Hex Technologies Hires Technical Sourcers at $265k Median to Find 'Undiscovered' AI Talent

Hex Technologies has embedded AI-assisted screening, fraud detection, and candidate enrichment into its technical sourcing pipeline, a human-in-the-loop system built to surface "undiscovered" engineering talent across AI, infrastructure, and product roles as the company scales past $100 million in backing from Sequoia, a16z, Snowflake, and Amplify. The move reflects a strategic shift toward skills-based hiring driven by rapid team growth and a market where generative AI rewrites a CV in minutes, fabricates project narratives, and polishes away the gaps that once signaled inexperience. Recruiting teams at high-growth startups now face a flood of applications that look perfect on paper but collapse under technical scrutiny, costing senior engineers hours interviewing candidates who never should have cleared the first gate.

Inside Hex's Human-in-the-Loop Pipeline

Hex's workflow operates as a continuous feedback loop. Incoming applications pass through AI tools that flag inconsistencies, verify identity markers, and score candidate profiles against role requirements. Recruiters then calibrate those outputs, adjusting thresholds and overriding false positives. No application is rejected solely on an algorithm's output, a policy explicit on Hex's careers page. Fraud detection has become a distinct layer: the rise of AI-assisted interview cheating, from off-screen coding tools to impersonators on video screens, forces companies to verify identity before technical assessment begins. As HTD Talent reports, proprietary verification software combined with internal engineering screens conducted by their team can verify candidates before they reach a hiring manager.

Candidate enrichment sits alongside screening. Rather than relying on LinkedIn keywords, the talent team maps GitHub activity, research publications, engineering blogs, and open-source contributions to build a fuller picture of technical depth. The Technical Sourcer role Hex is hiring for (salary range $120,000 to $150,000) explicitly requires the ability to "read a GitHub profile, skim a research paper or engineering blog, and understand what makes someone exceptional rather than just experienced." That literacy lets the team prompt AI sourcing agents with precision, turning vague role requirements into searchable criteria that surface candidates traditional keyword searches miss.

The stack is intentionally heterogeneous: Juicebox, LinkedIn Recruiter, Gem, GitHub, and Ashby all feed into a unified pipeline managed through the company's applicant tracking system. The sourcer owns the configuration: building prompts, templates, and outreach workflows the broader talent team reuses. Conversion and response rates are tracked, A/B tested, and used to pivot searches that go stale. New tools are evaluated continuously, with the expectation that what works this quarter may be obsolete next quarter.

This operational model reflects a broader shift: the technical sourcer is evolving from a Boolean-string specialist into a prompt engineer, data analyst, and domain expert rolled into one. Hex's job description calls for "fluency with AI tooling for recruiting" and "a point of view on where they help and where they don't." That skepticism is deliberate. As HTD Talent notes, bad actors are getting better at fooling the old system, and companies are building processes designed to respond without slowing hiring for the roles driving growth.

How the Map Gets Built

Hex's technical sourcer mandate makes the strategy explicit: "Map deep talent pools and competitive companies to surface 'undiscovered' candidates, not just the obvious ones on LinkedIn." Instead of filtering inbound applicants or running boolean strings on the same public profiles, the team builds maps of where specific engineering talent actually lives — competitor rosters, open-source contribution graphs, niche research communities — then uses AI tooling to traverse those maps at scale.

Each tool serves a different node in the discovery pipeline. Juicebox and Gem automate candidate identification and enrichment across public web sources; GitHub provides the technical ground truth (commit history, repository ownership, language depth) that a resume never captures. A traditional sourcer sees a keyword match; a technical sourcer sees a maintainer of a core dependency the infrastructure team uses daily.

The competitive mapping works because the AI infrastructure talent pool is both small and visibly clustered. SignalFire's data shows engineers leaving OpenAI for Anthropic at an 8-to-1 ratio and DeepMind at nearly 11-to-1, with Big Tech — Google, Meta, Microsoft, Amazon, Stripe — serving as primary talent pools for AI labs. When Hex hunts for an "AI Engineering Lead" or "Engineering Director, Agent Context," the relevant candidates are likely already building similar systems at those companies. The sourcer's job: translate hiring-manager calibration ("partner closely with recruiters and hiring managers to calibrate profiles in real time, then translate that into precise, searchable criteria") into a searchable competitor list, then run the AI stack against that list to surface passive candidates who aren't looking but might respond to technically credible, personalized outreach.

That outreach piece is where the "undiscovered" label earns its keep. The role requires designing "AI-assisted outreach that engineers actually respond to: highly personalized, technically credible, and human." Volume alone doesn't close; Hex pairs reach with a human-in-the-loop calibration loop so the pipeline stays aligned with the actual technical bar, not a proxy.

The result is a sourcing motion that looks more like market intelligence than recruiting. By continuously evaluating new tools ("continuously evaluate new AI sourcing tools and share what works, helping raise the bar for how the whole Talent team sources"), Hex treats the stack itself as a product it iterates. Each search refines the map, each outreach test improves the conversion model, each hire expands the internal network that seeds the next map. In a market where the best AI infrastructure engineers are already employed at the companies Hex competes with for talent, that systematic, tool-augmented mapping is the only way to reach them before they hit the open market.

