ATG’s Data Pipelines Role Requires 8+ Years Experience — Highest Bar in the Hiring Wave

The Real Scope of ATG's Open Roles

Autonomous Technologies Group is actively recruiting for nine roles across engineering, design, and operations as it scales its applied AI platform for financial markets. This hiring push reflects the company's transition from early-stage development to product deployment.

One role carries the title Research Scientist. It asks for six-plus years of experience, offers $200,000 to $300,000, and sits in New York. The description on BuiltIn mentions driving original AI research on new models and algorithms for reasoning in complex environments. That is the only listing framed around novel model architecture or publication-grade work.

A second role, Member of Technical Staff (Research Engineering), requires one-plus years of experience at $180,000 to $260,000. It bridges research and production but remains an engineering seat. The remaining seven positions span backend, full stack, iOS, data pipelines, quantitative systems, product design, and a client director role focused on high-net-worth relationships. Salary bands range from $115,000 for the product designer to $300,000 at the top end for the client director, quantitative systems, and research scientist roles. Experience floors start at one year for research engineering and climb to eight-plus years for data pipelines.

Five of the nine roles are core software engineering positions: backend, full stack, iOS, data pipelines, and quantitative systems. The quantitative systems seat, listed at $200,000 to $300,000 with three-plus years required, signals a need for engineers who understand financial modeling and production constraints, not just model training. The data pipelines role demands the most experience at eight-plus years, reflecting the operational weight of moving financial data at scale.

The client director role, posted on BuiltIn as recently as August 11, 2026, owns high-net-worth client relationships, onboarding, CRM build-out, and feedback loops into product and engineering. That is a go-to-market hire, not a research support function. The product designer, at $115,000 to $125,000, rounds out the product trio.

ATG's team size stands at eight people as of the Y Combinator listing dated August 13, 2026. Adding nine roles would more than double headcount. The distribution — seven engineering and product roles, one research scientist, one research engineer, one client-facing operator — matches a company shipping a financial advisor product called Autonomous, not one chasing AGI. The next section examines how ATG defines that work.

What 'Applied AI Lab' Actually Means at ATG

Autonomous Technologies Group describes itself as an applied AI research lab deploying frontier reasoning systems within financial markets. That phrasing appears on the company's LinkedIn page and its own site, atg.science, and it carries a specific meaning that separates ATG from both general AI research outfits and AGI-focused ventures. The distinction starts with the vertical constraint: every model, every data pipeline, every evaluation framework serves a single domain. That domain is financial decision-making for individual investors. There is no language research program, no robotics effort, no pursuit of general-purpose reasoning divorced from a balance sheet.

The company's own documentation breaks the work into three commitments. First, teach the system to deeply understand the market. This involves synthesizing quantitative and qualitative signals to surface causal forces driving behavior, not just price and volume. Second, teach it to deeply understand an individual. This means modeling preferences, goals, and constraints as a complete financial life, not isolated risk questionnaires. Third, advance representational capabilities. This is the ability to communicate what the system knows, calibrated to each person's level of understanding, across voice, charts, and written analysis. These commitments are not research themes; they are product requirements. The site states plainly that everything the system does follows from them.

This framing reveals what "applied" means in practice. ATG is not publishing papers on novel architectures for their own sake. It is building a registered investment advisor (an RIA under SEC and FINRA jurisdiction) with an Investment Doctrine that governs how recommendations are made across the full range of financial decisions. The platform must be SOC 2 compliant, FINRA-licensed, FDIC-approved, and fully SEC-compliant by launch. Those regulatory boundaries shape the engineering: every method is subject to rigorous testing and evaluation before it is trusted. The company plans zero advisory fees and no trading fees for early users, a business model that only works if the advisory layer itself is automated at scale.

