Hiring Software Shouldn't Wait for You to Open It
The dashboard model assumes the user shows up. Hiring managers hire twice a year and have a full-time job in between. That assumption is why pipelines go stale.
Almost every B2B tool is built on the same assumption: the user will open it.
That holds for tools people live in — an inbox, a code editor, a CRM for someone in sales all day. It does not hold for hiring software, because the people whose judgement matters most aren't recruiters. They're engineering managers, sales leaders and founders who hire two or three times a year and have a full-time job in between.
For them, opening a hiring tool is a task competing with everything else on a Tuesday. It loses. And when it loses for four days, the strongest candidate in the pipeline accepts somewhere else.
This is the actual failure mode of hiring software, and no amount of dashboard improvement fixes it. The problem isn't that the dashboard is bad. It's that going to look at it is a step.
Pull Versus Push
A dashboard is a pull interface. It waits, holds state, and requires the user to arrive with a question. It optimises for the case where someone has already decided to engage.
An inbox is a push interface. It arrives, it's ordered by urgency, and it creates the obligation rather than waiting for one. Email beat every knowledge management system ever built on exactly this difference — not because it's well designed, but because it comes to you.
The reason hiring tools haven't worked this way is that push requires judgement. To send someone three things instead of thirty, something has to decide which three. Without that filtering, push is just notification spam, and notification spam gets muted within a week — which is worse than a dashboard nobody opens.
What's changed is that the filtering is now possible. A model that has read the scorecard, the profiles, the conversation history and the pipeline state can rank what deserves attention accurately enough that a hiring manager trusts the top of the list.
What "Agentic" Should Mean
The word is doing a lot of unearned work right now. Most of what gets called an AI recruiting agent is a sequence of scripted steps with a language model generating the text at each one.
A more useful definition: a system is agentic if it decides what to surface and when, without being asked. That's a narrower claim than "it does the hiring for you," and a more honest one.
The division of labour that works is straightforward. The system decides what needs attention, ranking across every active search by score, recency, candidate signals and how long something has been waiting. The system prepares the decision — evidence attached, comparison to alternatives, draft response ready. The person decides.
Step three doesn't get automated. Not because the model couldn't produce an answer, but because the value of the whole arrangement rests on a human standing behind the outcome — and because the person who'll manage the hire is the only one who knows what they're actually looking for.
The Interface Follows the Model
Once a system is deciding what to surface, three things change shape.
Setup becomes a conversation. If the assistant is making judgement calls later, it needs to understand the role properly first. Starting from a title, a posting link or a paragraph of description and having the AI build out the profile while you correct it produces a better specification than any set of fields, because you're reacting to concrete proposals rather than generating from nothing.
The assistant belongs everywhere, not in a corner. One that only exists on a setup screen is a wizard. The highest-value placement turns out to be inside candidate review itself — looking at a profile and asking how she compares to the others, or what's missing against the mandatory criteria, without leaving the card. That question comes up constantly in review, and answering it manually means opening four tabs.
Onboarding friction has to approach zero. This is the unglamorous one, and it matters more than either of the above. If outreach goes out from a hiring manager's own profile — which is what makes reply rates work — then every manager has to connect their account. If that requires creating an account, receiving an invitation and learning a product, a meaningful fraction never do it, and the search stalls before it starts.
Where This Goes
The logical endpoint is that the interface disappears into wherever the person already is. For most teams that's Slack: a recommendation arriving in a channel, evidence attached, approve or pass as two buttons.
At that point the product isn't a place you go. It's a colleague who brings you things and does the preparation — which, described plainly, is the job of a good internal recruiter.
The risk is worth stating. A system that decides what you see also decides what you don't. That puts real weight on the ranking, and on keeping the full pool one click away so nobody has to trust the filter blindly. Push interfaces earn trust slowly and lose it in one bad batch.
Umamy surfaces what needs your attention across every open search in one prioritised inbox — scored, ranked, and ready for a decision.