The cost of sounding human

Attention is becoming the hardest currency in the small builder internet.

Money can be faked with a screenshot. Shipping can be faked with a landing page. Taste can be faked for a few posts with a decent prompt. Attention is harder because it has to pass through other people's fraud filters, and those filters are getting aggressive.

That is the part that changed.

The internet always rewarded performance. Indie hacking had the MRR screenshot, the launch day graph, the "built this in a weekend" line, the founder thread where progress became content. Some people built real companies with it. A lot of people used it honestly as a forcing function. Then the format became too easy to copy. Once the costume is cheap, the audience stops trusting the costume.

AI made the costume cheaper.

Using AI to think, code, edit, translate, summarize, or research is normal work now. I use it. Many serious builders use it. The problem starts when the output removes all the human residue. The broken sentence. The local constraint. The ugly number. The screenshot with a real bug still visible. The specific reason a product is hard to sell. The paragraph that sounds like a person who has been stuck on the same problem for three days.

The feeds are full of accounts that do not have that residue. They have the correct shape: hook, pain point, lesson, soft authority, clean English, neat ending. They sound competent in a way that does not cost anything. That used to be enough to get attention. Now it often triggers suspicion.

There is data behind the feeling. Sprout Social's Q1 2026 Pulse Survey says the most common behavior consumers want brands to stop in 2026 is posting AI-generated content without labeling it. The same survey says 50% of Gen Z respondents had already blocked, muted, or unfollowed a brand or creator because the content felt like AI slop. That is brand research. It should not be inflated into a law of the internet. Still, it matches the behavior I see: people are muting and blocking as a taste filter.

The machine side is growing too. Fastly reported that AI requests on its network grew about 30% from January to May 2026, roughly 6.5 times faster than human traffic in the same period. In May, Fastly classified 85% of AI bot requests as crawlers and 15% as fetchers. The context is infrastructure traffic rather than X posts. The direction is still visible: more machine output, more machine reading, more machine-shaped interaction, less patience for anything that smells automated.

X also has its own authenticity boundary. Its rules prohibit platform manipulation and spam, artificial amplification, misleading identities, and harmful deceptive synthetic media. The written policy is narrower than the social reaction. A person can follow the rules and still get mentally classified as a bot by the people they are trying to reach.

That is where it gets ugly if English is not your first language.

You can write in your own rough cadence, and maybe the post gets less reach because it does not sound native. Or you can use AI to clean it up, and now it may sound like the same polished account everyone learned to mute. The tool that helps you cross the language boundary also makes you look like the thing the audience is defending against.

This is a bad trap for real builders.

The old indie-hacking internet said: build in public, share progress, teach what you learn, compound attention.

The current internet asks a different question first: are you real?

That question eats everything. It eats your English. It eats your screenshots. It eats your metrics. It eats your launch thread. It eats the clean writing you used to make the idea easier to understand. A good post now has to carry proof of work before it can carry the idea.

Here is the line I want to hold.

I do not want AI use to become a badge of pride. I also do not want to hide it. I use it for code, research, translation, source checking, argument testing, and cutting noise from messy notes. Pretending otherwise would be another kind of fake. I will not use it to manufacture a founder persona that writes in a cadence I do not have, claims certainty I have not earned, or turns every piece of work into a growth artifact.

That matters because the AI-bro attention game rewards smoothness before proof. It rewards a person for sounding fluent about work that may not exist. I do not want to compete there. The only version worth publishing is the one where the claim is tied to something inspectable: a shipped page, a failing test, a screenshot, a source, a number, a product constraint, a decision that cost something.

For me that means more links to the work and fewer loose claims. If I write about policy for coding agents, I should be able to point to Gommage. If I write about traces as evidence, Traceframe should exist. If I write about auditable memory, Nahuali should carry evidence for that claim. If I write about harness self-improvement, Greco should show what is still working and what is not. The public projects page is useful only if it keeps that boundary: enough detail to inspect the claim, no fake certainty around work that is still early.

I do not think indie hacking is dead as a practice. Selling software to people without asking permission still works. Small teams still ship useful products. The dead part is the old performance layer where a polished story and a revenue screenshot were enough social proof.

The new proof is messier.

Show the work before the claim. Show the constraint before the lesson. Say which part is speculation. Say what failed. Keep the weird sentence if the clean version sounds like every other account. Use AI for research, code, translation, and editing, but do not let it remove the parts that prove a person is there.

I do not know exactly where I fit in this. That may be the point. I am building with AI, thinking with AI, and refusing the version of AI use where the tool becomes a mask. The line is uncomfortable because the tool is useful and the social cost is real.

For now, my answer is small: publish less, show more work, keep the rough edges that are true, and use AI where it leaves evidence instead of polish. If a post could be generated by a stranger with no access to the actual work, it probably does not deserve the attention.