One Vendor Screens 150 Employers — and Rejects One in Ten Applicants From Every Job

Five Roles, One Signal

Rollstack has five salaried openings on Zero G Talent, with two added in the past week, and the cluster reveals what frontier AI companies are actually filtering for. The Y Combinator W23 company, which automates the "last mile" between BI tools (Tableau, PowerBI, Looker) and presentation surfaces (PowerPoint, Google Slides, Notion) for customers including SoFi, Zillow, Whirlpool, and Tripadvisor, is staffing two distinct engineering tracks and a two-person sales pod. Three engineering roles, two sales roles, zero marketing or ops openings. That ratio signals a product still in build mode but entering a repeatable go-to-market motion. The screen candidates face next filters for exactly that profile.

Rollstack, backed by Insight Partners, Y Combinator, and a syndicate including Goodwater Capital, Notion, Kima Ventures, UpHonest Capital, and Roosh Ventures, has grown to roughly 30 people since 2022. Founders Saad Bahir, Nabil Jallouli, and Younes Jallouli run the team remote-first from New York. As of the latest board data, the board shows:

The dual TypeScript postings, separated by geography, not stack, show the team needs throughput on a known architecture, not exploration. The AI Software Engineer role, priced wider and distributed globally, reads as a bet on a new capability layer rather than core maintenance. The AE/BDR pair maps a classic enterprise sales motion: high-touch consultative selling fed by a high-velocity outbound engine. The geographic split is its own signal: Eastern US/Canada and UK/Portugal for core engineering; global for AI; Seattle-anchored for BDR; US/Canada for AE. This isn't "remote-friendly" as a perk — it's talent arbitrage calibrated to time-zone overlap with customers (US East Coast, Europe) and hiring pools where senior TypeScript and AI talent costs less than San Francisco. The BDR's Seattle requirement hints at a hub-and-spoke model for the outbound team, or a preference for in-person ramp on a specific tooling stack (Nooks is name-checked in the JD).

What the mix omits is equally revealing: no product marketing, no customer success, no DevOps, no data roles. Either those functions are covered, contracted, or deliberately deferred. A Product Marketing Manager role appeared on third-party boards in July at $120K–$150K; a Strategic Product Marketing Lead at 7+ years surfaced the same month. Neither is on the Zero G Talent board as of this writing.

AI Software Engineer: The Production LLM Practitioner

The AI Software Engineer role is the most technically specific. The Y Combinator posting calls for 3+ years of machine learning and NLP experience focused on generative models, plus proven experience researching, building, and fine-tuning large language models in production environments. That "in production" qualifier is the filter — it excludes researchers who haven't shipped and engineers who've only prompted APIs. The stack confirms the orientation: the generative-AI layer runs on OpenAI API, Gemini, LangChain, and Langfuse, while the surrounding product uses TypeScript/React on the frontend, Node.js with Prisma ORM and Temporal workflows on the backend, and Kubernetes on AWS deployed via Argo CD. Candidates who've built LLM agents for data analysis and content generation get bonus consideration, a direct signal that Rollstack is moving beyond retrieval-augmented generation into agentic workflows that act on BI data. The role also demands advanced Python skills alongside the TypeScript stack, a dual-language requirement that mirrors the industry's split between ML prototyping (Python) and product delivery (TypeScript).

Software Engineer (TypeScript): Two Geographies, One Bar

The two TypeScript roles share an identical technical spec but carry a $60K spread at the top of band (EUR 120K ≈ USD 130K vs. USD 240K). The UK/Portugal opening sits at EUR 60K–120K; the Eastern US/Canada role at USD 140K–240K. Both require at least 3 years of related professional work experience after graduation, Node.js backend experience, React frontend experience, cloud infrastructure and CI/CD fluency, and strong algorithms and data-structures fundamentals. The day-to-day work centers on building and maintaining integrations with BI tools (Tableau, Looker, Metabase) and content platforms (Google Slides, Docs, PowerPoint, Notion), plus hardening the update engine that keeps slides and documents in sync with live data. The posting emphasizes comfort with ambiguity and remote work. Rollstack is fully remote by design, with a "Zero silos, zero politics, zero unnecessary meetings" operating philosophy. The geographic split likely reflects hiring budget allocation and time-zone coverage for a team that already includes alumni from Amazon, Meta, Pinterest, Tesla, and AiFi.

