An AI agent that runs your job search end-to-end. Finds matched frontier roles, drafts tailored applications, schedules interviews, follows up. You set the brief in plain English. It does the work. You review and approve.
Cohire knows the application process at
// The premise
A job-match feed and a resume rewriter and an autofill button still leave you running the workflow. Open the dashboard, read the matches, click into the resume tool, draft, switch tabs, copy-paste, schedule the call, follow up next Tuesday, remember to ask about RLHF vs. DPO. Cohire is the agent that runs all of that, and just hands you the decisions.
// How it works
Cohire is built so you spend Sunday mornings reviewing decisions, not weeknights running the workflow. Setup is a 10-minute conversation; the rest is on the agent.
Link GitHub, Scholar, and any project write-ups. Cohire reads your real work -- not buzzwords -- and builds a working memory of your taste, depth, and direction.
Tell it in plain English what you want. "Senior post-training roles, $750K floor, Paris or remote, no hyperscalers." The agent asks clarifying questions until the brief is unambiguous.
Cohire scans matches nightly, drafts applications, schedules introductions, follows up. Every action is logged. Every send asks for your approval first.
Each morning (or Sunday, or whenever you set it) you see a short queue of decisions. Approve, edit, redirect, or tell the agent to drop a thread. It learns from each one.
// Capabilities
Each capability is wired to a specialized tool -- Resume AI, AI Job Match, Application Autofill, calendar, email -- that the agent orchestrates. You don't have to think about which tool to open.
Scans open and quiet roles across frontier labs every night. Scores against your taste profile. Surfaces only what's worth your attention.
For each pursued role, generates a Resume AI-tailored resume, a lab-specific cover note, and the right answers for every form question. Queues for your review.
Once you've approved an application, Cohire handles the coordination -- proposing times, juggling time zones, sending calendar invites, prepping you 30 minutes before.
Drafts replies to recruiters. Chases on day 7 if you've gone quiet. Negotiates scheduling with lab side. Every outbound message is approved by you before it goes.
Pulls relevant context from your work, the lab's recent papers, and the panel members. Generates a one-page prep brief 24 hours before any live round.
Every time you redirect the agent -- "shorter cover notes", "no London", "I'd prefer post-training over agents" -- Cohire updates its working memory. Future actions reflect it without you re-stating.
// Watch it think
Cohire is built like a careful engineer. Every decision shows its reasoning, every tool call is logged, every assumption is checkable. No black box, no surprises.
The agent's reasoning trace for one task. Each step is captured, replayable, and editable -- if you don't like an assumption, you correct it and the agent reruns.
ai_job_match.scan({"since":"7d","scope":"post-training","min_comp_usd":750000}) -- Returned 11 candidate roles across Anthropic, OpenAI, DeepMind, Mistral, Cohere.resume_ai.tailor({"role_id":1547}) for Anthropic role -- 12 edits -- confidence 96%. Repeated for OpenAI and DeepMind. All three resume drafts queued.app_autofill.prepare({"role_id":1547,"resume_id":892}). Populated 11 of 12 fields. One field flagged ("describe your most challenging trade-off") for user input.// Memory & context
Every conversation builds Cohire's working memory of who you are, what you want, and how you like to be helped. The next task is shaped by everything that came before.
Cohire keeps a structured memory of your preferences -- both the ones you've stated and the ones it has inferred from how you redirect it. You can read this memory at any time, edit any line, or wipe it and start over.
It separates preferences (cover note style, geography weight, comp floor) from facts (your stack, your recent work, your past matches) from open questions (things it's flagged as uncertain and wants to confirm).
The memory is yours. Cohire doesn't train shared models on it; no other user benefits from anything stored against your profile. Wipe with one command.
// By the numbers
Cohire is a newer feature. Numbers below reflect early-access network members from launch to date.
Median time saved per active job search week vs. unaided baseline.
Of drafted applications accepted by the user with no edits after review.
More frontier-lab applications submitted per cycle vs. the same engineer pre-Cohire.
Median agent cost per completed task. Covered by OpenTalent; free to members.
// Privacy & control
Cohire is a careful agent, not an autonomous one. Outbound action is always gated by your explicit approval. Your data is yours.
No application, message, calendar invite, or reply goes out without your explicit approval. The agent drafts and queues; you click send. This is structural, not optional.
While the agent is exploring, no lab or recruiter sees you're considering anything. Visibility flips only when you approve a specific outbound action toward that lab.
Cohire's memory of you is scoped to your account. We don't train shared models on it. No other network member benefits from anything stored against your profile. Read it, edit it, or wipe it any time.
Every decision the agent made -- every tool call, every assumption, every clarifying question it asked itself -- is captured and replayable. Export as JSON any time. Full retention policy in the Privacy Policy.
I went from running my own job search like a side project to checking a Sunday-morning queue of decisions. Eleven applications out the door in three weeks -- all of them tailored, all of them in my voice, all of them approved by me before they went. Cohire is the first AI tool that's actually saved me more time than it cost.
// FAQ
No. Approval is required before every send. The agent drafts, organizes, queues -- but the final "submit" or "send" action is yours. This is a structural product decision, not a toggle we'll remove.
The whole value of Cohire is removing the workflow drudgery while keeping you in the driver's seat for every word that goes out.
Those are the tools. Cohire is the agent that decides which to use, when, and in what order -- based on your brief and your memory. You don't have to open four tabs and manage the workflow; Cohire does that, and asks you for input only when a decision needs to be made.
If you prefer to run the tools manually, you can. Most network members use both modes.
It will make drafts and recommendations you can correct. Every step is logged with reasoning, so when you disagree, you can see exactly which assumption was wrong, fix it, and have the agent rerun.
The memory system means it learns from corrections -- telling Cohire "no London" once is enough; future runs honor it without you re-stating.
Cohire orchestrates calls across multiple LLM providers based on the task -- fast models for routing and search, capable models for drafting and reasoning. The choice is invisible to you, but cost-transparent ($0.07 median per task) and entirely covered by OpenTalent.
Yes. The memory page shows every preference, fact, and open question scoped to your account. You can edit any line, delete any line, or wipe the whole memory and start fresh. Cohire does not train shared models on your memory.
Free for OpenTalent network members. To join the network, apply through the five-stage screening -- less than 3% of applicants make it.
// The tools Cohire orchestrates
You can use any of these directly. Cohire just decides when to reach for which -- and chains them together for you.
Cohire is free for OpenTalent network members. Set the brief. Review decisions on Sunday. Get back to your real work.