← → SPACE to navigate

Data, Myths, and What Leaders Must Do Now

Is AI Killing
Open Source
Software?

Stormy Peters

01 / 31

A note before we begin

Confession


02 / 31

Data, Myths, and What Leaders Must Do Now

Is AI Killing
Open Source
Software?

Stormy Peters · LF Member Summit · February 2026

03 / 31

I Was Right ... And Wrong


Anyone
Can Do It
Lots of
Slop
It's Not
Mine
I Can't
Maintain It

My initial theory was partially right — but the reality is more nuanced.

04 / 31

The Canary in the Coal Mine

cURL

15B

Devices running cURL globally

Global internet infrastructure estimate · Source: UN verified

7

Maintainers

The core team behind cURL, securing critical infrastructure.

05 / 31

January 2026:
The Breaking Point

7 REPORTS
08:00 AMBug Report #4182 INVALID
09:15 AMBug Report #4183 AI SLOP
11:00 AMBug Report #4184 AI SLOP
12:45 PMBug Report #4185 INVALID
04:45 PMBug Report #4186 NO VULNERABILITY
08:00 PMBug Report #4187 AI SLOP
11:50 PMBug Report #4188 AI SLOP

"Taking care of this lot took a good while… none of them identified a vulnerability."

— Daniel Stenberg
7

Bug Submissions
In 16 Hours

06 / 31

We Need to Ensure
Our Survival

Bug
Bounty

cURL Bug Bounty Program:

Terminated.

6 years · $86,000 in payouts · February 2026

"The main goal is to remove the incentive for people to submit crap… we need to do something to prevent us from drowning."

— Daniel Stenberg
07 / 31

The Paradox: Adoption is Explosive

85% 75% 2024 2026 76% 84%

84% of Developers Now Use AI Tools.

GPT, Claude, and Gemini are now standard fixtures. Source: Stack Overflow Developer Survey 2025

08 / 31

…But Trust is Freefalling

84% 42% 2023 2025/26 76% 84% ADOPTION 42% 33% TRUST IN ACCURACY

Usage is ubiquitous. Trust is scarce.

09 / 31

The 'Almost Right' Trap

function processUserData(data) {
  const userId = data.user_id;
  const accountStatus = data.status;

  // ... complex logic ...

  return {
    id: userId,
    status: accountStatus
  };
}

Subtle, critical logic error: userId is exposed in the response

66%

Developers frustrated by solutions that are "almost right"


45%

Say debugging AI code takes MORE time than writing it

10 / 31

The Productivity Mirage

Expectation vs Reality · METR Study · Experienced OSS Maintainers

Predicted Speed
(Experts)

+39%
VS

Actual Speed
(Measured)

−19%

"If I am the dedicated maintainer… there is no way agent mode can do better than me."

— Dedicated OSS Maintainer, METR Study participant

Randomized Controlled Trial with 16 experienced OSS maintainers.

11 / 31

The High Interest Rate of Technical Debt

Human-only PRs vs. AI-co-authored PRs · CodeRabbit Analysis · 470 Open Source PRs

Overall Issues
BASELINE
1.7× More
Readability
BASELINE
3× More
Logic Flaws
BASELINE
75% More
12 / 31

The Pre-Existing Condition

60% Are Unpaid.
60% Consider Quitting.
44% Cite Burnout.

"I have debts, a full-time job, a young family… I have already sunk thousands of hours into this project."

— Anonymous OSS Maintainer Survey Respondent
13 / 31
GOVN

Section

The Governance
Response

How the open source community is fighting back against AI-generated noise.

14 / 31

The Response

The Ban
(Gentoo)

Banned

Expressly forbidden to contribute content created with Natural Language Processing AI tools.

  1. 1. Copyright — Unclear provenance
  2. 2. Quality — Plausible but meaningless
  3. 3. Ethics — Energy and labor
15 / 31

Did You Use AI?


95.2%

of all commits use AI

80.5%

Claude users disclose their AI use

29.5%

disclose AI use

9.0%

GitHub Copilot users disclose

Source: The AI Attribution Paradox: Transparency as Social Strategy in Open-Source Software Development

16 / 31

The Response

The Quarantine
(NetBSD)

AI-Generated
Code

17 / 31

Other Responses


18 / 31

Needs to Be
More Than
Governance!

Policy alone can't solve this. We need to change how we use, support, and value open source.

19 / 31

The Difference Between "Slop" and "Science"

Slop

Asking a bot and pasting the result. No understanding. No verification. No ownership.

Science

Stenberg praised Joshua Rogers for using AI-assisted tools to find a massive list of real bugs.

The difference? A clever human driving the tool.

20 / 31

Where Should We Be Using AI?

Finding Bugs
  • AI-assisted PRs have higher merge rates
  • Duplicate detection
  • Automated labels
  • Enforcing rules
  • Compliance monitoring
Triaging
  • Duplicate detection
  • Automated labels
  • Enforcing rules
  • Compliance monitoring
Learning
  • Accelerated onboarding
  • Knowledge base access
  • Skill development
  • 36% of devs use AI to learn
21 / 31
"
It used to be that we could mark GitHub issues as 'good first issue' and ambitious young engineers would show up… Now we file something as 'good first issue' and in less than 24 hours get absolutely inundated with low quality vibe coded slop that takes time away from doing real work. This pattern of 'turning slop into quality code' through the review process hurts productivity and hurts morale.
— Craig McLuckie, CEO of Stacklok
22 / 31

Editorial Crisis

It's Not All Doom:
The Learning Curve

36%

of developers use AI to learn new skills

(Up from 37% in 2024 · Stack Overflow 2025)

AI is a powerful tutor. It helps junior developers onboard and understand complex systems faster.

A tool for learning, not a replacement for knowing.

23 / 31
ACT

Section

What Can
We Do?

Practical actions for leaders, maintainers, and contributors.

24 / 31

Action

Have Excellent
Documentation for…

🧑‍💻
Experienced Developers
🌱
New Developers
🤖
AI Agents

Good documentation helps everyone navigate your project — and gives AI tools the context they need to be useful rather than harmful.

25 / 31

Action

Dependencies

01
Understand
Know what open source you actually depend on — including transitive dependencies.
02
Reduce
Fewer dependencies = smaller attack surface and lower maintenance burden.
03
Support
Fund and contribute to the projects your stack depends on.
26 / 31

Action

Is It Needed
Anymore?

Nolan Lawson maintains blob-util — a small JavaScript utility. With 80%+ of developers using AI, they'd just ask an LLM to generate blob utilities instead of finding a library.

blob-util
5M+

weekly npm downloads

AI may silently eliminate the small projects that grew the next generation of maintainers.

27 / 31

Support Open Source


01
Fund open source Donate to foundations and maintainers. Make it a budget line item.
02
Contribute time Let your developers work upstream on company time. Money doesn't review PRs.
03
Submit quality bug reports and PRs Whether you use AI or not — reproducible, well-researched reports have never been more valuable.
04
Attribute AI-assisted code Transparency builds trust. Disclose when and how AI helped.
28 / 31

The question

Is AI Killing
Open Source?

29 / 31
No,
but …
30 / 31

Discussion

Your
Thoughts?

31 / 31