Account Executive: Selling the "Last Mile" to the C-Suite

The Account Executive role carries the highest floor (USD 180K, as Zero G Talent's data shows) and matches the AI Engineer's ceiling (USD 240K, as Y Combinator's figures put) — a market signal that Rollstack treats enterprise sales as a lever as valuable as its core AI talent. The RemoteSource listing (dated July 28, 2026) frames the pitch: "Companies invest millions into dashboards and [Rollstack] automates the last mile into slides and documents." The customer roster indicates a mid-market to public-company ICP where deals run through security reviews (SOC 2 Type II, HIPAA, GDPR compliance are listed on the careers page) and require multi-stakeholder consensus. The role is remote across US and Canada; no explicit experience floor appears in the board data, but the compensation band implies a quota-carrying closer with a track record in data-infrastructure or SaaS sales cycles.

Business Development Representative: The Seattle Anchor

The BDR role is the only geography-constrained opening (Seattle, WA, with remote eligibility) and sits at the narrowest band (USD 85K–100K). That specificity suggests a deliberate hub strategy: a small outbound team collocated for ramp and culture, feeding the AEs. The board data doesn't publish a detailed spec, but the function is standard: top-of-funnel qualification for the same ICP, heavy on cold outreach, demo coordination, and CRM hygiene. At a 30-person company backed by Insight Partners and Y Combinator, this is likely the first or second BDR hire, a signal that Rollstack is formalizing its outbound motion after founder-led and inbound-driven growth.

What the Cluster Reveals

Five roles, two distinct engineering tracks, one AI specialization, and a two-person sales pod. The TypeScript roles outnumber the AI role 2:1, confirming that Rollstack's moat remains the integration layer — the "last mile" plumbing between BI tools and presentation surfaces, while AI is an acceleration layer, not the product itself. The salary spread (EUR 60K–USD 240K) maps cleanly to geography and function, not title inflation. And the 3-year experience floor on both engineering tracks filters for engineers who've shipped and maintained systems, not just built prototypes. For candidates, the playbook is legible: show production LLM work for the AI role; show full-cycle TypeScript/Node.js/K8s ownership and BI-tool integration scars for the software roles; show quota attainment in data-infrastructure SaaS for the sales roles. The screen isn't hidden — it's in the spec.

The Screen: How It Works

Nine in ten U.S. employers now run every application through an AI screener before a human sees it. Nearly every Fortune 500 company does. A Stanford HAI study tracking 3.4 million people across 1,700 postings found a single third-party vendor handling screening for 150 employers across 11 sectors. Market concentration is the hidden variable: when one vendor dominates, a candidate rejected by its model is effectively rejected by the whole industry. One in ten applicants who submitted four applications were rejected from all four.

The volume problem is real. Entry-level postings now draw nearly three times the applications they did in 2022. Large employers receive millions per year; a single posting can pull thousands. Recruiting teams cannot manually review that volume; resumes get seconds of attention, if any. AI screening promises speed, consistency, and scalability. It delivers a short list in minutes.

Three main types of screeners do the work. Keyword-based systems match terms from the job description. Grammar-based systems parse structure and syntax. Statistical models, the most common in frontier AI hiring, train on historical hiring data. They learn which job titles, schools, skills, and phrasing correlated with past hires, then weight those features for new applicants. The model improves each time a recruiter accepts or rejects its recommendation.

