Kastle Sells Autonomous Mortgage Agents. Its Screen Requires Human Production Mileage.

The 17-Role Scope: Where Kastle Is Scaling

Kastle's $24 million Series A, Y Combinator reported, closed in September 2026 and led by Insight Partners with continued participation from Y Combinator and Commerce Ventures plus new investors including Fifth Wall and financial-services executives, carries a specific mandate: scale the AI employee for consumer lending, starting with mortgage. The resulting headcount plan, 17 live openings on the company's own careers page as of this week, avoids the generic growth spreadsheet. It clusters around the operational seams where an autonomous agent meets a regulated mortgage workflow, and it imposes a narrow technical screen that rewards production agent experience over general credentials.

Zero G Talent, which ingests Kastle's feed directly, shows 12 salaried roles with a median band of $235,000. Engineering carries the heaviest load: Kastle's posting lists Staff AI Engineer ($250K–350K), roles.cc found Staff Engineer ($250K–320K), Kastle's posting lists Forward Deployed Engineer ($170K–250K), and a Software Engineer Intern. Two staff-level slots signal that the core agent architecture, the piece that processes payments, originates loans, and handles compliance conversations, is still hardening, not just maintaining.

Product sits adjacent but distinct. The per Kastle's posting Agent Product Manager ($150K–220K) and Kastle's posting lists Senior Product Manager ($200K–250K) are the only product roles listed. That ratio, two product leads for four senior engineers, shows the roadmap driven from the model layer outward, not from a feature backlog. The agent is the product; the PMs translate lender requirements into agent capabilities.

Go-to-market carries three slots: Enterprise Account Executive, Head of Sales, and Founding Growth Lead ($170K–200K). Kastle sells to the largest mortgage lenders in America — a buyer universe measured in dozens, not thousands. The Head of Sales role marks a shift from founder-led deals to a repeatable motion; the Growth Lead sits earlier in the funnel, shaping how lenders discover and evaluate an AI workforce.

Operations and deployment absorb another three: Kastle's posting lists AI Deployments Lead ($200K–250K), Mortgage Originations Process Specialist, and AI Trainer (Part-time) ($30/hr, remote). The Deployments Lead owns customer-side integration: the "ship it to top 10 lenders next week" velocity the job posts advertise. The Process Specialist brings domain fluency: someone who knows TRID, GSE guidelines, and servicing transfers well enough to teach the agent where the edges are. The part-time trainer role is the human-in-the-loop layer, grading agent outputs on live lender traffic.

Talent and founder's office round out the list. Kastle's posting lists Head of Talent ($200K–250K) and a Technical Recruiter (Contract) were both posted in the past seven days, the newest roles on the board, showing the recruiting engine itself is staffing up. Founder's Office ($150K–210K) and two Founder Intern slots ($100K/yr and $8K–8.75K/mo variants) reflect a team of two founders, Rishi Choudhary, Nitish Poddar, still close enough to the metal to need leverage on special projects.

The spread is lopsided by design. No marketing, no HR, no finance, no generalist ops. Every open role either builds the agent, deploys it inside a lender's contact center and compliance stack, or sells the next deployment. That concentration signals Kastle's bottleneck: not brand awareness, not back-office scale, but the technical and domain depth to make an autonomous agent reliable enough for a top-10 mortgage originator to trust with a borrower's largest financial decision.

Comparable Figures Summary

Past the Screen: What Separates Kastle Hires From Applicants

The board lists a Mortgage Originations Process Specialist alongside a Staff AI Engineer and an Agent Product Manager: three titles that define the intersection Kastle hires for: production mortgage operations, deployed agent systems, and the product judgment to stitch them together. Generalist AI engineers who have only built demos bounce. Mortgage veterans who have never shipped software bounce. The screen selects for the overlap.

The technical bar shows in the engineering titles. Staff-level bands track ICs who have taken LLM-backed systems from prototype to regulated production. The Forward Deployed Engineer and AI Deployments Lead roles confirm the pattern: Kastle needs engineers who embed inside a bank's contact center, wire agents into existing loan origination systems, and iterate under compliance review cycles. That is not a research skill set. It is a delivery skill set earned by shipping agents that real borrowers interact with.

Product roles mirror the same constraint. Those roles sit at $150K–$250K, but the title modifier "Agent" is the tell. Kastle is not hiring for generic SaaS product sense. It needs PMs who have defined evaluation harnesses for autonomous agents, negotiated guardrails with compliance officers, and prioritized roadmap bets when the failure mode is a regulatory finding rather than a churned user. The Founding Growth Lead and Founder's Office roles extend that logic to go-to-market: sellers who speak to loan-level pricing adjustments and servicing transfer timelines, not just "AI automation" slideware.

Domain fluency is the hardest filter to fake. The Mortgage Originations Process Specialist role exists because Kastle's agents originate loans for the largest U.S. banks; the company has processed over $1.8 billion in volume. The Enterprise Account Executive and Head of Sales carry the same requirement: they sell into mortgage banking executives who measure vendor risk in basis points and consent orders.

