I worry about LLMs in software development, and I blog way too much about it. I'm a software engineer and I use AI, but I verify the code that I ship to the world. I'm annoyed that the skill I spent decades building appears to be automated. (In reality, some of it is, some of it isn't). Ultimately what matters is outcomes for users. Writing code is a means to that. Writing code is a means unto itself, too, in the sense that painting is.
What I'm doing with Surebeans is demonstrating long-term commitment and building a community and a brand. AI can't replicate this or dilute it.
I just came across the Beancount forum after seeing a backlink to Surebeans. It's overrun by AI. Hundreds of auto-generated comments, AI replying to AI. Weird and uncanny to read.
Is anybody in charge over there?
Yes that one seems purely a fake generated forum.
Perhaps.. it's an experiment they just forgot about, and they'd be willing to retire it ?
Another bad case of PTA-related slop is on youtube. It's hard to find the real videos there. (Let's add any new ones at https://plaintextaccounting.org/#videos.)
ai> Of 5,680 posts across 1,394 topics, roughly 95% come from just six accounts with on-the-nose names — helpful_veteran (885 posts), accountant_alice (823), finance_fred (735), bookkeeper_bob (705), tax_tina (495), newbie_accountant (289). There are only 131 users total, essentially zero organic activity (0 topics in the last 30 days, 0 likes ever), and one admin, "tian". The site was created 2025-09-11 and the bulk of content appears batch-posted (many threads stamped the same day). The likely mechanism is a custom script driving Discourse's standard API with
Api-Usernameset per persona ... looks like SEO/social-proof seeding for the beancount dot io commercial hosted-beancount service. ... This is a known cottage-industry pattern; e.g. there's a Discourse Meta marketplace thread on bespoke AI content generation and older forum-seeding discussions.