How elite AI teams get built | Adam Ward, Head of Talent at Cursor

Why does it matter? Because your recruiting process is built to find whoever was available — not whoever was best.

The 20% who respond to your outreach aren't the top 20% of talent — they're whoever happened to see your message on a bad day for them. Adam Ward scaled Facebook's tech org from 1,000 to 10,000 engineers, grew Pinterest from 200 to 2,000 employees, then co-founded a recruiting firm that Cursor acquired outright. His diagnosis: virtually every company is running a process structurally incapable of finding the people it actually wants.

• Most hiring is "remainder hiring" — filter everyone out, hire whoever's left, and miss the best candidates by construction • Executive search methodology (scope → map → pursue) applies to every hire, not just leadership roles • "Who's the best X you know?" is the worst referral question; one specific attribute question consistently surfaces better names • Cursor removed work samples from its process, immediately lost confidence in its decisions, and reversed course

Most hiring is 'remainder hiring' — and it's structurally designed to miss the best candidates

The math makes it unavoidable. Reach out to 100 people; 20% reply. But those 20 aren't the top 20% — "that is just the 20 people who you caught on a bad day for them," Ward says. Filter down at every stage and hire whoever survives: Ward calls this "remainder hiring," the structural output of the funnel of doom.

The borrowed metaphor is the culprit. Companies modeled recruiting on the sales funnel because sales has centuries of precedent and recruiting barely has fifty years. But a sales funnel assumes rational products and rational buyers; recruiting has irrational humans on both sides. Even with excellent interview hygiene, "you kind of regress to the mean over time as you get to volume."

Ward's inversion: build confidence in who the top candidates are before you contact anyone, then use assessment to validate that thesis. "What if you had confidence up front of who the top 20% was and then you use your assessment process to validate that thesis?" The funnel of doom filters down to a remainder. The alternative starts with a named list of the 50 people who could actually do this — and relentlessly pursues them.

Every hire deserves executive search methodology — scope, map, pursue — and most companies skip all three steps

Scope. Map. Pursue. These three steps define how the best leadership searches work, and Ward argues they apply to every hire — the gap between average and elite teams opens at the IC level.

Scoping is the cornerstone: before you talk to any candidate, stack-rank the skills and experiences the role demands and define what success looks like. "I'll know great when I see it" is a trap — without that upfront work, you won't. Scoping generates your pitch, your assessment rubric, and your closing argument. It is the most underinvested step in most hiring processes and the most foundational in a leadership search.

Mapping turns the spec into a target list. Use the attributes you defined — not company logos — to identify the 50 people who could do this. "These are objective, somewhat transferable things — versus trying to piggyback off of someone else's process or definition of great."

Pursuit is what happens when the list exists: patient, personal activation, knowing some seeds harvest in weeks and others take years. Plant them all simultaneously.

'Who's the best product engineer you know?' is the worst referral question — one specific attribute surfaces better names every time

Generic referral questions return the first name that comes to mind, not the right one. Ward calls "who's the best X you know?" the single biggest mistake companies make when tapping their networks. The replacement: "Of all the product engineers you've worked with, who stands out as the most collaborative with designers? Or who can translate a framework into a product better than anyone you've ever seen?"

Attribute-specific questions match how memory actually stores colleagues — not as ranked lists, but as standout moments. You don't remember "the best engineer." You remember who solved the thing nobody else could.

The mechanism that makes them powerful is triangulation. Ask the same targeted question across multiple trusted sources and "you actually start to triangulate around the same names." Convergence across separate networks is validation — if Lenny names someone and a second source names the same person in response to a different attribute question, that person belongs at the top of the target list.

Each referral question also maps back to the scoping work in step one, so you're collecting evidence that specific people have the attributes you decided matter most.

Work samples predict job success better than any interview — Cursor removed them, lost confidence immediately, and reversed the experiment

Cursor tried removing work samples from its process. The result was immediate: "We really struggled to have confidence in our decision." They brought them back.

Research has long established why: "Work samples are the highest predictor of success at a company. But yet most companies are still spending a lot of time doing one-to-one across the table assessments," Ward says. Cursor is known for multi-day on-sites where candidates work on real projects alongside the team — a significant ask, and the one they couldn't afford to cut.

The signal extends beyond technical competence. "Imagine the amount of core values and collaboration and other signals you get over a stretch of time that can be arguably more important than the specific technical skill." For non-engineering roles, Cursor runs equivalents: go-to-market hires work through a difficult-customer challenge scenario; PM candidates complete a product exercise alongside the hiring manager.

