Every task a recruiter does, shaded by how much of it a machine can take off your plate in 2026. Of ~53 tasks, roughly 21 are mostly automatable, 17 about half, and 15 stay human - the closing, relationship, and accountability layer - plus 3 a machine can do but the law keeps a human on.
By the Team behind Glozo. Neutral map - Glozo is named only where it genuinely competes, always labelled.
desk
Click any task to see how far automation goes.
Agency full-desk — both candidate and client sides, closing-heavy. Employer branding & headcount planning are closed (not an agency job).
mostly humanmostly automatablesize = time spent closing thread~ nurture legally restricted (law keeps a human on the call)
Every task post on this map is built on the same ladder. "Automatable" isn't yes/no - it's how high you can climb.
Rung
What it means
Looks like
L1 · LLM one-shot
You prompt a model once and get a usable draft.
Paste a JD, get a candidate persona and search plan.
L2 · Playbook / project
A saved template, rubric or project runs the task consistently — you still launch it.
A saved rubric applied to every batch of profiles.
L3 · Full agent
A tool runs continuously with no touch — searches, matches, books, logs.
A scheduler that syncs calendars, books panels, handles reschedules.
Two different things: the map colour is how much of a task comes off your plate; this ladder is how it gets there. They are independent - a task can have an autonomous L3 rung and still be mostly human work (screening is the clearest case).
Where the ladder ends
Look across the map and a pattern appears: the tasks with no automation rung at all - pre-close, counteroffer, networking, live objection-handling - are the same kind of work. Someone has to own the outcome: hold the relationship, carry the liability, be the person the other side says yes to.
That is the spine of this map. Automation has eaten the mechanical, transfer-style work and keeps climbing into judgment-with-drafting. The durable human zone is not defined by what a machine can't do - machines keep getting better - but by who has to answer for the call. As capability rises, that zone gets more valuable, not less. Read the full piece on the accountability layer →
Common questions.
How many recruiting tasks can actually be automated in 2026?
Of ~53 distinct tasks, roughly 21 are mostly automatable today, 17 about half, and 15 stay mostly human — with 3 more a machine could do but the law keeps a human on. The automatable ones cluster in sourcing, scheduling, enrichment and admin. (Directional — see methodology.)
What recruiting work can't AI do?
Roughly 15 tasks — pre-close, counteroffer, negotiation, networking, live objection-handling, relationship coaching. Not because a machine can't attempt them, but because a human has to own the outcome.
What's the highest-leverage thing to automate first?
The three biggest time sinks that also reach L3: sourcing, resume screening, and scheduling. One scheduler or one notetaker often covers several tasks at once.
Is the map data exact?
The colour (how much a machine can technically do) is solid, grounded in shipping tools; the ⚖ flag marks where the law — not capability — keeps a human on the call. The size (time spent) is directional — four coarse tiers, one standard set, not measured to the minute — so the counts are ranges, not exact.
Methodology & honesty
Colour = technical capability - how much of a task a machine can take off your plate, grounded in tools shipping today. ⚖ = a separate legal layer, not a colour: where the law keeps a human on the decision. Size = directional - four coarse tiers, one standard set (we don't re-size per desk), not a stopwatch study.
This map is meant to be wrong in places and get better. If your desk doesn't look like this, tell us - we're running a short diary study with working recruiters to sharpen the time tiers.
Disclosure: Glozo is a talent-intelligence platform built by the author. This guide is neutral; Glozo is named only on the few tasks where it genuinely competes, and always labelled.