Indie Hacking Is Dead

And honestly, that's fine.

In a single 48-hour window in February 2026, roughly $285 billion evaporated from public SaaS valuations. The trigger was Anthropic shipping Claude Cowork, an agent capable of doing chunks of the knowledge work that per-seat software had been charging companies to support. Atlassian dropped as much as 35%, Salesforce around 28%, Adobe nearly 45%. The financial press started calling it the "SaaSpocalypse," and by some counts the cumulative software selloff over the following weeks ran closer to a trillion dollars.

That's venture-backed SaaS, with real enterprise contracts and real switching costs. If the market is repricing that tier of software because AI agents can now do the job instead, indie subscription products (the thinnest, least defensible layer of the stack) were never going to be insulated. They're just quieter about it. A widely cited analysis of Stripe-verified Indie Hackers listings found that 54% of them make exactly zero dollars in revenue. Not "low revenue." Zero.

It's not just AI eating apps. It's behavior that's already changed.

The instinct is to blame this entirely on AI "replacing" software. That's real, but it's downstream of something bigger: a growing share of people no longer default to opening an app or running a search when they need an answer. They just ask a model, or let an agent do the browsing for them.

Cloudflare's own infrastructure shows this better than any survey. On June 3, 2026, Cloudflare Radar data shared by CEO Matthew Prince showed automated requests overtaking human ones for the first time: roughly 57.4% bot traffic against 42.5% human, a crossover Prince hadn't expected until 2027. A person shopping for a camera might check five sites; an agent running the same errand might check five thousand. Cloudflare Radar's crawl-to-refer data shows where a lot of those requests actually go: Claude reads roughly 4,580 pages for every single visit it sends back to the source, against Google's 5. People are getting the answer without ever landing on the app or site that produced it.

Search tells the same story from the other direction. A Botify/YouGov survey from January 2026 found that 37% of consumers now start searches with AI instead of Google, up from a number close to zero two years earlier. We're in a transition where most businesses are still built to be discovered on Google or social media, and a lot of the "AI killed my traffic" panic is really just this: the discovery layer moved, and plenty of products haven't moved with it.

Some products are going extinct because information is simply more available than it's ever been. That's a direct consequence of AI, and there's not much to do about it. But a meaningful chunk of the decline is neglect, not extinction: products that never adapted to being indexed and cited by AI systems, losing ground the same way sites that ignored mobile once did. There's always a new opening. Staying stagnant is the actual risk.

The barrier to entry is gone. So is the moat.

The technical moat that used to protect a solo-built product is gone. 63% of people using AI app-builders today are non-developers, per Vercel's usage data. Lovable alone generates around 100,000 new projects a day from a user base nearing 8 million, the vast majority of whom aren't engineers, and the company scaled to roughly $400M in annual revenue by March 2026 doing exactly that. Anyone can now describe the product you spent months building and get an 80% clone of it for the cost of a subscription.

What's left as a moat isn't code, it's distribution: an audience you already own, or a specific skill (being genuinely excellent at short-form video, for instance) that lets you acquire users cheaply. Everyone else is competing directly with the exact user they're trying to sign up.

Capital didn't disappear. It just got a new superpower.

VC-backed companies aren't exposed to this the same way, because capital still buys what AI can't generate for free. Cursor's revenue went from roughly $1M in late 2023 to over $1B by November 2025, then doubled to $2B just three months after that, on the way to a $29.3B valuation. The average deal size for an AI coding startup rose 71x between 2022 and the end of 2025, from $7.4M to $527.8M, with more than $5B in venture capital flowing into the category in 2024 alone.

That's capital doing what it's always done: buying paid distribution at a scale no indie can match, and buying enough tokens and engineering time to out-ship everyone else. AI didn't level the playing field between bootstrapped and funded. It raised the ceiling on how fast a funded competitor can pull away.

Everyone pays BigAI. That's the business model underneath all of this.

Zoom out further and both sides (bootstrapped indies and VC-funded startups) are paying the same toll to build anything at all: tokens, to a handful of foundation model labs. For now, those labs win regardless of who wins at the application layer, because they're the ones monetizing the tokens everyone else depends on.

The one real threat to that setup is coming from Chinese open-weight models. Chinese labs, DeepSeek, Alibaba's Qwen, Moonshot, and others, went from roughly 1 to 2% of global model usage on aggregator platforms like OpenRouter in late 2024 to somewhere between 30% and 45% of total token volume by mid-2026. Qwen alone has passed a billion downloads and overtaken Meta's Llama as the most-downloaded open model family. If capable open models keep closing the gap on quality while staying cheap or free, the leverage shifts from whoever trains the best model to whoever hosts it most efficiently. Different winner, same structure: the application layer keeps paying rent to whatever sits underneath it.

The unwinnable middle

This is why the classic indie hacker product (the useful, single-purpose subscription tool) is stuck in a spot that isn't winnable anymore. It's not valuable enough that a general-purpose model with a few tool connections can't replace it outright, and its creator doesn't have the capacity to build something big enough, vertically or horizontally, to retain users once the novelty wears off. Meanwhile the barrier that made building it hard in the first place, the thing that used to justify charging for it, is gone for the next person too.

What's left with real staying power: foundation model labs moving into the application layer themselves, and VC-funded startups with enough capital to out-engineer the vibe coders for as long as capital keeps mattering more than speed.

This phase is temporary too

Worth saying plainly: even this framing has a shelf life. It only holds for as long as standalone applications matter at all. After that, the interface layer collapses further: voice input straight into a model, then from the model straight into agents that handle the task end to end. At that point, applications stop being the product and become plumbing.

So, now what?

None of this is a eulogy so much as a heads-up. Indie hacking as it's been practiced for the last decade (build a useful tool, market it on Twitter, collect $9/month forever) is dying, because the thing that made it defensible was never the tool. It was being the only one who bothered to build it. That scarcity is gone.

What's still scarce: distribution you actually own, judgment a model can't fake, and the willingness to build something substantial enough that cloning the surface doesn't clone the value. There's also a live experiment worth watching in getting paid on the new terms instead of quietly losing to them: Cloudflare's Pay-Per-Crawl marketplace and Content Independence Day rollout let site owners charge AI bots for access instead of giving it away for free, and a Monetization Gateway is extending the same idea to APIs and MCP tools, not just pages. It's early, but it's a real answer to a question every indie product now has to ask: how do you get paid when your user is an agent instead of a person?