The part-time AI Trainer at $30/hr remote reveals the data side of the screen. Training data for mortgage agents comes from real call transcripts and servicing records. Kastle pays a premium for that precision.

What the screen filters out is instructive. Candidates whose AI experience is academic (papers, benchmarks, model training without production serving) will not clear the technical bar for the Staff or Forward Deployed roles. Engineers who want pure backend or infrastructure work without customer-facing deployment cycles will churn fast; the board shows zero pure "Software Engineer" listings, only deployment-flavored titles. Remote-only applicants have effectively one slot: the part-time AI Trainer. The Software Engineer - Intern is on-site in San Francisco. Mortgage novices applying to product, deployment, or specialist roles will stall at the domain screen; Kastle's agents execute loss mitigation and collections calls. Finally, anyone confusing this Kastle with Kastle Systems (physical security) or Kastle International (legal/financial consulting) is reading the wrong jobs page — those career sites list security technicians and international counsel, not AI agent builders.

The roles on the board, especially the Head of Talent and Technical Recruiter (Contract) added in the past week, show the screen tightening, not relaxing. The company is hiring recruiters who can recognize the hybrid profile at volume.

Conflicting Counts: Why Job Boards Report 11 to 21 Roles

Depending on where you look, Kastle has 11, 13, 14, 15, 16, or 21 open roles. That ten-role gap isn't noise — it maps how each aggregator ingests, filters, and ages the same underlying feed.

Start with the primary source. Kastle posts exclusively on its own careers site, www.kastle.ai. EngRadar scrapes that page daily and counts only direct-apply roles: no recruiters, no reposts. Its latest snapshot, dated 2026-10-18, shows 15 open roles, a net gain of two over the prior 28 days and marked "accelerating." Two months earlier, on 2026-08-27, the same tracker read 13 roles with the same +2 net but labeled "cooling." The methodology hasn't changed; the velocity signal flipped because the underlying posting rate did.

HighFit polls Kastle's careers page every few hours and mirrors whatever it finds. On its company page the count has flickered between 14 and 16 roles in October 2026, with timestamps showing two postings in the past week and three in the past 30 days. That hourly cadence catches roles the moment they appear, and the moment they disappear, so its number sits closer to the live page than any daily scraper.

Landed.jobs and Fast AI Jobs both reported 11 roles as of mid-October (Landed specifies 2026-10-13). Both appear to sample less frequently and to apply their own seniority or location filters; Landed notes 9% remote and 33% senior among the 11 it sees. Jobby shows only 4 roles and flags the newest posting as two months old: a clear case of stale ingestion.

LinkedIn introduces geographic slicing. Its "San Francisco" view lists 21 roles; the "Worldwide" view lists 14. The delta suggests seven roles tagged to the Bay Area that either lack a remote flag or are duplicated across location facets. LinkedIn's "new jobs added daily" disclaimer reflects employer-side reposting behavior, not Kastle's actual publishing rhythm.

That board lists 17 live roles on Kastle as of this writing, with two added in the past seven days. That count sits between EngRadar's 15 and LinkedIn San Francisco's 21, and it updates faster than the weekly or monthly snapshots most public boards publish.

What drives the variance? Three mechanics. First, ingestion cadence: hourly (HighFit) vs. daily (EngRadar) vs. weekly or worse (Jobby, Landed, Fast AI Jobs). Second, filter logic: direct-apply only (EngRadar), all listings (LinkedIn), seniority-tagged subsets (Landed), or remote-only slices. Third, de-duplication: Kastle's own page shows each role once; LinkedIn may surface the same requisition under multiple location tags, inflating the San Francisco count.

Posting velocity is measurable. EngRadar's +2 net over 28 days (as of 2026-10-18) matches that board's two new roles in the past week. HighFit's "posted this week: 2" aligns. The signal is consistent: Kastle adds roughly one role every two weeks, not in bursts. The 21-role LinkedIn figure likely includes roles that closed weeks ago but remain indexed, plus location duplicates. The 4-role Jobby figure is the inverse: roles that closed months ago and never got purged.

For candidates, the takeaway is practical: the freshest view is that page or a board that polls it hourly. Any count older than a week trades on stale inventory.

KastleOS: The Mortgage Automation Product Behind the Hires

Kastle does not sell a generic voice bot. It sells KastleOS, an AI-agent operating platform built for consumer lending and mortgage servicing, starting with the high-volume, high-compliance contact centers that sit between borrowers and the institutions that own their debt. The product context explains the hiring spike: every open role maps to a component of a system that must operate inside regulated workflows, on top of legacy infrastructure, without breaking the rules that govern every borrower interaction.