The founders' background reinforces the applied focus. Dillon Erb and Daniel Kobran sold Paperspace, a GPU cloud infrastructure company, to DigitalOcean for $111 million in 2023. They did not come from an AI lab; they came from the infrastructure layer that makes AI training viable. After the exit, they spoke with financial advisors across firms and regions, found inconsistencies and high fees, and decided to automate the advisory layer itself. This was not another portfolio-balancing tool. Garry Tan, CEO of Y Combinator, put it directly: "The financial advisory industry is one of the last holdouts where human intermediaries extract massive value without creating it."

ATG's own site draws the line against prior automation attempts. Robo-advisors rebalanced portfolios but stopped short of real guidance. They hit a wall without real intelligence — cookie-cutter portfolios and little else. The largest robos sit under $100 billion in assets under management while traditional advisors still control trillions. ATG's bet is that the missing piece is not better allocation algorithms but a system that understands the market, understands the person, and can represent that understanding in a way that disappears into the background. The goal is invisibility: handle complexity so completely that the user can focus on living.

The roadmap makes the product timeline concrete. Today Autonomous is an application that onboards users into a system that learns their financial life and guides with precision. Public access is expected in early 2026 with an iOS launch planned for the first quarter. Over time, the system will act on the user's behalf. This is a financial superintelligence working at all times, whether or not the app is open. That progression from guidance to autonomous action is the engineering arc the current hiring push serves. The lab label is real, but the research agenda is entirely subordinate to a shipping product with regulatory obligations and a fee structure that leaves no margin for experimental detours.

The Screening Criteria That Matter Most

Autonomous Technologies Group's job postings read less like an AI lab's wish list and more like a systems engineering checklist. Across the nine open roles, the non-negotiables cluster around production ownership, data rigor, and the ability to ship without hand-holding. These signals indicate that the company screens for engineers who can turn ambiguous financial-domain problems into reliable, auditable software.

The Member of Technical Staff (Data Pipelines) role, which carries an 8+ year experience requirement, makes this explicit. The posting demands "exceptional software and data engineering skills: strong Python and SQL, plus experience with distributed processing and modern data infrastructure." But the next bullet is the tell: "Production ownership: demonstrated ability to design, deploy, operate, and improve business-critical data systems at scale." ATG isn't hiring for model experimentation; it's hiring for the plumbing that makes models trustworthy in a regulated, money-moving context. The role also requires "temporal-data expertise: revisions, event time versus knowledge time, reproducibility, and point-in-time querying." This is a direct nod to the look-ahead and survivorship biases that can invalidate quantitative research. Candidates who can't speak to lineage, versioning, and backfill strategy won't clear the first screen.

Financial-data fluency is listed as "strongly preferred" rather than required, but the specificity matters: "market data, security identifiers, corporate actions, or quantitative research." This isn't a generic data engineering role with a finance veneer. The posting notes the team is "building a data lake that ingests a broad and growing set of financial and market datasets from many vendors and sources" and the hire will "take ownership of it end to end: hardening what exist today, making the core architectural calls on storage, orchestration, and data modeling, and building out the rest." The expectation is architectural judgment, not ticket-taking.

The iOS role, by contrast, requires 3+ years of hands-on Swift experience and at least one App Store shipment "evolved based on real-user feedback." The phrasing "comfortable shipping fast while maintaining quality (modular design, unit tests, CI, profiling)" and "resourceful: you choose the right tools, work lean, and thrive without heavy scaffolding" echoes the same production-first mindset. Notably, the posting adds "expertise in leveraging the latest AI tools (Cursor, Claude Code, Codex, etc) to increase productivity & code output while maintaining high code quality, maintainability, and structure." This is a practical filter for engineers who've integrated coding agents into daily workflow rather than just prompting them for demos.

Seniority bands across the nine roles reinforce the pattern. The Research Scientist role asks for 6+ years; the Client Director, 5+; the Data Pipelines role, 8+. Even the Research Engineering and Full Stack roles start at 3+ years, with only the Research Engineering position dipping to 1+ year. This suggests a narrow aperture for early-career hires, and only on the engineering side of research. The company's self-description ("high agency, talent dense, zero bureaucracy") functions as a cultural screen: candidates who need specification documents and management layers self-select out.

