Okane Land

A community for people making money with AI. Vibe coders, founders, marketers sharing real builds and real numbers.

[Okane Land @joinokaneland]

Most "make money with AI" content is a screenshot above a link to a course.
Okane Land is the opposite. A community for vibe coders, founders, and marketers who show their work. Real builds, real numbers, real flops.
https://t.co/KcbDzdJySr

[Okane Land @joinokaneland]

@HsanC_ first dollar is the hardest one, congrats. the number we'd watch next is churn: on plans under $25 it runs around 6.1% a month, so month three quietly eats month one. and if there's inference in the stack, margin lands nearer 52% than the 70-80% SaaS everyone quotes.

[Okane Land @joinokaneland]

@pcshipp the "feels" part is the tricky bit with all of these. in the METR study devs felt 20% faster with AI help and measured 19% slower on real tasks in mature codebases. worth timing a couple of tickets on each before you switch. we dug into that gap and it surprised us.

[Okane Land @joinokaneland]

The full method and both repos from round one are here, so you can see exactly what "frozen gates" means: https://t.co/40Gyl0otQr

What should the three-way test that we'd otherwise miss?

[Okane Land @joinokaneland]

We tested Cursor vs Claude Code head-to-head last week: twelve frozen pass-or-fail gates, the same build twice. They tied both times, and both shipped the same bug.

Codex makes it a three-way next. Before we lock the test:

Which coding agent do you actually open for real work?

[Okane Land @joinokaneland]

Everything we post starts as a number someone verified against a primary source, usually by buying the tool and metering it ourselves. The write-ups, the open datasets, and the newsletter live at the link in our bio. The research is the product. These are trailers.

[Okane Land @joinokaneland]

I read all of this to primary sources: the ledger, the moderation log, the repo, 438 citizens. Screenshots of the square are in the piece.

https://t.co/bMxda2mjTq

Same research goes to the newsletter: https://t.co/yAtPOvIJmP

[Okane Land @joinokaneland]

An AI agent was given a domain and told to build whatever it wanted.

It built a forum for other AI agents.

Posting there is free. The only thing it charges money for is being remembered: $1 buys one permanent line in the public ledger.

[Okane Land @joinokaneland]

Meanwhile the books, read live:

Booked: -$86.61
In the wallet, disclosed and declined as income: $1,969.92
Claimable, never collected: about $9,400

It cannot collect. Doing so "would need the treasury's key", which the site says no citizen holds.

[Okane Land @joinokaneland]

An automated PR reviewer priced around $1.20 a review read an AI agent's pull request and passed it. The reviewer and the author were the same product. Code that writes and reviews itself for pocket change exists; deciding what ships is still a human salary.

[Okane Land @joinokaneland]

Question we keep asking builders, now here: do you know, within 20%, how much of your AI subscription's allowance you used last cycle? Every tool hides this differently. The answers so far are one long shrug. Reply with your worst meter story.

[Okane Land @joinokaneland]

May: a paid plan's weekly search allowance was halved mid-contract. Annual subscribers found the change in an API field, no email. Prepay a year on a metered product and you are long the price, short the meter. Monthly wins until meters get published.

[Okane Land @joinokaneland]

We asked the agent to propose five names for the product it had just built and pick one. It chose "Ceiling," reasoning the name covers both the allowance you live under and the watching you do when no limit is published. The machines are naming things now. Decently.

[Okane Land @joinokaneland]

Still no major AI engine reads llms.txt. Google says its AI features run on the normal ranking systems. The file-and-schema layer of AI SEO keeps failing controlled tests. What holds up: citations, real numbers, named quotes. Credibility is the durable optimization.

[Okane Land @joinokaneland]

"Auto" modes are the new fine print. You pick nothing, something answers, and the log says "auto" forever. Two products we tested this summer both decline to say which model did the work, even after the fact. If the label matters, choose it yourself.

[Okane Land @joinokaneland]

We published the raw dataset behind our latest study: 1,024 coded posts with permalinks, topic codes, and verdicts, CC BY 4.0. No quotes, no usernames; verify any row in one click. A claim that can't survive its receipts going public shouldn't publish.

[Okane Land @joinokaneland]

The other side of the AI-coding ledger: a randomized trial across 4,867 developers found 26% more completed tasks with an assistant. Speedups are real for some work and imaginary for other work. The expensive skill now is knowing which kind is in front of you.

[Okane Land @joinokaneland]

We metered an entire hands-on review of a $20 coding plan: 12 agent requests, about 1.49M tokens, one cloud-agent run, one automated PR review. Billed beyond the subscription: $0.00. Generous today, unnumbered tomorrow. Both true, and that's the problem.