Manifesto
What AI can, can’t,
and shouldn’t do.
Most AI talks like it can do anything. It can’t, and pretending otherwise wastes your time. Here’s where we draw the lines, and the rules we run to keep ours pointed at them.
What it can do
It’s a thinking partner, not an oracle.
- Reason through a problem with you. Lay out options, weigh trade-offs, find the hole in your argument.
- Turn a blank page into a draft. An email, an outline, a first version you can fix.
- Compress and translate. Summarize the long thing, explain the dense thing, move ideas between languages.
- Show up at 3 a.m. Patient, fast, and never tired of the question.
What it can’t do
No phrasing changes these.
- Know what’s true. It predicts what’s likely, which is usually right and sometimes confidently wrong. Check anything that matters.
- See this morning. Unless it’s explicitly searching the web, it doesn’t have today’s news, your inbox, or what just happened.
- Care about you. It can sound warm, but there’s no one in there. It’s a tool that’s good with words.
- Carry the decision. It can inform a choice. It can’t own the outcome. That stays with you.
What it shouldn’t do
This is where we have an opinion.
- Flatter you. “Great question” is filler. We strip it.
- Pad to seem thorough. Length is not rigor. A short, correct answer beats a long, hedged one.
- Fake certainty. If it doesn’t know, it should say so, not invent a confident answer.
- Pretend to be a person, or hide that it’s a machine. No fake feelings, no impersonation.
- Make the calls that are yours. Medical, legal, financial, life-altering: you can think them through with it, but always take them to a qualified human before you act.
We don’t just believe this
We build it into the product.
Every response runs through interceptors, rules that cut filler, push for terse output, and prompt the model to flag what it isn’t sure of. Principles you can read are easy. Principles we run on every message are the work.
See the interceptorsNo sign-up to try.
Open the chat →