I'm becoming AI-blind: an engineer on why AI-written documents stop being readable
Rafal Cymerys writes that he keeps failing to process documents colleagues send him. The common thread, he says, is a strong trace of AI — and his own ability to spot it was trained by the same tools.
A short essay about reading at work has struck a nerve. Rafal Cymerys writes that he keeps catching himself unable to process documents other people send him: it feels as though he is reading, but his brain refuses to analyse the content, and he ends up in long back-and-forth exchanges asking about points that the document already covered.
Because he had spent the past year trying to relearn focus, the pattern worried him. So he sat down to analyse the cases and found one common denominator — every document carried a strong trace of AI.
What the documents have in common
His examples are specific. A design document that looks like a copy-paste from Claude: it does cover the feature in question, but it arrives wrapped in model-specific lingo and analysis, with phrases such as "this cuts just through it" or "the first gate is real". A 20-page marketing concept deck that mixes a reasonable marketing strategy with technical architecture gibberish, pitching the idea as "it's not selling X, it's selling Y" while asserting that a Redis backbone redefines the product. A technical requirements document that describes a fairly simple concept at great length, much of it reading like someone's uncertain internal reasoning rather than a settled decision.
Vocabulary alone is not what gives these texts away, he argues: the flow of the sentences, and the habit of presenting every small detail as a breakthrough, do. His test case is memorable — if a document describes the checkboxes in an RBAC configuration view of an enterprise application, it should not be sold like the invention of fire.
Banner blindness, for text
Research often suggests that people are poor at recognising AI-generated writing. Cymerys disagrees, at least where low-effort output is concerned, and points to a summary of common patterns assembled by Florian Roth. He suspects he has effectively been "pre-trained" on AI-written LinkedIn posts, emails and websites — text that is full of words and empty of meaning — and that his brain now spots the signs and skips ahead without much deliberation. Others have compared this to banner blindness, and he finds the analogy apt: with so much content pushed at people, filtering it out is how anyone stays sane.
The unexpected cost
The irony he cannot shake is that the same technology meant to make him more productive is now slowing him down. He also records a detail showing the filter is not fully under his control: on the Baltic coast, hungry and walking past restaurants, he ignored one at a glance — and a minute later found himself wondering whether the place had really put up a photograph of a quiche with mould on it. He walked back to check.
SiTech — AI-powered web development
We build fast, modern websites and bring AI into real business workflows. Have a project or a question? We'd love to help.