How Work Gets Done at Credal.ai
Two Palantir alumni started a company to control how AI models touch enterprise data. Three years later the team is small: Y Combinator lists 14 people; BuiltIn counts 10 product and tech employees; Ravin Thambapillai said "still a five-person team" in a February 2025 interview. The founders still approve architectural decisions before they ship, a founder-led operating model that prioritizes execution speed and technical depth over pedigree, and shapes every hiring decision, product choice, and daily ritual.
Jack Fischer and Ravin Thambapillai founded Credal.ai in 2022 after building data-access tooling at Palantir. Their pitch — an "Okta for AI" governing which models read which internal systems — landed a $750,000 contract with the U.S. Department of Health and Human Services, USAspending's data shows and enterprise deals with MongoDB, Comcast NBC Universal, Checkr, and Wise. The company has raised more than $10 million, Credal.ai's careers page reports. Eight salaried roles on Zero G Talent job postings show a median band of $190,000. Headcount stays small by design; both founders said they prefer a dense, senior team over rapid hiring.
Authority rests with Fischer and Thambapillai. Weekly all-hands serve as the primary decision forum. Product direction, hiring priorities, and technical trade-offs are debated there, not in a layered management chain. Employees describe direct access to both founders as routine. No intermediate VP layer clears a spec or escalates a blocker. Open roles — Software Engineer, Forward Deployed Software Engineer, and Customer Deployment Engineer — all report into the founding pair.
The pace reflects the market. Credal operates in the model-agnostic governance layer where every major cloud provider and a dozen well-funded startups race to define the standard. Internal feedback collected by Built In characterizes the environment as "really, really hard" work with late nights and rapid iteration. That intensity is concrete: the product ships over 1,000 data connectors, scopes agents to minimal context windows, and publishes nearly a nine-tenths reduction in context per query alongside roughly three-quarters better accuracy for customers like Checkr. Those numbers come from the company's own benchmarks; independent verification is limited.
Rituals reinforce the founder-led model. Lunch & Learns run weekly, often led by Fischer or Thambapillai on topics from MCP server design to go-to-market motion. Strategy discussions stay open: roadmap docs are shared company-wide, and engineers challenge assumptions in real time. The office is in-person only, meals provided, with a year-end shutdown and unlimited PTO framed as recovery mechanisms, not perks.
The structure creates a specific tension. High autonomy and end-to-end ownership are real. A Founding Frontend/UX Software Engineer or Founding GTM Deployment Strategist owns a surface area that at a larger company would span three teams. But the same flatness means support structures vary by manager and role. Limited public review volume makes it difficult to gauge how consistently the culture holds under pressure. The operating principles Fischer and Thambapillai codified to manage that tension are the subject of the next section.
Values and Operating Principles
Credal's operating principles read less like a values poster and more like a survival manual written by two Palantir alumni who watched enterprises choke on AI adoption. The through-line is trust, not as a compliance checkbox but as the prerequisite for any AI system touching regulated data. "Security & compliance: Move fast without compromising control," the company's site states. "AI adoption always outpaces trust. Credal sits as a gateway between your agents and your systems, so permissioning, auditing, and legal review happen once and apply to every team, instead of being rebuilt slightly differently by each one." That framing — centralized governance as an accelerant, not a brake — shapes every product decision.
The founders' Palantir tenure surfaces repeatedly. Fischer deployed computer-vision models at the Department of Defense; Thambapillai kept hospitals from overflowing during COVID at the Department of Health. Both environments demanded deterministic behavior, audit trails, and permission fidelity. When they founded Credal in 2022, they carried that constraint set into a market rushing to bolt chat interfaces onto sensitive data. The result: a platform that mirrors source-system permissions automatically — a user querying Salesforce sees only the records their Salesforce role allows — and logs every LLM call with full data-lineage. A 2023 TechCrunch profile noted Credal had processed over 250,000 LLM calls and ingested roughly 100,000 documents across 11 customers, several on six-figure contracts. By February 2025, Thambapillai said the team was handling over a million LLM queries monthly and adding a household-name enterprise every two to three months.
