The Engine Room
At most AI startups, engineers build the model and salespeople sell it. At Hebbia, former investment bankers and transaction lawyers sit inside customer firms, such as BlackRock, KKR, and Centerview, and write the logic that makes the model useful for that specific shop. They don't sell. They encode. The distinction rewrites how a 150-person company operates across New York, San Francisco, and London.
George Sivulka founded Hebbia in 2020 while finishing his PhD at Stanford. He describes the company as sitting "between what an Anthropic would do and what a Palantir would do." That positioning, part frontier research lab, part embedded implementation partner, sets the daily rhythm. The research team publishes papers on autonomous evaluator consensus and builds financial AI benchmarks. The forward-deployed team, drawn from the very workflows the product targets, translates each firm's "institutional secret sauce" into Hebbia's primitives: deterministic scaffolds that wrap generative reasoning in verifiable, auditable steps. The LLM handles the hard reasoning; the skeleton holds the workflow together. One without the other fails in finance, where a missed clause in a credit agreement carries real liability.
Decision-making mirrors that split. Engineers and forward-deployed staff ship directly. When a customer needs a new primitive (say, a reconciliation workflow that cross-references a loan agreement against an amendment filed two years earlier), the pair who understands both the code and the covenant threshold builds it. Sivulka has said the goal is to use generative reasoning only where necessary; everything else should be rules-based and verifiable. That philosophy pushes authority to the people closest to the problem. A former Hebbia user who joined the forward-deployed team can design the primitive, test it against live data rooms, and push to production.
The pace is set by the customer side. Hebbia processes roughly 200,000 prompts a day across more than 1,000 production use cases, ingesting 1.5 billion pages and 2 trillion tokens for firms managing $30 trillion in assets, Hebbia's site reports. When a deal team at Apogem Capital cuts diligence review from 12 hours to two, the feedback loop tightens. Revenue grew 15-fold, TechCrunch reported, in the 18 months before the July 2024 Series B, reaching $13 million ARR, TechCrunch's data shows, at profitability, a 54x multiple, according to TechCrunch, that gave the company runway to hire aggressively. Headcount doubled from the 15-person team reported in 2022 to the 51–200 range LinkedIn shows today, with open roles on Zero G Talent spanning AI Strategist, Solutions Engineer, and Partnerships Manager at bands between $170,000 and $300,000.
The flat structure has sharp edges. The forward-deployed bankers juggle customer embedding, primitive design, and internal tooling. The company's own marketing emphasizes "enterprise controls since day one" (SOC 2 Type II, ISO 42001, bring-your-own encryption keys, sovereign cloud deployments for UAE clients), and the head of security came from Bridgewater, a firm legendary for vendor scrutiny. Meeting that bar while shipping new primitives requires a tolerance for ambiguity that org charts usually absorb.
Sivulka frames it as earning trust incrementally: "You really have to earn that trust before you can start removing those checkpoints." The autonomy gap — between an agent that reads a document and one that sends an email or updates a financial model feeding a real decision — is where Hebbia lives. Closing it means the people who understand the workflow must also understand the code. That collision of domains is the operating rhythm. It moves fast because it has to. It stays flat because hierarchy would slow the translation. Whether that velocity is sustainable for the people doing the translating is the question the next sections examine.
The Load-Bearing Values
Hebbia publishes four values on its about page, each phrased as a command. Run Hard: operate with intensity and purpose, own outcomes like a founder, move fast and raise the bar. Readiness: learn fast, adapt faster, thrive in ambiguity, seek feedback, grow ahead of the curve. Compassion: respect others' perspectives, invest in team success, operate with emotional awareness and maturity under pressure. Cogency: drive clarity, structure thoughts, distill complexity, communicate clearly in writing, meetings, and decisions. The list is short, deliberate, and absent the usual corporate filler: no "integrity," no "customer obsession," no "excellence."
The values reflect a deeper operating philosophy Sivulka has articulated repeatedly. "In finance, rigor is everything," the about page states. "A single position can touch thousands of documents, dozens of data sources, and a research process that runs through models, memos, and presentations. Being slightly wrong costs the same as being completely wrong, so a system that's right most of the time isn't one institutional finance can use." That standard — zero tolerance for approximate answers — shapes how Hebbia builds. The company doesn't just ship models; it encodes the judgment of bankers, investors, lawyers, and consultants into the product itself. "We've spent more than six years building for that, alongside people who spent their careers at the world's leading institutions," the site reads. The result is described as "an intelligence platform built for how finance operates."