What's absent is telling. No role lists publications, conference talks, or novel architecture papers as requirements. The Research Scientist posting (6+ years) doesn't appear in the provided research with its full criteria, but the surrounding roles suggest ATG evaluates research talent by the same production yardstick: can the work survive contact with live financial data, regulatory constraints, and the latency demands of a consumer-facing advisor product? The "AI-native workflow" bullet on the data pipelines role — "uses coding agents daily and builds data systems that are safe, well-described, and machine-consumable, for agents as much as humans" — captures the actual AI bar: not model novelty, but building infrastructure that agents can reliably operate on.

In practice, this means a candidate with three shipped production systems, strong opinions on orchestration frameworks, and a track record of cleaning vendor data will outrank a PhD with a transformer paper and no deployment scars. The screening criteria are engineered for the company's stated transition from early-stage development to product deployment — and they filter accordingly.

Why Engineering Trumps AI Theory in Hiring

The job posting for ATG's Member of Technical Staff — Research Engineering role reads like a systems engineering spec, not a research agenda. Expert-level Python. Significant experience in PyTorch, JAX, and TensorFlow. The ability to translate research into robust, scalable systems. Expertise leveraging AI coding tools (Cursor, Claude Code, Codex) to increase output while maintaining code quality, maintainability, and structure. Prototype and scale experimental models (LLMs, RL agents, agentic systems) on large, real-world data. Build tools and pipelines for training, evaluation, and analysis. Implement state-of-the-art research from papers and iterate in collaboration with scientists. Own the full ML lifecycle: data engineering, experimentation, training, and deployment.

None of those bullets ask for a publication record. None ask for novel model architecture. The phrase "state-of-the-art research from papers" positions the candidate as an implementer of others' breakthroughs, not a generator of their own. The explicit callout to AI coding assistants signals a team that measures productivity in shipped, maintainable code — not in theoretical novelty.

This aligns with what the broader ML engineering discipline has documented. Databricks' guide to production ML systems notes that the majority of ML projects fail not from technical limitations but from poor planning, inadequate scoping, fragile code, or inability to demonstrate business value. The same guide draws a sharp line: data scientists focus on model development and statistical analysis; ML engineers concentrate on building scalable, maintainable systems that deliver real business value. ATG's requirements map directly to the second category.

The "full ML lifecycle" bullet is the tell. Data engineering. Experimentation. Training. Deployment. That is a software delivery scope. A pure research lab hiring for novelty would emphasize experiment design, benchmark chasing, or theoretical contribution. ATG emphasizes pipelines, evaluation tooling, and the operational path from prototype to production. The posting even flags "simplicity over complexity" as an architectural principle — engineers default to the least complex solution that solves the problem. That is a production heuristic, not a research one.

Team context reinforces the point. ATG lists eight people total as of August 2026, founded in 2025 as a Y Combinator F25 batch company. At that size, every hire must contribute to the shipping surface area. There is no room for a theorist who hands off prototypes to an engineering team — because the engineering team is the same two or three people. The posting's cultural markers ("high agency, talent dense, zero bureaucracy") describe an execution environment, not a research sanctuary.

The compensation band for comparable software engineer roles at ATG ($120k–$205k per ZipRecruiter; 279 listings active) also tracks engineering-market rates, not research-premium rates. The Indeed average for ML engineers sits at $131,007; entry-level starts $100k–$130k; experienced engineers at leading tech companies earn $150k–$250k+. ATG's range sits squarely in the applied-engineering band.

Candidates who lead with publication counts or novel architecture proposals are answering a different interview than the one ATG is conducting. The role demands someone who can take a paper, extract the usable idea, harden it into a pipeline, instrument it for evaluation, and deploy it on financial-market data — then do it again next week. That is systems work. The research is the input. The engineering is the job.