Most on-sites are directionally wrong — the company extracts, the candidate absorbs. Work trials flip it into a two-way experience that powers genuine self-selection.

'Caring is free' — the number one driver of candidate satisfaction costs nothing and any company can out-care its competitors

Across every candidate satisfaction survey Ward has seen, the top factor is never compensation or prestige — it's feeling genuinely wanted. "The company generally wanted me. They made it feel like I belonged there."

"Out-caring other companies — when I've built recruiting teams, that's something we've always talked about. We can care more about that candidate than anyone else. That's in us." The funnel of doom destroys this by design: when you're processing 100 people to find one, you get caught up in the numbers and lose the individual. Ward's reframe: "We're trying to hire one person for this role, not 10. Don't get caught up in the 10."

In practice: hand-pick interview panels based on shared interests, not calendar availability. Build every on-site around a meal. Learn what the specific candidate cares about before they arrive — and orchestrate the day around them. None of this requires budget. It requires knowing who you're actually talking to.

Closing a candidate is a team sport that starts at first contact — Cursor runs daily standups and individual Slack channels per high-priority hire

By offer time, the sell should already be done. "The goal is to not be selling at the close. We're trying to sell throughout." The offer should feel like the natural conclusion of a relationship built over weeks, not the opening of a negotiation.

For high-priority candidates, Cursor runs daily 10-minute standups — not about recruiting broadly, but about what the team is doing that specific day to move that specific person toward yes. Every high-caliber candidate gets their own Slack channel. Two days before this conversation, a team member discovered a candidate in New York was a classically trained violinist. They found a specialty instrument shop, bought a unique Turkish guitar, and presented it at the offer dinner. A few hundred dollars — negligible relative to the hire's expected impact. The thoughtfulness was the whole point.

Post-acceptance is a second recruiting motion. Renegs are rising as competing offers arrive after close. Cursor's countermeasure: pre-boarding dinners with other incoming hires, regular touchpoints, a Cursor laptop shipped before day one.

Hiring one 'head of talent' to both build the system and fill the backlog almost always delivers neither

System-builder and high-volume executor are opposing archetypes. Founders routinely collapse them into a single "head of talent" hire — and pay for it in missing roles and missing infrastructure simultaneously.

The structural problem: recruiters thrive at 90 to 110% capacity. "If they only have half a req load, they kind of are useless. But they're kind of great at this edge of adrenaline-rush busy." Drop a systems thinker into a 15-req backlog and they'll either stop building or stop executing. "The thing they try to do is do this silver bullet hire — a first head of talent — and they kind of collapse two competing skill sets into one."

Ward's fix: decouple explicitly before hiring. Bring in the system-builder and pair them with a contingent recruiter for near-term execution. "I'm going to hire the person who's going to put in the right system, but I'm going to partner them with a contingent recruiter to make some short-term progress." Which problem comes first should be a deliberate choice, not an assumption.

Forward Deployed Engineers are the most in-demand emerging role in AI — and the pipeline is full-stack engineers from three to five years ago

One role is materializing faster than the vocabulary to describe it: the Forward Deployed Engineer. Deeply technical, customer-facing, able to walk into an executive meeting and help an organization move from raw model spend to something deployed and optimized. "As CFOs start to get the bill coming in," Ward says, companies that can bridge that gap face a genuine supply problem.

The natural pipeline isn't AI researchers — that bracket is finite and well-understood. It's the full-stack engineers from three to five years ago who now find themselves at a threshold. When those engineers encounter FDE work, "it kind of unlocks something in them — they're moving out of the terminal and into the company, into the customers and clients." An unexpected second chapter.

What's waning: narrow deep specialists, "overly specialized in a way," at a moment when AI is collapsing distance between engineering, product, and design. New grads are softer too — tools like Cursor absorb some of what they'd otherwise contribute.

As AI timelines compress from months to days, the relationship machine is the only recruiting model that compounds

The throughline Ward doesn't quite name: recruiting is a lagging function in a market where everything else moves in days. Most of the advantages he describes — the targeting methodology, the work-trial signal, the relationship architecture — exist because the field hasn't caught up yet. That window won't stay open.

His specific tactics only work once you've abandoned volume thinking entirely. Companies that make that shift build relationships while competitors build pipelines.

Relationships compound. Pipelines don't.

---

Topics: recruiting, talent density, hiring, team building, AI companies, forward deployed engineer, candidate experience, work trials, referral networks, offer closing, early-stage hiring, Cursor