That feedback loop is where bias enters. If a company's engineering workforce has been predominantly male, the model learns to favor male-associated signals, even with gender stripped from the data. Extracurriculars, writing style, educational pedigree, geographic markers: all become proxies. LLMs infer gender, race, and socioeconomic status from names, schools, neighborhoods, and word choice. In Brookings simulations across three LLMs and nine occupations, resumes with white-associated names were preferred in five of six tests. Black-associated names led in fewer than one in ten. For Black men, the intersection was brutal: selected never, compared to white men's names.

The screeners also miss what matters. They filter out strong candidates whose resumes don't match the exact phrasing of a job description. Different terminology, unconventional structure, skills the model doesn't prioritize: all become disqualifiers. Qualities critical but hard to quantify: communication, adaptability, leadership potential — go unmeasured. Non-traditional backgrounds get disproportionately discarded.

Frontier AI companies have moved beyond basic keyword matching. Tools like Tofu build personalized screening agents from bespoke criteria: "PMs with 3+ years at Series A fintechs" or "Founding AEs at seed-stage startups." The system enriches every profile with company trajectory data, evaluating real experience rather than keyword density. A fraud detection layer, trained on four billion data points and 18 million analyzed profiles, flags manipulated submissions the moment they hit the ATS. Some applicants embed hidden instructions in white text to game the model. The arms race is live.

Human oversight exists in theory. Best-practice guidance from Fisher Phillips urges employers to use AI to inform, not replace, judgment, especially for borderline cases or unusual profiles. But the same Stanford study found that when a single vendor's recommendations are pooled across positions, adverse impact disappears. Break it down by position (the legally relevant unit) and the disparities reappear. The screen is where the decision effectively happens.

Regulation is catching up. New York City and Colorado mandate audits. Maryland, Illinois, Colorado, and NYC require applicant consent. California recognized intersectionality as a protected identity in September 2024. The EEOC has said existing anti-discrimination law applies equally to AI-driven selection. Employers cannot outsource liability.

The screen is no longer a gate. It is a filter trained on the past, optimizing for signals that correlate with who got hired before — and who didn't. Getting past it means understanding what the model weights, what it ignores, and where the human still has a vote.

The Playbook: What Moves the Needle

The cold application is effectively dead at frontier AI startups. A warm introduction from someone in the founder's network clears a two-day response; a cold application might wait two weeks or disappear entirely. That pattern holds across companies with fewer than 20 engineers, most VC-backed AI startups, where every hire directly affects the product and a bad hire isn't just a cost but a delay to the entire roadmap.

Founders aren't looking for "smart engineers" anymore. They need specific engineers who solve their specific, large-scale AI problem. Technical interview rounds for AI engineering roles have increased 40 percent over the last 30 days compared to 2022. The median time-to-hire for U.S. software engineers sits around 41 days, but lean startups where founders are involved in every hire can extend offers in two to three weeks when mutual interest is genuine. Founder-led processes that close inside six weeks win offers at twice the rate of slower loops.

The skills that actually move the needle have narrowed. Critical today: LLM engineering, including fine-tuning, prompt engineering for complex tasks, RAG system design, multi-modal integration, agentic workflows. MLOps and production AI: building and maintaining reliable, scalable ML pipelines. Data-centric AI: expertise in collection, labeling, augmentation, and synthesis for model improvement. Performance optimization: quantization, distillation, model compression, GPU programming (CUDA, Triton). Deep framework mastery in PyTorch or JAX. Reinforcement learning for agentic applications. Cloud infrastructure across AWS, GCP, Azure.

What doesn't work: generic data science (Python, Pandas, Scikit-learn, basic regression). Academic ML without application, knowing the papers but unable to implement or deploy. Pure software engineering without strong understanding of AI systems design and ML fundamentals. "AI enthusiast" projects like a simple classification model on Kaggle. Those are starts, but they won't land a role at a top-tier AI startup.

