On Monday, the June 2026 edition of “Ask HN: What Are You Working On?” landed on Hacker News and racked up 642 comments by midweek — a number that would have read as a vibrant maker culture flex five years ago. Today, it reads like something else. Scroll past the first hundred entries and a pattern solidifies: solo founders on their third pivot, engineers quietly noting which projects are “[NO-AI]” as though that’s now a content warning, and a striking number of people building tools for other builders — CI/CD dashboards, PR review aids, yet another lightweight Kagi alternative. A lot of scaffolding. Not many houses.

This is not the thread as it existed in 2019 or even 2022. The monthly ritual used to be equal parts portfolio and party invitation. Now it scans like a support group for people whose primary output is the tooling required to keep producing output. The optimism is still there — it always is on HN — but it’s increasingly directed at the infrastructure of making rather than the things being made.

What changed is not mysterious. The window between “I shipped an AI wrapper” and “OpenAI shipped the same feature as a platform update” has narrowed to roughly a quarter, and everyone in that thread knows it. The rational response, for a certain kind of engineer, is to move one layer down the stack — stop building the thing users touch, start building the thing builders touch. It’s defensible. It’s also a trap.

The Platform Risk Has Moved Indoors

The conventional anxiety about building on someone else’s platform used to mean Twitter’s API or Apple’s App Store. In 2026, the platform that can kill you is the foundation model provider whose next release makes your differentiation evaporate. The HN thread is littered with projects that are one GPT-5.2 or Gemini Ultra away from being subsumed: RAG pipelines, agent orchestration layers, “multi-agent chat” for verticals like wellness — all of which are sensible ideas that happen to sit precisely where the big labs are aiming their commoditization cannons.

So builders retreat. They build for other builders. The logic is that developers are the one customer segment that never gets abstracted away by a dropdown menu. And there’s truth to that. Stripe and Twilio proved the model works. But the 2026 version is different: it’s not just API-first companies, it’s entire ecosystems of micro-tools — docker images for isolated AI coding environments, testing-session recorders for pull requests, custom Go stacks to replace Siemens industrial controllers — that are deeply useful to maybe 4,000 people and economically viable for exactly none of them.

One developer in the thread described his project as “continuously improving my docker image for running OpenCode in an isolated and security-focused environment.” That’s a real comment, from a real person, and it is the platonic ideal of where this energy is going: inward, recursive, admirable in craft, and unlikely to ever meet a customer who isn’t also a contributor.

The [NO-AI] Tag Is the New Organic Label

A small but telling detail: several projects in the thread are now labeled “[NO-AI],” including one search-engine alternative whose creators explicitly flag it. This is not a Luddite gesture. It’s a market signal. The same way “organic” emerged as a differentiator once industrial agriculture saturated the shelf, “[NO-AI]” is becoming a differentiator for a segment of users who have decided that AI-generated content is the high-fructose corn syrup of the information diet.

Whether that segment is large enough to sustain a business is an open question. Kagi, the paid search engine that inspired at least one of these projects, claims roughly 25,000 paying subscribers as of late 2025 — a rounding error in Google’s world but proof that a premium, human-curated or at least AI-abstinent experience can find a floor. The HN thread suggests more builders are betting that floor can become a ceiling.

But here’s the tension: many of those same builders are using AI tools to build the [NO-AI] products. Copilot writes the code for the search engine that promises no AI-generated results. It’s not hypocrisy — it’s the logical endpoint of a world where the toolchain is saturated and the product layer becomes the only place you can still make a philosophical stand.

What the Thread Doesn’t Say

What’s largely absent from the June thread is the thing that used to animate these posts: the unreasonable project. The thing that sounded slightly deranged and might not work. The hardware startup that required actual manufacturing. The consumer app aimed at people who don’t read Hacker News. The bet that wasn’t hedged by being a tool for the person two desks over.

The shift is rational. It’s also a loss. The monthly thread was never a representative sample of the economy, but it was a decent leading indicator of where ambitious technical energy was flowing. Right now, that energy is flowing toward making the act of building software marginally more pleasant for the people who already build software. That’s not nothing. It might even produce a few real companies. But as a temperature check on what the smartest people in the room think is worth doing, it suggests a generation of builders who have internalized the lesson that the safest thing to make is a shovel — even if everyone else is also making shovels, and nobody’s digging a hole.

“I don’t think half the people in that thread would call what they’re doing a business,” a former YC founder told me in a Slack DM while the thread was still climbing the front page. “They’d call it a project. And there’s a difference they’re very careful about now.”

The carefulness is new. So is the volume. Six hundred and forty-two comments is a lot of people showing up to say, essentially, “I’m still here, I’m still building.” The subtext is that staying in the game has itself become the achievement. Whether that’s resilience or a collective failure of imagination is the question the thread poses but cannot answer.

Sources appended by editor.

Sources