# Eightfold's AI Teammate Needs a 5-Line Brief. One Line Finds 552 People

_Eightfold's AI Recruiting Teammate takes a plain-English goal. A title-only goal finds 552 profiles. Five brief lines reach the 58,344 with the skills._

- Canonical URL: https://www.refolk.ai/blog/eightfold-ai-recruiting-teammate-five-line-sourcing-brief
- Published: 2026-11-22
- Reading time: 11 min
- Keywords: Eightfold AI Recruiting Teammate, AI recruiting agent prompt, how to write a sourcing brief, plain language candidate search, AI sourcing agent intake

Give Eightfold's new AI Recruiting Teammate a one-line goal like "senior backend engineer" and the job title is all it has to work with. In Refolk's index of professional profiles, a literal title search finds 552 US people. Search on the skills instead and you reach 58,344 senior US engineers.

Eightfold announced AI Recruiting Teammate on Nov 12, 2026. You give it a hiring goal in plain English, and it runs sourcing, screening, scheduling and follow-up until it reaches that goal. That makes the goal your search query. The five lines you write into it decide who the agent can reach at all.

## What is Eightfold AI Recruiting Teammate, and is it live?

Eightfold AI Recruiting Teammate is a goal-based recruiting agent. You describe the outcome you want in plain language, and it handles sourcing, screening, scheduling and follow-up until it gets there. Eightfold announced it from Santa Clara on Nov 12, 2026. It is still in preview, and customer design partnerships are due to start in late November.

A goal-based recruiting agent is software that takes an outcome ("fill this req") instead of a list of steps and works out the steps on its own. Eightfold's pitch is consolidation. Recruiters, it argues, are stuck briefing each tool separately and then stitching the work together by hand. One agent fixes that. But you still have to brief the one agent you have left, and that brief now decides everything downstream.

Eightfold isn't the only company building this way:

- **Eightfold** has raised $410M from Lightspeed, General Catalyst and IVP, at a $2.1B valuation. CEO Joe Korngiebel is quoted in the launch release.
- **LinkedIn** says more than 20,000 companies use its agentic hiring products, including AMD, Cisco, Ochsner Health, Palo Alto Networks and Zillow. The current version is covered in my earlier post on [Hiring Assistant 2's crowding problem](https://www.refolk.ai/blog/hiring-assistant-2-connected-apps-source-engineers-outside-linkedin).
- **Fountain** released Orchestra, which lets companies build hiring and scheduling agents by describing the work in plain language. It's aimed at frontline hiring.

Each of these products starts from a sentence you write. The quality of that sentence now matters more than the quality of any Boolean string you used to write.

## Why does a one-line goal find the wrong people?

A one-line goal gets read literally, so the job title ends up doing the filtering. Titles are the least reliable thing on a profile, which means the agent searches a small and skewed slice of the market and treats it as the whole market.

To test this, I ran one role three ways against Refolk's index: a senior backend engineer in the US who works in Go and Kubernetes.

| Brief style | Filter | Matching US profiles |
|---|---|---|
| One-line goal (title only) | Title "Backend Engineer" | 552 |
| Evidence of skill | Senior + Go AND Kubernetes, any title | 58,344 |
| Skills + synonyms | Senior + Go + Kubernetes + 4 title synonyms | 1,334 |
| Skill-based vs title-only | 58,344 ÷ 552 | ~106x |
| Skilled pool kept by title synonyms | 1,334 ÷ 58,344 | ~2.3% |

Two caveats. First, the title-only row has no seniority filter. A like-for-like version would shrink that 552, so the real gap is larger. Second, title matching in the index is strict. Read these numbers as a picture of how literal title filters behave, not as a headcount of the whole market.

> **106x** More senior US engineers found by searching Go and Kubernetes skills than by searching the title "Backend Engineer"
>
> 58,344 profiles versus 552 in Refolk's index, and the title-only count wasn't even limited to senior people.

The cause is simple. An agent with nothing else to go on uses the cheapest signal that matches your words, and that signal is the title field. "Backend Engineer" is a title some companies use and many don't. Top employers among the 552 include Spotify, Walmart, IBM and Tatari. They're real engineers. They're also a small, arbitrary sample.

## Why do the best engineers have the least descriptive titles?

Frontier employers use flat titles that say nothing about the work. So a title-based goal ranks the most sought-after engineers lowest, or misses them entirely. "Member of Technical Staff" (MTS) is the clearest example: one title that tells you nothing about what the person builds.

In the skills-only query (senior, Go, Kubernetes, any title), the most common titles were Member of Technical Staff, Principal Software Engineer and Senior Software Engineer. Top employers included OpenAI, Anthropic, AWS, Datadog, Cloudflare, Cisco and Palo Alto Networks. None of the 25 profiles I sampled held the title "Backend Engineer".