The evaluation itself is shifting. HackerRank's Chakra platform combines what used to be three rounds (recruiter screen, take-home assessment, follow-up interview) into a single session where candidates work through a real-world code repository with an AI assistant built in. The platform measures "AI fluency": how well a candidate frames a problem for AI, judges its output, and steers it toward a solution. Suspicious-activity flags ran roughly three-quarters lower in Chakra interviews than in traditional assessments. The logic: giving candidates access to AI reduces the incentive to secretly use outside tools during the interview.

PRDs and ADRs are the new source of truth for system behaviors. AGENTS.md files stuffed with best practices plus agentic review pipelines now govern how code gets written and reviewed. Candidates who treat code like assembly language, specifying intent in natural language, letting agents handle syntax, advance. Those who cling to line-by-line authorship stall.

The Pattern: Where Frontier AI Hiring Is Headed

Rollstack's five-role push reads like a condensed map of where frontier AI companies are placing bets. Technical roles sit beside commercial ones. The salary bands cluster around the premium the market now pays for people who can ship AI into production, not just train models in isolation.

That pattern repeats at scale. Microsoft committed $2.5 billion and 6,000 employees to a new subsidiary, Microsoft Frontier Co., built around forward-deployed engineers who embed with clients. Amazon put $1 billion behind its own FDE initiative. Anthropic and OpenAI both stood up FDE groups in May, partnering with private-equity firms, banks, and consultancies. The signal: the bottleneck has moved from model access to implementation capacity. Companies that treat AI as a business strategy — not an IT initiative — outperform. Frontier firms, the top 10% of AI users by monthly token volume, generate 8.3× the output tokens per active user of typical firms, up from 2.6× in January. They also deploy plugins at more than double the rate (21% vs. 9% weekly active users) and are extending usage beyond engineering into legal, sales, recruiting, and marketing, where weekly active Codex users grew 108×, 41×, 41×, and 26× respectively since February.

Hiring data bears this out. Ninety-five percent of frontier firms are hiring for new role categories (AI trainers, data specialists, agent specialists) and nearly half of leaders prioritize AI-specific upskilling. OpenAI's pre-IPO recruitment of Noam Shazeer and former White House AI policy official Dean Ball shows the same logic: frontier labs are staffing for governance, recursive improvement, and labor-market impact, not just research. Rollstack's account executive role mirrors the enterprise sales motion that Wells Fargo, Accenture, and Estée Lauder now run through AI agents, with three-quarters of searches handled by AI, response times cut from 10 minutes to 30 seconds, collections automated, consumer data consolidated.

Geography matters too. Rollstack hires across the U.S., U.K., Portugal, Mexico, Brazil, and Canada. Chinese open-weight models accounted for 41% of Hugging Face downloads this spring, surpassing U.S. models. Half the Fortune 500 use Hugging Face to deploy private and open-source models. A new repository appears every seven seconds on a platform hosting nearly three million public models and one million datasets. The talent market is global, and the companies winning are the ones that can integrate across borders and model families, not lock into a single provider.

The macro frame is a recalibration, not a crisis. U.S. employers cut 23,000 jobs in July, yet construction for AI data centers added nearly 20,000 roles and semiconductor manufacturing grew. Frontier firms report 4× better results in brand differentiation, efficiency, growth, and customer experience. Seventy-one percent are thriving versus 39% globally. Organizational factors — culture, manager support, talent practices — account for 67% of AI impact, more than double the 32% from individual mindset.

Rollstack's screen filters for exactly that integration capability. The TypeScript tracks signal product-engineering fluency. The AI engineer role signals model-to-production fluency. The commercial roles signal go-to-market fluency. Together they form the profile of a frontier firm: cross-functional, agent-ready, and hiring for the operating model that compounds intelligence faster than competitors can catch.

The next time Rollstack posts a sixth role, it won't be another engineer. It will be the product marketing lead the founders have been carrying themselves — the hire that turns a technical moat into a market narrative. The screen will have shifted again. The candidates who pass will be the ones who already moved.

Working in AI? Zero G Talent tracks the openings: see every open Rollstack role, browse AI jobs, the companies hiring, and the people building the field.