About

The AI Downside is a blog about the parts of the AI boom that the launch videos leave out: the pricing changes, the rate limits, the hallucinations that never quite got solved, the dark patterns, the lock-in, and the steady erosion of products that used to just work.

We are not anti-AI. Half of us use these tools every day. We are anti-hype — the gap between what is promised on stage and what actually lands in your account. Somebody has to write that gap down, so it may as well be us.

What you'll find here

Two kinds of writing, and it isn't all grumbling. There's the news and criticism — the pricing changes, outages, dark patterns and broken promises, documented as they happen. And there's the evergreen analysis: longer, calmer explainers on the questions that outlast any one launch — AI regulation, bias and fairness, copyright and training data, safety, and cyber security. The news tells you what just went wrong; the analysis explains why it keeps happening and what would actually fix it.

Both are held to the same bar: evidence first, fair to the other side, and useful to the person on the receiving end. Criticism without understanding is just noise, and we're trying to be the opposite of noise.

What we stand for

Evidence over outrage. Every claim points to something publicly documented — a changelog, a screenshot, a status page, a policy diff.
Pro-consumer. We measure products by what they do to the person paying the bill, not the valuation.
Critical but fair. We criticise decisions and practices, not people, and we say plainly when a company gets something right.
No sensationalism. If a headline needs a scandal that isn't there, we don't publish it.

Who writes this, and how

Everything is published under one byline — The AI Downside — because we would rather be honest than invent a newsroom of fake reporters. Articles are researched and drafted with AI assistance, then edited and fact-checked by a human against the sourcing standards in our writer guide before anything goes live. The confidence of a language model is not evidence; every checkable claim is verified against a primary source, and a person owns what we publish.

We are open about the AI assistance for the same reason we criticise everyone else's dark patterns: it would be hypocritical not to be.

Corrections

We get things wrong sometimes. When we do, tell us via GitHub Issues and we'll fix the post and note the change. Being critical of other people's mistakes only works if we own our own.

How it's built

Static HTML, CSS and a little vanilla JavaScript. No frameworks, no build step to deploy. The whole thing is open source on GitHub.

How it's funded

Ads and reader support. We run Google ads to cover hosting — that's the extent of the commercial relationship: no sponsored posts dressed up as articles, no selling editorial judgement to the highest bidder. If that ever changes, it will be labelled plainly, because we hold ourselves to the standard we hold everyone else to.

Prefer fewer ads on us? You can chip in via Buy Me a Coffee. Reader support is what keeps the editorial line independent of any advertiser. Entirely optional; the site stays free either way.