Next I added the synonyms a careful recruiter would list: Backend Engineer, Software Engineer, Platform Engineer and Member of Technical Staff. That gave 1,334 profiles, and 24 of the 25 sampled were MTS. Seven were at Anthropic and five at OpenAI. Others were at Replit, Modal and Databricks.

> **2.3%** Share of the skilled senior Go and Kubernetes pool left after adding four title synonyms
>
> Synonyms help, but used as a hard filter they still cut 58,344 people to 1,334.

That's the finding most intake advice misses. Synonyms are necessary, but they can't stand in for evidence. A brief that leans on titles, even well-chosen ones, throws away most of the people who can actually do the job. The "evidence of skill" line has to do most of the work, and titles should only nudge the ranking.

## What are the five lines every AI sourcing brief needs?

A good brief for an AI sourcing agent has five lines: must-haves, dealbreakers, synonyms, evidence of skill, and where the person shows up. A sourcing brief is the written definition of who you're looking for, specific enough that a person or an agent could pick the right profiles without asking you.

1. **Must-haves:** two or three things you can check, not adjectives.
2. **Dealbreakers:** what rules a candidate out, however good they look otherwise.
3. **Synonyms:** the other titles and terms the same person might use, treated as soft signals.
4. **Evidence of skill:** what proof of the skill looks like on a profile or in public work.
5. **Where they show up:** the places this person leaves traces, such as repos, review threads and talks.

### 1. Must-haves

Keep these short and checkable. "Strong communicator" can't be checked against a profile. "Has run Go services on Kubernetes in production" can. If you can't say what would confirm a must-have, it's a nice-to-have, and the agent will treat it like one whatever you call it.

### 2. Dealbreakers

Dealbreakers are the "boundaries" Eightfold says its agent works inside. The release describes the agent handling routine work "within the boundaries each organization sets" and passing judgment calls to a human. If you don't write the boundaries, the agent fills them in from defaults. Typical dealbreakers: can't work US hours, currently at a client account, contract-only. If the agent scores candidates against these lines, check whether that turns sourcing into a screening step under [NYC's AI hiring law](https://www.refolk.ai/blog/nyc-local-law-144-ai-sourcing-screening-step).

### 3. Synonyms

List every title the person might hold, and say explicitly that titles are a ranking hint, not a filter. The 1,334 result shows what happens when synonyms are used as a gate. If you're hiring outside the US, the [native title forms and level traps](https://www.refolk.ai/guides/local-title-search-reference) for each country belong in this line too.

### 4. Evidence of skill

This line carries the brief. Spell out what counts as proof: a merged contribution to a Kubernetes operator, a production Go service named in a project write-up, on-call ownership of a platform. It helps to separate skills that are [claimed, practiced, or proven](https://www.refolk.ai/guides/skill-evidence-proficiency-tier-reference). A skill listed on a profile is the weakest kind of evidence. Shipped work is the strongest.

### 5. Where they show up

Tell the agent where this person leaves traces: GitHub repos, code review threads, conference talks, maintainer lists. For senior infrastructure engineers, I'd weight review activity at least as heavily as commits, which is the reasoning behind a [reviewer-first senior slate](https://www.refolk.ai/guides/reviewer-first-senior-engineer-slate). An agent that only searches one network can't use this line at all. [Refolk](https://www.refolk.ai/) is built for this line: you describe the person in plain English, and the search runs on skills and public evidence rather than the title field.

> Titles should nudge the ranking. The moment they filter, you lose most of the people who can do the job.

## Before and after: one goal versus five lines

A one-line goal returns the title pool. A five-line brief returns the skill pool, ranked by evidence. That's the difference between 552 people who happen to hold one title and a pool led by OpenAI and Anthropic.

**Before (the one-line goal):**

- "Find a senior backend engineer for our platform team."

**After (the five-line brief):**

- **Must-haves:** runs Go services on Kubernetes in production; has owned a platform or infrastructure service.
- **Dealbreakers:** can't work US hours; currently at one of our client accounts; contract-only.
- **Synonyms (rank, don't filter):** Backend, Software, Platform, Infrastructure Engineer, Member of Technical Staff.
- **Evidence of skill:** merged work in Go or Kubernetes-ecosystem repos; production systems named in talks or write-ups.
- **Where they show up:** GitHub, code review threads, infrastructure conference talks.

The after-version lets the agent start from a pool roughly 106 times bigger before it begins ranking.