Model agnosticism is a second principle, baked into architecture rather than marketing. "No model lock-in. Switch between Claude, ChatGPT, Gemini, or any model without reconfiguring a single permission or prompt," the site promises. Connectors, permission rules, and skills work across Claude Desktop, Cursor, Slack, and custom apps without extra setup. That neutrality let Credal win regulated European accounts like Wise after registering under the U.S.-EU Data Privacy Framework, a credential Fischer explicitly tied to those deals in 2023. The same neutrality appears in the MCP (Model Context Protocol) server layer: domain experts bundle only the tools, data, and instructions a given workflow needs, avoiding the "over-provisioned agents" Thambapillai calls "expensive and inaccurate for the same reason: too much context."
Customer-driven iteration replaces theoretical threat modeling. "What we're really focusing on is asking our customers, hey, what are the actual things you're trying to do? What are the security concerns you actually have in real life? Which are theoretical and which are you completely worried about?" Thambapillai said on the Frontlines podcast. The pivot from "AI chief of staff" to pure security gateway came after a Latchel pilot where executives signaled $10–20/month willingness to pay, "off by two orders of magnitude," Thambapillai recalled. They kept the security thesis, discarded the application layer, and let inbound demand pull the roadmap. "For us at least inbound was definitely the primary driver of our growth and hopefully it stays that way."
The hiring bar operationalizes the same selectivity. Thambapillai described a reference-check question that stops most processes cold: "Hey, is this person actually the best person you've ever worked with? Maybe one of the best two or three? If we don't get references saying yes, this is one of the best people I've ever worked with, we just can't make the hire because that's what amazing people want to work with." As of early 2025 the team stood at roughly 14 people (Y Combinator lists 14; he reiterated that the team remained at that size in the same February 2025 interview, suggesting the count may reflect contractors or very recent hires). Open roles signal a split between core product and deep customer embedding:
| Role | Salary Band |
|---|---|
| Software Engineer | $190K–250K |
| Forward Deployed Software Engineer | $140K–220K |
| Customer Deployment Engineer | $120K–160K |
The latter reflects Fischer's line about "embedding deeply with a small number of amazing enterprises and actually solving their problems end-to-end."
Mission focus narrows the aperture further. Thambapillai named healthcare, education, financial services, and housing as "the really critical industries... the industries that are going to care the most about the security and compliance component of how they deploy AI." The customer roster — Wise, Checkr, Lattice, IFRS — maps to that thesis. So does the product trajectory: unstructured data first (documents, PDFs, images, video), then structured sources, "your data warehouse, your production databases, the sensors on your drones," as Thambapillai put it. The long-term vision is an "access controlled data lake for your AI employees" governing thousands of specialized agents across and between companies. "Google will need to talk to the AI at Google's customers and vice versa," he said. "And that exchange will need to be governed because I want the AI to help brief you, but I also don't necessarily want it to accidentally reveal that I'm trying to get Ilia fired or that I'm in a dispute with Satya Nadella or whatever."
In practice, these principles collapse into a single loop: ship the minimal governed primitive that lets a platform engineering lead — Credal's stated ICP — deploy an agent into production this week, then iterate from real logs, not hypotheticals. The values are the guardrails that keep the loop from spinning into chaos.
What the Hiring Bar Selects For
Credal.ai's interview process runs three stages in roughly seven days (faster than the ten-day median for comparable B2B productivity companies) and each stage filters for a different signal. The intro call (30 minutes, usually with a founder or hiring manager) tests culture fit and role expectations. The technical deep dive (60 minutes, led by a technical founder or lead) examines past projects and problem-solving approach. The final round (45 minutes, with the founding team) evaluates team fit and handles offer discussion. No take-home assignment exists; the company explicitly avoids puzzle-style screens in favor of assessing whether a candidate can operate without heavy process at a team of fewer than 50 people.
