Development statistics for the 7.2 kernel

📝 Discussion Summary (Click to expand)

Prevalent themes in the discussion

Concern about LLM‑generated “slop” degrading code quality
“While great in theory, the flip side is all the slop patches and PRs that both the kernel and open source in general now has to deal with, often with only agents as responders.” – Matl
“LLMs still suck at programming and output slop.” – bitwize
Debate over whether understanding/competence matters when using AI tools
“Linus's claim that use of LLM tools is uncorrelated with understanding or competence is ludicrous.” – senderista
“He’s saying that understanding or competence are more about the person than their AI use.” – munchbunny
Teiferer frames the split: if we don’t require human understanding, AI can be embraced fully; if we do, humans must stay in the loop.
Pragmatic optimism that AI is just another tool to be adopted, like past innovations
“The point is, new technology arrived that allowed developers to do much more… But the sensible engineers adopted the technology.” – bitwize
“Linus has always been pragmatic as opposed to dogmatic.” – dist‑epoch
“Yes, Linus is quite forward thinking, and we can credit that with why Linux has stayed relevant for so many decades.” – nvme0n1p1

🚀 Project Ideas

Automatically builds, runs static analysis, and executes property‑based tests on LLM‑submitted kernel patches before they reach maintainers.
Returns a confidence score and highlights potential regressions, reducing “slop” and increasing trust in AI‑assisted contributions.

Key	Value
Target Audience	Kernel maintainers, subsystem maintainers, CI engineers
Core Feature	Pre‑merge validation pipeline (compile, Coccinelle/Sparse, KUnit/fuzz, diff‑based regression detection)
Tech Stack	Rust (for fast binary analysis), Python (orchestration), GitHub Actions / Buildkite, LLVM‑based sanitizers
Difficulty	Medium
Monetization	Revenue-ready: SaaS subscription ($15/mo per maintainer)

HN commenters worry about “slop patches” and lack of correctness: “LLM usage is anti‑correlated with understanding or competence.” – senderista
Provides the safety net maintainers need to accept AI help without fear of regressions, directly addressing the desire to keep humans in the loop while still benefiting from AI speed.
Could spark discussion on what level of automated testing is sufficient for kernel patches and encourage better test coverage from contributors.

Interactive web tool that guides newcomers through kernel contribution, requiring them to explain AI‑generated code, answer probing questions, and produce minimal test cases before a patch can be submitted.
Ensures that contributors truly understand what they are submitting, turning AI into a learning aid rather than a crutch.

Key	Value
Target Audience	New kernel contributors, students, hobbyist developers
Core Feature	Guided checklist: code explanation, Q&A (generated from LLM), test‑case generation, and reviewer‑ready summary
Tech Stack	TypeScript/React frontend, Node.js backend, PostgreSQL, LLM API (open‑source or hosted)
Difficulty	Medium
Monetization	Revenue-ready: Tiered pricing (free for individuals, $9/contributor/mo for orgs)

Reflects the sentiment that “the kind of expectations I have is that people do their best and I expect people to understand what they are doing.” – mchbunny
Forces the human to demonstrate understanding, addressing the concern that AI use correlates with superficial competence.
Would likely generate lively HN debate about the balance between accessibility and maintaining high standards in kernel development.

Scans incoming patch emails or PRs for linguistic and statistical markers typical of LLM output (e.g., repetitive phrasing, lack of “Signed‑off‑by”, low comment density) and flags them for extra reviewer scrutiny.
Suggests concrete improvements (add tests, improve commit message, reference relevant documentation) to raise patch quality.

Key	Value
Target Audience	Kernel subsystem maintainers, mailing‑list moderators, release engineers
Core Feature	Real‑time LLM‑likelihood scorer with actionable feedback displayed in patch review UI
Tech Stack	Python (scikit‑learn, spaCy), Elasticsearch for patch corpus, Redis queue, webhook integration with Patchwork/lore.kernel.org
Difficulty	Low
Monetization	Hobby (open‑source, optional donations via Open Collective)

Directly tackles the frustration expressed by oggofart: “We get a lot of PR made by other's people agents, which can be a bit overwhelming.”
Quotes like “The agent helps a lot to fix bugs… but we get less bug reports” show a need to differentiate genuine contributions from AI‑generated noise.
Provides a lightweight, deployable tool that maintainers can adopt today, fostering discussion on what constitutes “slop” and how to detect it automatically.