Tired of Talking to AI, PMF Found, and DuckDuckGo's 28% Surge

Good morning. It's Wednesday, May 27, and we're covering the backlash to AI-mediated everything, the evidence that AI tools have crossed the PMF threshold, and DuckDuckGo's unexpected traffic spike after Google doubled down on AI Mode.

A rare Wednesday where the top three HN stories were all about AI - but approaching it from almost opposite angles.

THE BIG ONE

"I'm Tired of Talking to AI"

The day's highest-scoring post - 2,013 points, 951 comments - was not a launch or a benchmark. I'm Tired of Talking to AI is a personal essay from a writer who describes the slow erosion of genuine discovery when every search, every question, and every creative block gets routed through a chatbot. The piece is careful: it is not anti-AI, it is specifically about the friction of conversational interfaces for non-conversational tasks, and the exhaustion of managing a model's caveats, hedges, and hallucinations when you just want a straight answer or a human voice.

The HN thread is one of the better discussions of the week - a mix of people who recognized the feeling immediately and engineers who pushed back with "then use it differently." The post connects directly to Thursday's Claude Opus 4.8 framing around reducing sycophancy: Anthropic is clearly hearing this feedback at scale.

MODELS

Simon Willison: Anthropic and OpenAI Have Found Product-Market Fit

Simon Willison published a 1,900-word analysis (1,094 points, 1,245 comments) that did the rare thing of putting real numbers to an intuition everyone had. His argument: Anthropic is strongly rumored to be approaching its first profitable quarter, stories are circulating of companies shocked by their LLM bills from employee usage, and the Uber CTO shared that Claude Code can blow AI budgets without guardrails. This is not hype-driven spending - it is utility-driven spending that is hard to cut because the productivity gains are real.

Willison also flagged that SQLite gained an AGENTS.md file five days prior - not for their own development, but to tell agents pointing at the SQLite codebase what the project will and will not accept. The most notable line: SQLite strengthened its statement to "does not accept agentic code" (removing the "(currently)" qualifier), while simultaneously welcoming agentic bug reports with reproducible test cases.

INFRA

DuckDuckGo Saw 28% More Visits After Google Said People Love AI Mode

DuckDuckGo's AI-free search saw nearly 28% more visits in the week following Google's public insistence that users embrace AI Mode (1,075 points). The timing is notable: Google's blog post leaned into AI search as the future of the product, and a measurable slice of users responded by switching to the one major engine that explicitly markets itself as not doing that.

WHAT ELSE IS HAPPENING

• YouTube to automatically label AI-generated videos (1,322 pts): YouTube's expanded AI labeling policy will auto-detect and label synthetic content even when creators do not disclose it - the system uses both audio and video signals.
• Tech CEOs are apparently suffering from AI psychosis (722 pts): TechCrunch on the growing body of evidence that AI-related grandiosity and disconnection from reality has become a C-suite pattern, not an outlier.
• Claude Code as a Daily Driver (451 pts): A detailed practitioner guide covering CLAUDE.md structure, skills, subagents, plugins, and MCPs - one of the more thorough "how I actually use it" posts to date.
• GitHub major outage (335 pts): An incident affecting pull requests, issues, git operations, and API requests - the status page thread became a real-time tracker for teams mid-sprint.
• Go: Support for Generic Methods (311 pts): A new GitHub issue on the Go repo proposes adding generic methods - a long-requested feature that the language team has historically resisted on implementation complexity grounds.

FROM THE SITE

AI Chat Fatigue Is a Workflow Design Bug argues that the exhaustion described in the day's top post is not a sentiment problem but an architecture problem - chat is the wrong interface for most of what people are trying to do with AI, and verifiable workflow outputs are how you get out of the loop.

---

Every link above goes to a primary source. This brief is part of the Daily Brief archive.