Job postings and founder communications consistently name five traits that predict success. First, strong engineering fundamentals: the ability to reason about architecture, abstractions, performance, reliability, and complexity rather than fluency in a specific framework. Second, experience scaling engineering teams through systems, tools, and processes that create clear swim lanes and reduce complexity for others, not just by writing code. Third, high ownership: taking projects, outcomes, and customer unlocks from an ambiguous starting point to a clean finish. Fourth, bias toward action in fast-paced environments rather than waiting for direction. Fifth, craft and execution that elevates teammates, described in the Y Combinator listing as being "known as a creative, trusted, and resourceful teammate who elevates those around you."
These traits map directly to the company's stated mission: accelerating complex operations at impactful institutions while weaving security and governance into every product layer. The team of 13 (nine engineers per the latest Y Combinator posting) ships against seven-figure ARR with greater than 100 percent net revenue retention. The hiring bar selects for people who can sustain that velocity without guardrails.
The interview verifies genuine expertise. Candidates whose résumés claim five years of experience but cannot answer basic field questions create a "loophole" exposed quickly. Consistency across CV, application, and live discussion matters more than pedigree; the team includes alumni of Palantir, Cruise, Meta, Uber, and Google, but the screening language emphasizes "relevant experience and ability to operate autonomously" over brand names.
Practical signals that clear the filters: a tailored CV mirroring the job description's language (generic résumés are noticed and rejected fast at this size), opening with what Credal is dealing with right now rather than what the candidate wants, and a day-five follow-up that reportedly doubles response rates. References are checked on every hire to confirm future coworkers are "world class and great to be around." The final round carries a reputation as the most challenging stage not because of algorithmic puzzles, but because it tests whether the candidate's operating style aligns with a founder-led, high-autonomy environment where the default is shipping.
Employees Describe High Autonomy, Grinding Pace
Independent employee-review volume for Credal is extremely limited. As of October 2026, BuiltIn's culture analysis notes that "independent employee-review volume is extremely limited, making it unclear how consistently the stated culture is experienced" and that "sparse third-party sentiment means enthusiasm may not be representative across roles." No Glassdoor or Blind threads with meaningful sample sizes surface in public search. What exists are aggregated "Feedback suggests" patterns from BuiltIn's workplace-perception and culture-values pages (summaries derived from whatever internal signals the platform captures) and one documented episode of employee pushback reported by Wired.
On the positive side, BuiltIn's analysis describes a team that experiences such autonomy and ownership. "Feedback suggests small, senior teams are given such autonomy and ownership, with direct access to leadership and decision-making forums like weekly all-hands." The same source notes "regular Lunch & Learns, apprenticeship-style access to leaders, and open strategy discussions are part of the operating rhythm," indicating intentional investment in skill development. Benefits land in the feedback too: "comprehensive health coverage, unlimited PTO, and a year-end shutdown are positioned to support rest and wellbeing. Daily in-office meals and team activities also aim to reduce friction and strengthen cohesion." The workplace-perception page adds that roles emphasize "founding scope, end-to-end ownership, and direct customer exposure," framing the environment as "low-bureaucracy with individuals shaping priorities and outcomes."
Countervailing signals are specific. BuiltIn flags workload intensity: "Feedback suggests a fast-paced, early-stage environment with descriptions of working 'really, really hard,' including late nights and rapid iteration. Such intensity can create pressure and fatigue despite supportive perks." Change fatigue appears in the workplace-perception digest: "Fast iteration, forward-deployed roles, and customer-driven priorities suggest frequent pivots and shifting focus. Notes about rapid product cycles and hack weeks indicate a cadence where plans can change quickly." Job insecurity is called out explicitly: "The company is characterized as very small and early-stage, with startup risk and limited structural redundancy. Despite stated runway and notable customers, the inherent volatility of a young venture can translate into uncertainty." The culture-values page warns that "the in-person, small-team setup can vary by manager and role, potentially leading to uneven support structures. Limited formal processes at this stage may leave some employees without consistent cross-team backing."
