Hey PH — Danny here, founder of Avery. Before I built anything, I noticed the real failure mode wasn't "managers don't get advice." A general AI assistant already gives decent advice. The gap shows up right where the job gets hard: it doesn't tell you how sure it is, it doesn't tell you when the call stops being yours alone, and it doesn't show its work. So Avery is built to answer the three things a hard people-conversation actually needs: what's going on (with the evidence attached), how sure are we, and when do you need to pull in HR before you act. I didn't want to just claim that's better, so we ran a real, pre-registered eval: the same manager's case, the same evidence, given to Avery and to a general AI assistant, blind-graded by two independent judge models. The honest result — read the actual side-by-side here: [LINK: landing /eval section — pending confirmed URL, see gap above] What we found, straight: the general assistant gave genuinely warm, sensible advice. The gap wasn't kindness. It didn't state a confidence level, didn't flag when the situation needed HR, didn't show its evidence. Avery does all three, on the same case. One thing we won't do: dress this up as a win rate. The eval method is real and pre-registered; the human ratings aren't in yet, so there's no score on the page until we've actually earned one. Would love feedback from anyone who manages people day to day — especially whether the honesty framing here helps or just reads as another launch disclaimer.

Avery — the senior at your ear

A senior advisor that shows its work and knows its limits

A senior advisor that shows its work and knows its limits