Two Metros, Two-Thirds of the Talent

The numbers don't flinch. As of mid-2025, New York and San Francisco together hold nearly two-thirds of the U.S. AI engineering base: 65 percent, per SignalFire's State of Talent report, a threshold Tim Lockwood's November 2025 LinkedIn analysis confirms. The Bay Area alone accounts for roughly four in ten of all tech job openings tied to AI, both remote and in-office, CBRE's June 2025 Lightcast data shows. Job growth tells the same story: between 2021 and 2024, the New York metro added nearly 48,000 tech-talent roles, the Bay Area almost 37,000, the first and fourth largest gains in North America, per CBRE. Nationally, AI-related postings have nearly doubled from 11 percent of tech openings in mid-2022 to 20 percent in June 2025. In the Bay Area, that share is 42 percent. The infrastructure layer (GPU clusters, distributed training, model serving, data pipelines) is being built where the capital, the compute, and the senior practitioners already sit.

Remote work didn't die; it settled. Stanford's WFH Research shows work-from-home days stabilized at roughly a quarter of all paid workdays in the U.S. as of early 2025. Only 44 percent of surveyed workers said they would comply with a full return-to-office mandate; the rest would quit or look. Yet SignalFire's data reveals a counterintuitive pattern: "proximity over presence" has replaced the binary remote-versus-office debate. Companies aren't demanding five days a week. They're clustering around anchor days (two or three in-person touchpoints) and talent is moving to be within commuting distance of those anchors.

The migration data bears this out. Miami posted a 12 percent jump in AI roles, San Diego a 7 percent rise in Big Tech headcount; but both are niche gains. Austin saw a 6 percent drop in VC-backed startup headcount in 2024. Houston fell 10.9 percent. SignalFire attributes the Texas cooldown to lagging infrastructure, cultural mismatch, fluctuating housing costs, and a renewed emphasis on hybrid RTO policies that pull employees back toward traditional hubs.

Hex's own hiring board reflects the gravity. Its open roles, including Staff Software Engineer Backend (NYC or Remote US), Engineering Director Agent Context (NYC), AI Engineering Lead (San Francisco), and Fullstack and Backend engineers (SF or NYC), carry salary bands that tell the story plainly.

The compensation signals are clear: the highest-leverage infrastructure work still commands a Bay Area or New York premium, even when the contract says "remote eligible."

The talent pipeline reinforces the loop. New graduates now account for just 7 percent of hires across tech, down a quarter from 2023 and over half from 2019. At startups, new grads are under 6 percent. The Magnificent Seven have cut new-grad placement by more than half since 2022. Employers are filling junior-labeled requisitions with senior ICs — an "experience paradox" that favors metros dense with senior practitioners. Anthropic retains 80 percent of its talent; the same talent flow patterns persist. Those companies serve as the primary feeding pool for AI labs, and those labs are overwhelmingly headquartered in San Francisco and New York.

Remote flexibility expanded the addressable map. It did not flatten it. The infrastructure stack — hardware provisioning, low-latency networking, model optimization, secure data pipelines — still demands physical proximity to compute clusters, capital partners, and the thin layer of engineers who have shipped production systems at scale. That layer lives in two metros. Until the next paradigm shift moves the compute or the capital, the map stays drawn.

Gem and Juicebox: Two Bets on Automation

The AI recruiting platform market has split into two distinct architectural philosophies, and Hex's sourcing strategy sits between them. Gem has built a unified recruiting operating system (ATS, CRM, sourcing, scheduling, and analytics in a single platform) serving over 1,000 companies including Zillow, DoorDash, and Asana. Juicebox operates as an AI-native sourcing and outreach layer that plugs into existing ATS infrastructure, claiming 3,000-plus customers across major AI labs, Fortune 500 firms, and recruiting agencies with $10 million ARR and 20 percent month-over-month growth as of June 2026. Both target the same bottleneck: finding qualified candidates fast. But they diverge sharply on automation depth, data ownership, and outreach architecture: differences that illuminate where Hex's approach fits.

Gem's bet is consolidation. The platform includes unlimited AI agents at no extra cost, omni-channel outreach across email, InMail, and SMS, and unified analytics spanning the full funnel with complete ATS data integration. Its candidate rediscovery feature (surfacing past applicants from CRM and ATS records) now accounts for 46 percent of sourced hires, up from 26 percent in 2021. Gem integrates with more than 20 ATS systems including Greenhouse, Lever, Workday, and iCIMS, and offers SOC2, GDPR, and CCPA compliance with 99.9 percent uptime. Pricing starts around $325 per user per month; customers typically consolidate five to eight tools into Gem, cutting total recruiting tech spend by 30 to 50 percent while boosting recruiter productivity up to fivefold. The trade-off is implementation complexity. Gem's "meet you where you are" positioning (use it alongside your ATS or replace your ATS entirely) acknowledges that rip-and-replace is a heavy lift for enterprise teams.