The only named, on-the-record employee action in public reporting comes from Wired's January 2025 investigation into HHS grant-review contracts. The article states that "employees have questioned on Slack whether the company can put pressure on ICE, and pushed for the company to provide more transparency on its work with the agency." No employees are named; the reporting describes internal Slack activity without attributing quotes to individuals. That episode sits apart from the BuiltIn summaries, a concrete instance of staff organizing around an ethical concern, not a sentiment survey.
Taken together, the evidence forms a split picture. The company's own channels and BuiltIn's aggregated feedback describe a high-trust, high-autonomy environment with strong learning rituals and wellbeing perks. The same sources acknowledge the pace is grinding, priorities shift fast, structural safety nets are thin, and manager-dependent support can leave gaps. The Wired report adds a dimension neither summary captures: employees willing to challenge leadership on client work. With no named former employees on record and no public review corpus to mine, any characterization of "what employees say" remains provisional, grounded in platform-level pattern detection and one leaked Slack thread, not a chorus of identifiable voices.
Who Thrives and Who Burns Out
The personality that lasts at Credal.ai looks like the founders who built it. Jack Fischer and Ravin Thambapillai both came from Palantir, where deployment-forward engineering, high autonomy, and comfort with ambiguity are baseline expectations. That DNA shows up in the roles Credal posts on Zero G Talent (Forward Deployed Software Engineer, Customer Deployment Engineer, Founding GTM Deployment Strategist) titles signaling the work sits at the customer boundary, not behind an API wall. People who need a spec before they write code, or who expect a product manager to translate business problems into tickets, will struggle. The team is 10 product and tech employees total; no one exists to hand work to.
What the research describes as that dynamic rewards engineers who treat a vague customer request as a design problem they own from discovery through production. Weekly all-hands and direct access to Fischer and Thambapillai mean decisions happen in real time, but only for people willing to surface problems upward without being asked. Such mentorship and regular Lunch & Learns help, but they are opt-in. If you wait for a mentor to be assigned, you've already lost months.
The burnout profile is equally clear. Built In's aggregated feedback describes the same grinding pace noted earlier. That intensity is not a bug; it is the price of shipping an enterprise AI governance platform to customers like that agency, MongoDB, Comcast NBC Universal, and Wise with a ten-person technical team. Those benefits exist, but in a deployment-heavy role where a customer's production issue blocks their AI rollout, taking time off requires coordination the lightweight process does not guarantee. The pet-friendly office, daily meals, and free snacks reduce friction. They do not reduce scope.
Manager variance compounds the risk. The research notes the same structural concern flagged in the culture analysis. With no formal HR layer and limited cross-team process, a supportive manager is a force multiplier; an absent one leaves you isolated. The promotion-from-within track and visible growth plans help retain people who can self-direct, but they assume you survive the first 12–18 months of deployment cycles.
Equity and pay transparency (salary bands of $111k–$250k, median $190k) attract people betting on the company's trajectory, not just a paycheck. That bet pays off only if you can sustain the pace long enough for the equity to mean something. Sparse third-party sentiment (Built In flags "extremely limited" independent review volume) means you cannot rely on Glassdoor to tell you whether the culture matches the marketing. You decide from the signals: deployment-heavy roles, founder-led, Palantir alumni, in-person Brooklyn, 10 technical people moving fast.
Thrives: Engineers who have shipped to production in regulated or enterprise environments, who prefer undefined problems to defined tickets, who communicate upward proactively, and who treat customer friction as their problem to solve.
Burns out: People who need structure to function, who expect work-life boundaries enforced by policy, who want mentorship handed to them, or who view deployment as someone else's job. The distinction is not talent but tolerance for operational ambiguity at high velocity.
The Brooklyn office still fits the whole team. But the next household-name enterprise contract will require another connector, another permission model, another late-night deploy, and the founders will still approve the architecture before it ships.
Working in AI? Zero G Talent tracks the openings: see every open Credal.ai role, browse AI jobs, the companies hiring, and the people building the field.