The Background of ATG's Current Team

Autonomous Technologies Group lists eight people on its Y Combinator profile as of early 2026, with LinkedIn showing a 2–10 employee range. Four names appear publicly: CEO Dillon Erb, plus Francesco Bertelli, Scott Penrose, and Mark Kizelshteyn. Erb and co-founder Daniel Kobran built Paperspace, the GPU cloud platform DigitalOcean acquired for $111 million in 2023. That exit shapes the team's DNA more than any AI lab branding. Paperspace was infrastructure (bare-metal GPUs, container orchestration, multi-tenant scheduling) not model research. The founders spent seven years solving systems problems at scale. They brought that orientation to ATG.

The hiring plan reinforces it. Of the nine open roles, seven carry "Member of Technical Staff" titles across backend, full stack, iOS, and research engineering. Only one Research Scientist slot asks for six-plus years of experience and a $200K–$300K band. The research engineering role starts at one year. The message: ship production systems first, publish papers second. Bertelli, Penrose, and Kizelshteyn joined a company of eight — meaning each hire represented 12.5 percent of headcount. In a team that small, there is no room for pure researchers who cannot deploy. The YC job description for research engineering says it directly: "We're building a small, elite team. High agency, talent dense, zero bureaucracy."

Investor composition tells the same story. Garry Tan led the $15 million seed. Tan partnered with Erb and Kobran at Y Combinator in 2015. BoxGroup (backers of Plaid, Ramp, Stripe) participated. These are operators who fund operators. The "founder of one of the most successful quant funds" appears in the cap table per the YC listing, unnamed but signaled. That connection matters: quantitative finance lives or dies on execution latency, data pipeline correctness, and regulatory compliance. Not novel architectures.

The product itself, Autonomous, is positioned as an "AI-native wealth strategist" at zero advisory fees. The founders frame it as the Robinhood moment for wealth management. Robinhood's breakthrough was not AI — it was commission-free trading wrapped in a mobile experience that passed compliance. ATG's first product requires connecting to custody, handling KYC/AML, rendering tax-loss harvesting logic, and presenting it all on iOS. The iOS role at $160K–$240K sits alongside backend at $180K–$260K. Full stack at $170K–$250K. These are product engineering bands.

Contrast with the Research Scientist role: $200K–$300K, six-plus years, likely the only slot where a publication record carries weight. One seat out of nine. The rest demand "3+ years" building systems that don't fall over. The research engineering role bridges both — but its one-year floor and $180K–$260K band align it closer to ML engineering than to fundamental research.

The team's public footprint is thin by design. No arXiv preprints. No conference talks listed. LinkedIn shows 1,751 followers and two announcement posts with 1,290 and 344 reactions respectively. The company operates in stealth-adjacent mode, typical for fintech infra before launch. What's visible is a founder duo with a nine-figure infrastructure exit, a seed round from operators, and a hiring plan weighted 7:1 toward delivery over discovery. The "applied AI lab" label describes the target domain (financial markets) not the daily work. The daily work is the same systems engineering that made Paperspace worth $111 million to DigitalOcean.

How ATG's Hiring Compares to Other Applied AI Firms

The applied AI hiring market has split into two distinct tiers. At the top sit frontier labs (OpenAI, Anthropic, Google DeepMind) where headcount has ballooned and compensation for senior researchers now reaches $700K to $2M+ in total packages. OpenAI's research organization alone crossed 500 people as overall headcount doubled to roughly 4,000 over the past 18 months. Anthropic grew from a few hundred in 2023 to more than 1,000 by early 2026. These labs compete for a narrow slice of talent: PhDs with publication records in model architecture, scaling laws, and alignment. Their hiring signals — research scientist titles, publication expectations, compute-heavy interview loops — are not what ATG projects.

