An AI recruiting agent that checks the work, not just the resume

2026-10-29 · Avi Agola, The Seattle Compute Company

One of our customers is a technical recruiting agency. Its recruiters used to spend their mornings opening the same tabs: the search platform, a candidate's LinkedIn page, their personal site, their GitHub. A Smith agent now does that loop and hands back a shortlist with the evidence attached.

This page walks through what the agent does, in order, and why a job like this needs accounts and a record of who did what.

What the agent actually did

1. 1.
   Takes the candidate list

   The recruiter points the agent at a list of candidates on the agency's search platform, in Slack. The agent signs in to the platform and reads the list.

2. 2.
   Researches each candidate behind the login

   Using computer use, the agent opens each candidate's LinkedIn profile and personal site the way a person would, and notes roles, projects and anything that doesn't line up.

3. 3.
   Runs their public code

   It clones the candidate's public GitHub repositories onto its own machine and runs them to check that the experience on the resume shows up in real work.

4. 4.
   Posts a shortlist with sources

   The agent replies in the Slack thread with a detailed shortlist. Every claim links to the page or repo it came from, so the recruiter can check it in a click.

Why the code check matters

A resume says what someone claims. A repo shows what they shipped. For technical roles that gap is the whole job of a screener, and it is slow work: find the repo, clone it, work out how to run it, see whether it does what the README says.

The agent does this on its own machine, not yours. Each Smith agent has its own isolated computer with a real browser, so running unknown code from the internet never touches a recruiter's laptop. The point of running the code, not just reading the README, is that the shortlist can say what actually worked.

Why this needs gated accounts and an audit trail

Most of what makes the shortlist good sits behind a login: the agency's search platform, LinkedIn profiles beyond the public summary. That is the same reason a plain scraper or a chat window can't do this job. The agent has to sign in, and signing in raises two questions that a recruiting firm has to be able to answer.

Which account was used? For LinkedIn, the profile itself is the asset. It has a network and a reputation, so the agent works under the recruiter's own account (delegated identity). The password stays in an encrypted vault. The vault fills in the sign-in on the right site, so the agent never sees it. For tools where the account is interchangeable, such as GitHub, the agent can have an account of its own (agent-owned identity), which you can revoke without touching anyone's login. Delegated vs agent-owned identity for AI agents covers the tradeoff.

Who looked at what? Candidates are people. When a recruiter, a client or a candidate asks "who viewed this profile, and why?", the answer needs to be a record, not a guess. Smith traces every action to the agent that took it and the account it used. You can see which agent viewed which profile, under which account, and when.

You also keep a hand on the wheel. You can stop an agent at any time, and you can take over its live screen when a step needs a person.

The small LinkedIn connect workflow

The same setup also covers a smaller job: a LinkedIn connect workflow. LinkedIn is the textbook delegated-identity case, because the profile is the asset, so the agent works from the recruiter's own account with the password kept in the vault. It is logged like everything else, so there is a record of every request sent and which account sent it.

How the team gets work to it

Nobody opens a new app. The agent has its own Slack handle and email address. A recruiter tags it in the channel where the team already talks, drops in the candidate list, and reads the shortlist in the thread. Follow-up questions ("what did you find on candidate three's second repo?") go in the same thread.

If you want to set up a whole recruiting team, with a sourcing agent, a screening agent and an outreach agent that hand work to each other, read how to set up a recruiting agent team in Smith. To see how research hands off to action, read research, then act.