Sivulka extends the logic beyond product into organizational design. In a LinkedIn post titled "AI is your new worst employee," he outlines seven parallels between agent and human workforces. The provocation: "AI just gave your worst employee infinite headcount and infinite budget. For the first time in history, humans are cheaper than software. It's changing how we work." The framing reveals a principle that runs through Hebbia's hiring and management: human judgment is the scarce resource, and the organization should be structured to maximize its leverage. That means flat teams, minimal process, and a bias toward shipping over coordinating.
The values also explain the product trajectory. Matrix began as a search and summarization tool, then evolved into an AI analyst that can ingest unlimited files and return answers in a tabular, spreadsheet-like format with linked citations. The 2025 FlashDocs acquisition, a startup automating slide deck generation from structured prompts, extended the platform from retrieval into full artifact generation: investment memos, diligence reports, board presentations. Each step encodes more of the financial workflow into the system, reducing the need for human handoffs. The recent launch of "Max, the first AI teammate that works the way your firm works" continues the pattern: the product absorbs the firm's processes so the firm doesn't have to adapt to the tool.
Compassion and Cogency, the two values most often treated as soft, function here as operational constraints. Compassion — that directive, especially under pressure — acknowledges the intensity that Run Hard demands. Cogency — that formulation — is the mechanism that keeps a flat, fast-moving team aligned without heavy process. Writing replaces meetings. Clarity replaces consensus. The values are not culture posters; they are the load-bearing walls of a structure designed to move at a pace that would fracture a conventional org.
The Hiring Filter
Hebbia's hiring bar reads like a filter for a specific intersection: deep financial domain fluency paired with genuine technical depth, wrapped in the kind of autonomy that doesn't need managing. The evidence sits in who they've actually hired. The AI Strategy team lists alumni from Permira, Barclays, Sixth Street, Credit Suisse, people who've run live deals and priced assets. The technical staff includes a former Microsoft engineer, a Y Combinator founder, and designers and marketers from Stripe and Amazon. The pattern isn't accidental.
The four stated values function as behavioral selectors. "Run Hard" selects for velocity tolerance: headcount quintupled in a year while revenue grew 15x over 18 months. "Readiness" maps to the platform's model-agnostic architecture, as the team integrates new LLMs (Claude Opus 5, Gemini 3.6 Flash, GPT-5.6 variants) as they drop, meaning engineers and strategists must evaluate and ship against moving targets. "Cogency" reflects the product's core constraint: in institutional finance, zero tolerance for approximate answers is the standard. That filters for people who treat hallucination rates as a blocker, not a metric to optimize later.
Current openings on Zero G Talent's board make the profile concrete. AI Strategist roles (Principal, Strategic Accounts) carry bands of $200k–$300k and explicitly seek candidates who can "encode your firm's processes and judgment into every system they build," language that mirrors the company's self-description. Solutions Engineer roles ($170k–$260k) sit at the product-customer boundary, requiring both technical implementation skill and the credibility to sit across from Centerview Partners or Fenwick. Strategic Partnerships Manager roles (Software, Content) at $170k–$300k signal a push to expand the connector library that already spans SEC filings, FactSet, S&P Capital IQ, PitchBook, Preqin, and a dozen more data sources.
| Role | Salary Band |
|---|---|
| AI Strategist (Principal, Strategic Accounts) | $200k–$300k |
| Solutions Engineer | $170k–$260k |
| Strategic Partnerships Manager (Software, Content) | $170k–$300k |
Across 22 salaried roles, the board's median band sits at $200k with a $90k–$274k range, compensation that assumes immediate leverage, not ramp time.
The flat structure means the hiring bar must select for people who define their own scope, negotiate priorities directly with counterparts, and ship without heavy process. The work-life balance score of 4.2 on Glassdoor, the lowest of the major categories, hints at the friction beneath the high marks. Candidates who need onboarding programs, clear promotion ladders, or protected focus time self-select out before the offer stage. The bar isn't just high; it's shaped to exclude the very people who would struggle in the environment the founders built.
