Poe's proposition is decent. Rather than picking a model provider and building against one API, you build a bot once and it can run on many underlying models, and Poe handles distribution to an existing audience and pays you when people use what you made. The creator documentation supports that properly, with a clear progression from prompt bots, which are essentially a configured system prompt, through server bots that run your own code, script bots and canvas apps. The bot protocol is specified rather than hand waved, and the fastapi-poe library is documented well enough to build something real.
Monetisation is the part most platforms are vague about, and here it is concrete, paying per message and per subscription, which at least lets you reason about whether the work is worth doing. Group chat arrived in November 2025 supporting up to two hundred participants, so the platform is still being developed. I have to be straight about the risk, because anybody considering building here should weigh it. On the twenty first of April 2026 Quora announced layoffs, with Adam D'Angelo saying that Poe must stand on its own financially.
That is a significant statement from a parent company about a product, and it should factor into a decision to invest weeks of work in bots that live only on this platform. Separately, the compute point economics have moved against users more than once, with the free tier's daily allowance reportedly cut sharply in early 2026 and the Premium tier's monthly points reduced later in the year at an unchanged price. I would treat the specific figures with caution because they come from third party trackers rather than an official pricing page, and the direction of travel is not in dispute. Creators building for an audience whose allowances keep shrinking have a harder time than the documentation implies.
The technical gap is evaluation, and it is the same one everywhere in this industry. The documentation covers building and shipping a bot thoroughly and says essentially nothing about knowing whether it works. Anyone running a bot for real users needs a way to tell whether a prompt change improved things, whether the bot fails badly on particular inputs, and what to measure when somebody says the answers got worse. Without that you are shipping on vibes, and on a monetised platform you are shipping on vibes for money.
The strategic caution is the broader version of the same point. Building here means renting distribution. Poe owns the relationship with your users, sets the economics, and can change both, and the documentation naturally presents none of this as a consideration. Three point two.
Clear, genuinely useful developer documentation with a real payment model, marked down for platform risk that a builder should think hard about and for the usual silence on evaluation.