Is vibe coding killing indie hacking?

It is killing a moat. It is not killing the fact that someone will pay for a product that does a job they do not want to reassemble themselves. The obituary gets written because the moat was the part a lot of people thought the job was.

In February 2025 Andrej Karpathy posted the name. Vibe coding, in that post, meant describing what you want, accepting the generated code, and not reading the diffs. He framed it as fine for throwaway weekend projects. The name left the post. By 2026 it was how people talked about shipping almost anything small with an agent in the loop.

ChatGPT had already opened that door on 30 November 2022. Vibe coding was the slang for the loose end of a change that was three years old: people who could not write much code could still put a product in front of a stranger. Copying got cheaper in the same years. A working idea filled with lookalikes.

The moat Levels says is gone

On 30 July 2026 Pieter Levels wrote that indie hackers might be the first developers to feel AI take execution away as a moat. He put the new price of building at about $20 a month. He described customers vibe-coding a private copy instead of paying for a thin tool. Both the price and the customer behavior are his claim. Read them as an argument about defensibility.

The older “indie hacking is dead” line, from 2023, was about a subculture becoming normal. The 2026 line is about a skill. This site’s dead page keeps those two corpses apart. Extinct, in the later post, is not a headcount of people with products. It is “I can build it and you cannot” as a reason to pay.

If that reason was the whole business, the business is in trouble. A $29 wrapper around a prompt is a screenshot away from a free replica. Posting the roadmap in public makes the replica cheaper still. Build in public and vibe coding collide there: transparency feeds the clone. The founders who notice are quieter about the parts that are hard to copy, and louder about distribution they already own.

Rushab Jain’s February 2026 post says the same thing without the extinction costume. Micro-SaaS is not dying because demand left. It is dying, in his wording, because the barriers collapsed. Building is trivial. Old channels are tired. Users will not pay for a sloppy MVP. His complaint is about a playbook, and it matches the public record.

What the tools did to a real stack

The other Levels story is the cost side, and it cuts the other way. On 9 September 2026 he said he had replaced about a dozen paid tools and put the savings at roughly $25,000 a month. Scraping, maps, vision, moderation, support. The vibe-coded stack story is the receipts, all founder-reported. The manifesto a few weeks earlier was shorter: cancel the subscriptions, keep domains, servers, storage, and model APIs.

That is not the death of indie hacking. It is a solo operator using the cheapness on his own bill. The products still run. The vendors in the middle do not. A high claimed margin, later tied to the $10 million mix of revenue and investment gains, depends on that cost path in his telling. Vibe coding can raise the keep rate of a company that already has customers. It does not grant the customers.

Marc Lou’s portfolio shows the product-market version of the same shock. ShipFast sold the shortcut to a Next.js app with login, payments, and email. CodeFast sold the coding lesson. By March 2026 he said AI had hurt both, and he posted ShipFast at $9,200 and CodeFast at $8,300 in a month led by TrustMRR and DataFast. A Steal What Works look at ShipFast argues the steep drop started before vibe-coded app builders were a full substitute, and that the leftover pitch — saved setup time — got weaker as those tools learned auth, payments, and deploy. The boilerplate did not need to be murdered in one week. It needed to stop being scarce. Marc Lou’s record is a portfolio that moved, not a single product that vanished.

What did not get cheap

Distribution did not get cheap. If anything it got more expensive, because more products are asking for the same attention. The homepage era after AI shipping is verification, then bidding: proof, then paying to be seen. Ads and bidding are what a scene does when shipping is no longer the story.

Taste did not get cheap in the way a prompt suggests. Karpathy’s original framing was throwaway projects. A product a stranger trusts with a card is a pile of edge cases, copy, refusal states, and a reason to exist next year. Agents draft those faster. Someone still decides which draft is the product. The founders who treat the first working screen as finished are the ones the clone wave eats. The ones who sweat the job are harder to prompt-copy, because the spec is in the customer’s week, not in a tweet.

Data did not get cheap. A model can imitate a directory. It does not own the listings, the reputation, or the history. Levels’ nomad products are accumulated data with a front end. Postiz already had a publish path across networks when agents needed one. The agent-as-customer story is a pivot into a new buyer, not a claim that any scheduler would have caught the wave by prompting harder. Postiz was in the slot an agent looks for.

Trust did not get cheap. Revenue screenshots got less believable, which is why TrustMRR connects a read-only Stripe key. AI makes fake products and fake numbers easier. Buyers who have been burned by wrappers ask for proof, or they stay with a name they already use. Being new is a weaker advantage than it was when new meant “this took six months, so it might be real.”

Who is exposed

Exposed: a single-feature tool with no workflow depth, sold to a general internet audience, differentiated by the founder’s launch thread. Also exposed: a course or starter kit whose value was “you cannot set this up yourself.” When the buyer can set it up with an agent, the info product has to sell judgment, not steps. Some still do. Many were steps.

Less exposed: a product inside a job, with switching costs; a community or marketplace with its own liquidity; a brand people already search; a tool an agent must call because the networks, the auth, and the failure modes are already handled. Less exposed is not safe. A larger company can still enter. “Less exposed” means the weekend clone is not the whole threat.

Oversaturation is the supply side of this page. Vibe coding is why the supply can appear in a week. The two arguments are one argument.

What people are doing instead of mourning

They are shipping with the tools and charging for something the tools do not automatically have. Shipping fast is the ordinary 2026 method. The scarce sentence is why anyone comes back.

They are cutting their own SaaS bills the way Levels describes, when they have the stomach for the operational mess. That is a cost story. Copying it without his traffic is how people delete a vendor and discover they have rebuilt a worse one.

They are selling to the new user, when the new user is software. An MCP server, a CLI, a skill file: boring interfaces that make a product callable. Anthropic’s Model Context Protocol, announced 25 November 2024, is one public socket. The buyer might be an agent. A human still pays, in the Postiz telling. The category moved. It did not close.

Vibe coding killed the romance that the hard part was making the thing exist. The hard part is a customer who stays, a cost base that does not eat the month, and a reason the clone is not enough. That was always the business. The editor just stopped hiding it.

FAQ