Inside the Room
Public sentiment about working at Hebbia comes almost entirely from the company's own curated testimonials, six employees quoted by name on hebbia.com/about, all emphasizing intellectual intensity, peer quality, and the pull of a hard technical problem. Jenna Neher, who leads AI strategy after stints at Permira and Barclays, frames the experience as "solving hard problems with your friends." Beshoi Genidy, a technical staff member formerly at Microsoft, calls out "exceptional talent working with intensity and humility to build something disruptive." Charlie Pickell, also in AI strategy after Sixth Street and Credit Suisse, joined because "finance was changing forever, and I needed to help build it." Adithya Ramanathan, a Y Combinator alum and former founder, wanted "the smartest people" and a place to build. Kevin Shankar, marketing lead from Stripe and Adobe, cites "connecting our frontier technology with our customers" as the daily thrill. Arjun Mahesh, design lead from Stripe and Amazon, points to "a world-class team, develop patented interfaces, and drive innovation in finance."
These accounts share a through-line: people who left structured, high-prestige roles in finance or big tech for a smaller, faster environment where they can own outcomes end to end. The former-company pedigrees, including Permira, Barclays, Microsoft, Sixth Street, Credit Suisse, Y Combinator, Stripe, Adobe, and Amazon, signal a hiring bar that selects for proven performers in demanding settings. The language mirrors Hebbia's stated values and the founder's emphasis on speed and intellectual ownership covered in earlier sections.
What's absent from the research provided is any critical public review data: no Glassdoor trends, no Blind threads, no Reddit discussions, no departing-employee posts. The company's LinkedIn page (51–200 employees, primary hubs in New York, San Francisco, London) shows growth announcements like Avi Upreti joining the AI Strategy team after a decade in investment banking, but no open discourse about workload, management, or culture friction. That silence is itself a data point: at this stage, with the Series B closed in July 2024, Hebbia is small enough and new enough that public review surfaces haven't accumulated a critical mass. The testimonials on the company site are real, attributed, and consistent — but they are also selected. Candidates should treat them as signal about who thrives here, not a representative sample of every experience.
The Autonomy Tax
Glassdoor numbers tell a split story. As of mid-2024, 88 percent of Hebbia employees would recommend the company to a friend, and the overall rating sits at 4.4 out of 5 across 16 reviews. Culture and values scored 4.8. Career opportunities scored 4.8. The work-life balance rating matches the earlier 4.2.
That 4.2 is not a crisis. It is a signal. BuiltIn's workplace profile describes a flexible schedule and unlimited PTO, but adds that "feedback suggests policies allow employees discretion in planning time away." Discretion is the operative word. No one stops you from logging off. No one reminds you to. The OKR model defines what success looks like; the path to get there is yours to cut.
People who stay and advance share a cluster of traits. They treat ambiguity as a design space, not a defect. They ship without a spec. They argue directly with the founder and expect their ideas to be judged on merit, not title. The flat structure rewards this.
The hiring bar selects for this profile. Hebbia's own job postings, such as AI Strategist roles at $200k–$300k and Solutions Engineers at $170k–$260k, signal that the company pays for ownership, not execution. The median salary band on the board is $200k. The roles are titled "Principal" and "Strategist," not "Analyst" or "Associate." The language assumes you will define the work, not wait for a ticket. That 4.2 work-life balance score recurs.
The 4.2 work-life balance score is not an accident. It is the tax exacted by a culture that treats every employee as a founder of their own domain. Sivulka's "7 Parallels of Agent and Human Workforces" frames humans as the expensive, high-leverage layer, the ones who decide what the agents should do. That framing empowers. It also means the human layer never gets a break from deciding. The work-life balance metric underscores the same friction.
The quarterly engagement surveys and employee awards exist, but they recognize output, not effort. Promotion is from within, but the ladder is built by the people climbing it.
If you want a job where the boundaries are drawn for you, Hebbia will frustrate you. If you're looking for a role where the boundaries are yours to draw, and you have the evidence that you can draw them well, the same frustration becomes the reason you stay.
Working in frontier tech? Zero G Talent tracks the openings: see every open Hebbia role, browse frontier tech jobs, the companies hiring, and the people building the field.