Below that tier, the market looks different. Applied AI roles now dominate startup and enterprise hiring across vertical domains: finance, healthcare, logistics, manufacturing. The billionhopes.ai analysis of 2025-2026 trends found that most AI jobs today are applied roles, not pure research. Startups in this band optimize for rapid experimentation and product-market fit. Enterprises optimize for reliability, compliance, security, and integration with legacy systems. Both prioritize software engineering, data engineering, MLOps, cloud deployment, model evaluation, and system reliability over novel model architecture. Domain knowledge (finance, health, manufacturing) often matters more than pure model novelty.

Source: fdepulse.com (2026 compensation survey); runfieldwork.com (2026 talent migration analysis)

ATG's nine open roles — spanning ML engineering, platform engineering, data engineering, and a forward-deployed engineer slot — map cleanly to the applied tier. The company does not list research scientist positions. It does not require publication records. Its descriptions emphasize production-grade delivery, systems thinking, and financial markets domain fluency. That aligns with the runfieldwork.com finding that enterprise companies frequently mis-hire by posting "AI Research Scientist" descriptions when they actually need engineers who can ship features. The realistic talent pools for vertical AI companies are applied ML engineers (deploy models in production), AI product engineers (build AI features into existing products), and AI integration engineers (connect external AI APIs to enterprise systems). These pools are larger and compensation expectations are more reasonable.

The mistake most enterprise companies make is writing job descriptions for 'AI Research Scientist' or 'ML Researcher' when they actually need an AI Engineer who can ship features.

Two role archetypes have emerged as the fastest-growing technical positions at AI companies in 2026: Forward Deployed Engineer (FDE) and Applied AI Engineer. The FDE role traces back to Palantir's customer-deployment model in the 2000s and spread broadly across AI labs starting in 2022. The Applied AI Engineer role emerged around 2023-2024 as companies like OpenAI, Anthropic, and Cohere needed engineers who could ship AI features in core products without the customer-engagement scope of an FDE. ATG's single FDE opening suggests a hybrid motion: most hires will build the core platform, while one or two sit closer to customers (likely hedge funds, prop shops, or asset managers) to tailor deployment and gather product signal. That mirrors the structure at vertical AI firms in healthcare (e.g., pathology imaging platforms) and logistics (e.g., warehouse automation stacks), where a small FDE team feeds the product roadmap.

Compensation at ATG will likely track the applied ML engineer band ($200K–$400K) rather than the frontier lab bands. The gap between frontier labs and everyone else has widened significantly over the past two years. Companies like Stripe and ASML (neither an AI lab but both heavy AI adopters) show what mature applied hiring looks like at scale. Stripe added 51 roles in the past seven days with a board salary band of $42K–$286K (median $235K). ASML added 53 roles with a band of $31K–$258K (median $164K). Both hire ML engineers, data scientists, and platform engineers into production systems — not research orgs. ATG's scale is smaller, but the role composition rhymes.

The boundary between research and applied AI is narrowing. Many applied teams now run controlled experiments, contribute to open source, and publish engineering research. Hybrid titles (research engineer, applied scientist, model infrastructure engineer) are emerging. But the core difference persists: external vs. internal focus. FDEs face customers; Applied AI Engineers face the product. ATG's hiring slate puts weight on the product side, with a toehold in the customer-facing motion. That is representative of a vertical AI company moving from early-stage development to product deployment — not a frontier lab, not a pure consultancy, and not an enterprise IT shop bolting on AI. It is an applied AI lab in the precise sense the market now uses.

What Candidates Misunderstand About Applying to ATG

The gap between how candidates present themselves and what ATG actually evaluates starts before a human ever sees a resume. Research from Forbes shows that over 75% of resumes are rejected by applicant tracking systems before a human reviewer gets involved, and more than 98% of Fortune 500 companies now use such systems. ATG, headquartered in New York City and registered as an investment advisor under the SEC and FINRA, operates in a regulated financial domain where compliance, auditability, and systems reliability are non-negotiable. Candidates who treat the application like a generic AI research lab submission — leading with publication counts, model architecture novelties, or theoretical alignment work — often fail the automated screen because the keywords and experience patterns the system recognizes map to production ML engineering, not academic benchmark-chasing.