> **Example search:** "Senior US engineers who run Go services on Kubernetes in production, any title including Member of Technical Staff, with merged work in Kubernetes-ecosystem repos on GitHub." - [run it on Refolk](https://www.refolk.ai/s/bkaqj26nr3)
>
> A ranked shortlist matched on skills and public evidence, with title used only as a ranking hint, so flat-title engineers at frontier labs aren't filtered out.

## If the agent asks clarifying questions, do you still need a brief?

Yes. An agent that asks questions still answers them from defaults when you skip them, and those defaults come from your job description and past hiring patterns. LinkedIn's Hiring Assistant "asks clarification questions, offers recommendations, and adapts based on context". It also pulls evaluation criteria from the job description's must-haves and nice-to-haves. But a job description is a marketing document, not a search spec.

What happens next is predictable. If nobody writes the dealbreakers line, the agent's defaults take its place. If the defaults come from the last search, the new search repeats the old one, title filter included.

Intake data points the same way. Metaview's 2026 AI & Hiring Alignment Report surveyed 505 recruiting leaders and hiring managers. Where AI is central to hiring, 68% of searches start with strong agreement on requirements. Where teams don't use AI, 49% do. My reading is that the gap comes from writing the criteria down somewhere both sides can see them, more than from the model itself.

Calibration drift is when a search slowly moves away from what was agreed at intake. Metaview traces most of it to one cause: the must-haves live in one recruiter's notes and in the hiring manager's head. On a hard-to-fill senior role, every round of drift means more sourcing time and another awkward debrief where the hiring manager rejects a slate that matched the brief as written.

A written five-line brief fixes this because it's shared. The recruiter, the hiring manager and the agent all work from the same text. When the hiring manager rejects a candidate, you can point to the line that let that candidate through and edit it, instead of arguing from memory.

## How should you judge what the agent sends back?

Judge the shortlist by who's on it, not by how short it is. A smaller review queue only helps if the right people are in it. LinkedIn reports that charter customers reviewed 62% fewer profiles in September 2025 and 81% fewer by January 2026, with InMail acceptance gains of 69% and 66%. Those numbers are company-reported, and none of them tells you whether the shortlist came from the right pool.

Hari Srinivasan, LinkedIn's VP of Product, describes the goal as helping recruiters "go from a billion people to that one perfect hire." That's the right ambition. But the table above shows a one-line goal can give you a short, confident shortlist drawn entirely from the 552 and none of the 58,344.

Three checks before you trust any agent's slate:

- **Title mix.** If every candidate holds the exact title you typed, the agent filtered on title. For infrastructure roles, expect to see MTS, Principal and Senior Software Engineer.
- **Employer mix.** For this role, a pool with no OpenAI, Anthropic, Datadog or Cloudflare profiles probably missed the flat-title segment.
- **Evidence per candidate.** Every shortlisted person should map to at least one item from your evidence line. If you can't see why someone was picked, the agent can't either.

For a proper comparison between agents, run them all on reqs you've already filled and score them against the people you actually hired. The [people-search bake-off playbook](https://www.refolk.ai/guides/people-search-tool-bake-off) lays out that process. Run the five-line brief through each tool and keep whichever one finds the people you already know are right.

## FAQ

### What should I type into an AI recruiting agent prompt?

Type the five lines, not the job title. Write two or three checkable must-haves, the dealbreakers that rule people out, a list of synonym titles marked as ranking hints only, what counts as evidence of the skill, and where the person leaves public traces. A one-sentence goal makes the agent fall back on the title field. In Refolk's index, that means 552 US "Backend Engineer" profiles instead of the 58,344 senior US engineers who list both Go and Kubernetes.

### How long should a sourcing brief be?

Five lines, each one sentence or a short list. Write mostly exclusions and evidence, not adjectives. Anything longer usually means you're copying the job description, and the agent already reads that. The brief's job is to add what the job description leaves out: what rules people out and what proof looks like.

### Should I include job titles at all?

Yes, but only as soft signals. Even four well-chosen synonyms used as a hard filter cut the skilled senior Go and Kubernetes pool from 58,344 to 1,334, about 2.3%. Frontier employers like Anthropic and OpenAI use the flat title "Member of Technical Staff," so a title gate pushes their engineers down the ranking. Tell the agent to rank on titles, never to filter on them.

### Does this apply to LinkedIn Hiring Assistant and Fountain Orchestra too?

Yes. Eightfold's AI Recruiting Teammate, LinkedIn's Hiring Assistant and Fountain's Orchestra all start from a plain-language description of the work. Whatever your description leaves out, the agent fills in for you. LinkedIn's agent, for example, draws its criteria from the job description. Whichever agent you brief, the five lines decide who it can find, and skipping them hands that decision to the defaults.

---

_From the Refolk blog. More posts: https://www.refolk.ai/blog/index.md_
