The assistant shows vacancies as cards, with its reasoning on each one

Ask the assistant for roles and you used to get a wall of text with cards wedged into it. It wrote out the title, the company and the location, the card underneath repeated all three, and its actual reasoning — the part worth reading — trailed off in a loose paragraph below. Three blocks per vacancy, two of them saying the same thing, and prose splitting the list wherever the model felt like a sentence.

Now a recommendation is a set of cards and nothing else. They arrive as one stack under a heading, in the order the assistant ranked them, and each card carries its own reasoning inside its border: one sentence on what the role actually is, the concrete overlaps with your experience, and any real caveat it found. Nothing is repeated — the card already shows the title, the company, the location and how fresh the posting is, so the assistant spends its line on what those cannot tell you.

Two small things go with it. The assistant can no longer hand you a vacancy that does not exist: every posting is checked against the catalogue before its card is drawn, so an invented link cannot reach you. And when it separates a shortlist from a wider set, those are two labelled groups rather than a run-on list.

Everything on a card is still the real posting, pulled from the catalogue — save it, open it, track it exactly as anywhere else on the site. Try it on the assistant.