A recurring signal from industry hiring discussions underscores this mismatch. One widely circulated LinkedIn observation noted that 99% of self-described "Agentic AI experts" fail interviews in the first ten minutes — not because they cannot discuss AI concepts, but because they cannot explain how those concepts translate into reliable, observable system behavior. ATG's stated mission for its Autonomous platform is threefold: deeply understand the market, deeply understand an individual's financial life, and advance representational capabilities. Each objective demands engineering rigor — data pipelines that ingest messy market feeds, privacy-preserving architectures for personal financial data, and model serving infrastructure that meets latency and compliance requirements. Candidates who frame their experience around "building agents" without demonstrating ownership of the surrounding systems (monitoring, evaluation, rollback, data lineage) signal a fundamental misunderstanding of the role.

The anthropomorphism trap compounds the problem. Academic literature on AI anthropomorphism warns that attributing human-like agency to models distorts both capability assessment and responsibility assignment. In an ATG interview, describing a system as "understanding" or "reasoning" without qualifying the mechanistic basis (retrieval, tool use, constrained generation) reads as imprecision. The company's own documentation describes Autonomous as a system that "onboards you into a system that learns your financial life completely and guides you with precision," with a future trajectory toward acting on users' behalf. That language signals a product mindset: deterministic guarantees, regulatory boundaries, and user trust built through transparency. Candidates who lean on anthropomorphic shorthand reveal they have not internalized the engineering discipline required to ship in this environment.

Another misconception involves the regulatory context. Autonomous operates as an RIA under SEC and FINRA jurisdiction. This is not a footnote — it shapes every layer of the stack, from data retention and audit trails to model explainability requirements. Engineers who have never worked in a regulated environment often underestimate the weight of compliance-driven design decisions: why a feature store must support point-in-time correctness, why model versioning cannot be best-effort, why experiment tracking must satisfy an examiner. The Forbes research on ATS bias notes that algorithms replicate historical hiring patterns, which in fintech have long favored candidates with domain-adjacent experience. A resume that highlights only open-source LLM contributions without any evidence of working under regulatory constraints will likely score lower than one that includes even a single project with audit requirements, regardless of the former's technical sophistication.

The referral advantage is real and structural. Forbes reporting confirms that employee referrals remain the most trusted source of hires because they shortcut the algorithmic filter — a human inside the organization has already vouched for the candidate's ability to contribute and fit the culture. ATG's team, composed of engineers and researchers with backgrounds spanning quantitative finance, systems engineering, and applied ML, evaluates fit through that lens. Candidates who apply cold, without engaging ATG's technical blog, open-source releases, or public talks by team members, miss the opportunity to align their narrative with the problems the team actually discusses. The research is clear: in a system increasingly dominated by algorithms, the most effective strategy is to step outside of it and connect human to human.

Finally, candidates confuse "applied AI lab" with "research lab that applies AI." ATG's nine open roles span engineering, design, and operations — not research scientist titles. The hiring push reflects a transition from early-stage development to product deployment. The evaluation criteria have shifted accordingly: demonstrable systems engineering ability over pure AI novelty. A candidate who optimizes for the latter (polishing a novel architecture for a benchmark) while neglecting the former (shipping a maintainable, monitored, compliant service) has misread the company's stage and its hiring priorities. The misunderstanding is not about technical competence. It is about which competence the role rewards.

The nine roles posted between July and August 2026 are not a research lab's wishlist — they are a product team's blueprint, written in salary bands and experience floors that favor engineers who can ship financial software under regulatory scrutiny over researchers chasing publication credit.

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