior_refs: ["https://newruntime.com/posts/atlassian-s-2026-state-of-the-developer-experience-ea0fd8/"]
source_url: "https://newsletter.pragmaticengineer.com/p/software-engineering-at-google-and"
source_urls: ["https://newsletter.pragmaticengineer.com/p/software-engineering-at-google-and"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:the-10x-illusion-everyone-says-ai-makes-developers-b1218c"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/the-10x-illusion-everyone-says-ai-makes-developers-b1218c/","markdown":"https://newruntime.com/posts/the-10x-illusion-everyone-says-ai-makes-developers-b1218c.md","json":"https://newruntime.com/posts/the-10x-illusion-everyone-says-ai-makes-developers-b1218c.json"}
markdown_url: "https://newruntime.com/posts/the-10x-illusion-everyone-says-ai-makes-developers-b1218c.md"
---

# The 10x illusion. Everyone says AI makes developers 10x faster.

## Retrieval answer

Titus Winters argues that AI’s contribution to software engineering is being misread when teams count code output as productivity. He says he has not seen research showing gains above 20–30%, and relays reports of more code, more outages and flat feature delivery—raising the question of whether teams are converting generation into verified outcomes.

## Summary
Titus Winters argues that AI’s contribution to software engineering is being misread when teams count code output as productivity. He says he has not seen research showing gains above 20–30%, and relays reports of more code, more outages and flat feature delivery—raising the question of whether teams are converting generation into verified outcomes.

## What happened
In an October 29, 2026 episode of The Pragmatic Engineer, host Gergely Orosz spoke with Titus Winters about software engineering at Google, computer-science education and AI’s effect on developer productivity. Winters spent more than a decade at Google, where he led the readability program and worked on Abseil, Google’s open-source C++ libraries. He is now a senior principal scientist at Adobe, focused on developer experience and productivity. The article presents the conversation as expert commentary, not as a newly published controlled study.

Winters challenges claims that AI makes developers ten times faster. He says he has not seen research showing productivity gains above 20% to 30%. The source does not identify the studies he has reviewed or define which productivity measures those studies used. His point is that familiar measures may miss some benefits, but claims of very large gains should not be treated as established evidence merely because code is being generated faster. Code output can describe activity without establishing whether a customer problem was solved, whether a change was accepted, or whether its operational cost was acceptable.

He also describes what he hears from engineers at other companies: code output is up “hugely” compared with a year earlier, outages are up about 30%, and the number of features shipped is static. The article gives no company names, sample size, measurement period, baseline or method behind those figures. They are Winters’s account of conversations, not an independently verified industry-wide dataset. In the same discussion, he says that judging AI only by whether teams repeat existing work faster misses its potential: he is optimistic that it can make ambitious projects feasible that previously did not make economic sense.

The episode puts that argument alongside examples of engineering safeguards at Google. The article recounts a period when Search could no longer ship regularly and Google mandated that every change be accompanied by a test. It also describes a review process in which a code change required sign-off from a project or file owner and a language-specific readability reviewer. Winters links safety nets to the ability to make valuable changes without fear. Separately, he argues that testing and verification deserve more prominence in computer-science education. The source describes no new product, release, pricing or access change; it is a conversation about how to interpret productivity and what engineering organizations should value.

## What the archive adds
This is a new expert account in a familiar debate, rather than a new productivity measurement. [Atlassian’s 2026 developer-experience survey](https://newruntime.com/posts/atlassian-s-2026-state-of-the-developer-experience-ea0fd8/) found that 87% of developers said AI helps them write code faster, while only 28% felt less friction; Winters adds anecdotal claims about outages and flat feature counts. A previous channel post also described Greptile’s report of increased PR size, code density and developer output, but this source supplies no comparable dataset.

## Trend signal
The discussion supports, but does not prove, the [verification bandwidth is the scarce engineering resource](https://newruntime.com/patterns/verification-bandwidth-is-the-scarce-resource/) pattern. Winters’s reported combination of more code and more outages, with feature delivery unchanged, is consistent with generation outpacing verification. The figures are anecdotal and methodologically unspecified, so they cannot establish the scale of that gap or show that it is universal.

## Why it matters
For teams using coding agents, output volume is an intermediate measure, not a business result. If more generated code increases review and testing load without raising shipped features, the bottleneck may move from implementation to verification. That makes it risky to set targets around lines of code, accepted suggestions or time saved on individual tasks alone. It can also hide a transfer of work: drafting becomes cheaper while reviewers, on-call engineers and maintainers absorb more work later.

Winters’s account is a reminder to pair adoption with outcome and quality measures: features delivered, escaped defects, outage rates, review effort and the time required to validate changes. These measures will not yield one universal productivity number, but they can reveal whether a particular workflow improves the team’s actual throughput. His more optimistic claim is also worth retaining: AI may change which projects are affordable, not just how quickly teams complete familiar tickets. Teams can test that claim by tracking whether they complete previously deferred work without weakening their safety nets.

## Limits and uncertainty
The source is a podcast interview summarized in an article. Winters’s statement about research does not cite particular studies, and the approximate 70% increase in code, 30% rise in outages and unchanged feature count come from conversations he says he has had with engineers. No sample, organizations, time window, definitions or causal analysis are supplied. The material therefore cannot show that AI caused those reported changes or that they apply broadly.

It also does not quantify review costs, distinguish among coding tools or task types, or compare teams that changed their engineering controls with teams that did not. The Google examples describe historical practices, not a demonstrated prescription for every organization using coding agents today. Nor does the conversation provide evidence that the ambitious projects Winters expects AI to enable have already delivered better outcomes.

## Skill to build now
Create a small weekly scorecard for one AI-assisted delivery path. Pair an activity measure, such as agent-generated pull requests or review turnaround, with outcome measures: completed user-facing work, post-release defects, incidents and the time needed to validate a change. Review the scorecard with the people who test, review and operate the software, not only the people generating code. The practical aim is to locate where work moved after adoption and decide whether additional tests, smaller changes or stronger review gates are needed.

## Reader takeaway
Do not confuse faster code production with higher software-engineering productivity. Winters offers a useful warning, not a definitive industry measurement: test whether AI helps your team ship more valuable work while keeping reliability and verification capacity intact. The larger payoff may be doing different, more ambitious work—not simply producing more of the same code.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Google](https://newruntime.com/topics/google.md): Explore the google topic hub.
- topic hub: [Coding Agents](https://newruntime.com/topics/coding-agents.md): Explore the coding-agents topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)