Juicebox took the opposite architectural path. Built AI-native from day one, it focuses exclusively on top-of-funnel automation: agents that construct structured search queries without Boolean logic, automatically assess 800 million-plus profiles across 30-plus sources against role criteria, and execute dynamic multi-step outreach sequences that deliver up to three times more replies than manual methods. The platform integrates with 60-plus ATS and CRM systems and can be operational in 60 seconds. Pricing is transparent and modular: a free tier for testing, Starter at $99 per month billed annually, Growth at $129 per month, and an optional AI Agents add-on at $300 per month. Juicebox's agents handle sourcing, ranking, and engagement autonomously; a single recruiter can manage three to five times more active searches compared to manual work. Early customers like Silo generated their first qualified pipeline in under 30 minutes versus days or weeks with traditional methods. But Juicebox doesn't offer candidate rediscovery, limits outreach to email and LinkedIn, and its analytics stop at the top of the funnel. The handoff to ATS (copying activity, checking duplicates, reconciling notes, re-validating contact data) remains a friction point Gem explicitly designed to eliminate.

Hex's strategy differs from both. Rather than buying a platform or bolting on a point solution, Hex built an internal AI-assisted workflow: these tools feed a human-in-the-loop review process where the same policy applies. This mirrors the "decision-quality AI" framing Gem uses, but applied to Hex's own proprietary pipeline rather than a vendor's. Where Juicebox emphasizes autonomous agents that replace sourcer labor, Hex is hiring technical sourcers: roles requiring AI tool fluency, GitHub literacy, and research skills to validate engineering talent in deep technical communities. The distinction matters for AI infrastructure hiring. Juicebox's high-signal filters target impact and achievements across 800 million profiles; Gem's rediscovery engine mines existing candidate databases. Hex's approach (mapping competitor companies and non-obvious talent pools to find passive candidates with relevant skills) relies on human judgment augmented by AI enrichment, not replacement. In a market where nearly two-thirds of AI engineers concentrate in those metros, and where AI-related postings hit a similar proportion in the Bay Area, the sourcing bottleneck isn't volume — it's signal. Hex's bet: the signal lives in communities and codebases that neither a unified CRM nor an autonomous agent can fully parse without a technical sourcer who knows where to look.

The Sourcer Who Reads Code

Hex Technologies' funding didn't just fund product development; it triggered a hiring mandate that exposed a new class of recruiting role. The company's careers page now explicitly seeks candidates "fluent with AI tooling for recruiting: you've already used (or are eager to master) AI sourcing tools, agents, and automation, and you have such a perspective." That language signals a shift: the technical sourcer is no longer a junior researcher feeding reqs to recruiters. It's a hybrid operator who can prompt an LLM to generate a Boolean string, validate the output against a GitHub commit history, and decide whether a passive candidate's side project actually maps to the distributed-systems problem the engineering team faces.

First-party board data underscores the scale of the engineering build-out. As of the latest ingest, Hex lists 24 salaried roles with a median band of $265k: roles like AI Engineering Lead (San Francisco, $246k–$329k), Engineering Director, Agent Context (NYC, $262k–$348.5k), and Staff Software Engineer, Backend Platform (NYC or Remote, $221k–$348.5k). Each of those hires requires a sourcer who can read code, not just keywords. The old model — boolean search on LinkedIn, spray-and-pray InMail — fails when the target talent pool is nearly two-thirds concentrated in SF and NYC and the differentiating signal lives in a GitHub Actions workflow, not a job title.

This mirrors the broader inflection documented across the industry. The shift of the past three years has devalued traditional credentials in favor of practical, skills-based competencies: GitHub portfolios, internships, and demonstrated AI literacy now outweigh pedigree. AI tools accelerate the shift, but only if the human operator knows how to interpret the scoring. A sourcer who treats the AI output as gospel will surface false positives; one who treats it as a lead list and then cross-references commit frequency, issue-thread depth, and dependency-graph centrality will find the "undiscovered" engineers Hex's strategy targets.

Hex's own process makes the dependency explicit. The company states that "all AI-generated recommendations are reviewed by a member of our recruiting team before any hiring decision is made" — a human-in-the-loop guardrail that only works if the reviewer possesses enough technical fluency to audit the model's reasoning. That reviewer is increasingly the technical sourcer. They're the ones who notice when the model flags a candidate because of a keyword match but misses that the candidate's actual contribution was updating configuration files, not designing control-plane operators.

The compensation data reflects the premium. While the board doesn't break out technical sourcer bands separately, the engineering median of $265k sets a floor: a sourcer who can credibly evaluate a Staff+ backend candidate's architecture decisions commands a salary that approaches the lower end of the engineering band. Startups competing for the same AI infrastructure talent (Juicebox, Gem, and the wave of Series A–B companies in NYC and SF) are posting similar hybrid requirements. The role has become a leading indicator: when a company starts hiring technical sourcers with GitHub literacy and AI-tool fluency, it's signaling that its engineering hiring bar has moved from "knows the stack" to "can prove it in the open."

The resume is still a moving target. But the map Hex is building (code-first, human-calibrated, tool-augmented) is the first one that moves with it.

Working in frontier tech? Zero G Talent tracks the openings: see every open Juicebox role, browse frontier tech jobs, openings at Hex Technologies, and the people building the field.