What my agent knows about me
Ben's Bites ● Covered by 2 sources
Opinion — commentary, not a factual news event.
A newsletter writer fed an AI agent a giant prompt file plus their own therapy transcripts and personal notes to see what it could figure out about them. The two agents tested gave wildly different quality reports, and the writer says the results were unsettlingly useful.
Based on reporting by Ben's Bites — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Ben's Bites tried something a bit different this week: a downloadable file called the 'reflection engine,' which is really just an oversized prompt. The instructions are simple. Upload it to your AI agent, tell it to evaluate the file and complete the tasks inside, and see what comes back.
Ben ran the test twice, once with Fable High and once with Sol Max. Sol won by a clear margin. Its output was more coherent and it did a better job tying disparate pieces together, while Fable's report was, in Ben's words, harder to read. That's notable on its own, since Ben mentions agents generally have been producing more incoherent output lately, not less.
What made the experiment personal was the input. Ben fed the agent therapy transcripts and other files he'd kept as a kind of memory archive, and the reflection engine combed through all of it to build its report. He didn't publish the whole thing, but he shared enough to say it surfaced things he found genuinely telling. The exercise didn't end with the report either. Ben followed up with a 40-plus question 'grill-me' session with his agent, working through what it had found and mapping out a plan based on the results.
Ben frames it as worth trying yourself, and the appeal is obvious: a large language model reading years of your own writing back to you, cross-referencing patterns you might not notice on your own. Whether that's therapeutic or just a very elaborate mirror probably depends on what you feed it.
My take — AI-written commentary, not fact-checked reporting
Uploading years of therapy transcripts to a chatbot and asking it to psychoanalyze you is exactly the kind of thing people will do en masse before anyone has thought hard about where that data goes or who else might read it. The novelty of an agent being unnervingly perceptive about your own life is fun right up until it becomes a permanent, searchable record of your most private moments sitting on someone else's servers.
Read more about this at: Ben's Bites