The platform deploys a roster of narrow, pre-trained agents rather than a single do-everything model. Each agent mirrors a role that already exists on a servicing floor. The Servicing Agent handles payment reminders, takes payments, and answers routine questions on escrow, balances, and statements around the clock. The Collections Agent engages delinquent borrowers and negotiates within compliance guardrails to bring past-due accounts current. The Compliance/QA Agent scores every call and document against the regulatory rulebook — FDCPA, TILA, UDAAP, RESPA, TCPA — replacing sampled spot-checks with near-full coverage. The Loan Officer Assistant qualifies new inquiries, generates warm transfers, and clears the manual follow-up loan officers dread. A Loss Mitigation Agent extends the same approach into hardship and modification workflows. An omnichannel layer keeps the conversation consistent across voice, SMS, email, and webchat, handing off to a human the moment a call needs one.

Compliance is not a bolt-on. The agents are pre-trained on the alphabet soup that governs lending so the guardrails are built in. The company carries SOC 2 Type I and Type II certification and is GDPR-ready. Every interaction, whether handled by AI or human, is scored in real time with automated scorecards, targeting 99.5% QA coverage. The compliance engine is the moat: horizontal voice tools stall at the door of a regulated industry because they cannot demonstrate per-interaction adherence to federal lending law.

Integration is the other half of the product. KastleOS connects directly to ICE Mortgage Technology's MSP, the servicing system of record that a large share of the industry runs on, plus consumer-direct CRMs and the major contact-center platforms (Five9, Talkdesk, Genesys, NICE). The pitch to servicers is explicit: no core replacement, no multi-year migration. Agents map to existing workflows and go live in days. The platform includes an Agent Builder that turns SOPs into AI agents using natural language, a Simulation Testing framework that uses AI to stress-test edge cases before production, and a proprietary inference layer that makes every AI response explainable: a requirement for audit trails in financial services.

The operational proof points are specific. Carrington Mortgage Services, with over $200 billion in unpaid balances across more than a million loans, chose Kastle in July 2026 to deploy autonomous and agent-assist AI across its servicing contact centers. Kastle is in production at 10 of the top 25 US servicers and Y Combinator's data shows it has processed more than $1.8 billion in transactions. Stated benchmarks: roughly 70% lower cost per call than a staffed seat, 40% lower average handle time, 90%+ customer satisfaction. The Series A — $24 million led by Insight Partners with Y Combinator, Commerce Ventures, and those investors participating — is explicitly earmarked for expanding engineering, product, and go-to-market teams and accelerating deployments with leading banks across North America.

For applicants, the product domain means two things. First, the technical bar is shaped by regulated production: inference latency, explainability, compliance scoring, and integration reliability matter more than model novelty. Second, the roadmap is vertical. The founders describe the longer-term ambition as an "operating system for consumer lending": mortgage first, the rest of the lifecycle later. The hiring spike reflects that sequence: forward-deployed engineers who ship inside a servicer's environment, agent product managers who understand loss-mitigation workflows, AI trainers who curate compliance-critical data, and sales leaders who speak servicer economics. Anyone joining builds infrastructure that a $200B servicer runs on daily.

Who Should Apply — and Who Should Self-Select Out

The screen Kastle runs is narrow by design. Its open roles cluster around three constituencies: engineers who have shipped autonomous agents into production, product people who have owned agent-facing roadmaps, and operators who know mortgage servicing workflows cold. The first-party board data from Kastle makes this visible: every senior engineering role lists eight-plus years of experience and a compensation band that starts at $250K (Staff AI Engineer, Staff Engineer). Those roles sit at the customer boundary; they require engineers who can debug a voice agent on a lender's call floor, not researchers who hand off a model card.

Product roles tell the same story. Those listings both ask for five-plus years and explicitly tie the work to mortgage workflows: borrower servicing, payment collection, loss mitigation, collections across those channels. That role has no years floor but demands domain fluency: you write the playbooks the agents follow. The AI Trainer (Part-time) is the only remote role on the board at $30/hr; it targets people who have labeled or evaluated conversations in a lending contact center, not generalist annotators.

Sales and growth roles mirror the buyer. Head of Sales, Enterprise Account Executive, and Founding Growth Lead all sit in San Francisco and carry five-plus year requirements. The buyers are America's largest banks and mortgage servicers — Kastle's site notes $1.8 billion in processed transactions — so the sales motion is enterprise, multi-stakeholder, and compliance-heavy. Those roles exist to keep this flywheel spinning; they need recruiters who can close senior AI talent in a market where everyone is hiring.

The Founder's Office and Founder Intern roles are the exception that proves the rule. They take candidates with "any" experience level (new grads included) but the signal they screen for is high-agency execution in ambiguous, high-stakes settings.

Who should self-select out: the same filters apply: candidates with only academic AI backgrounds won't clear the technical bar; engineers seeking pure backend work without customer-facing deployment will churn; remote-only applicants have only the part-time trainer slot; the intern role is on-site; mortgage novices will hit the domain screen; and those confusing Kastle with the security or consulting firms are looking at the wrong jobs page.

What This Story Does Not Claim

This article examines Kastle's hiring screen: the technical and domain filters that separate candidates who advance from those who don't. It does not make claims about compensation benchmarks, workforce displacement in mortgage servicing, or the physical security company that shares the Kastle name. Each of those narratives exists in the public record; each is distinct from the screen this piece documents.

Start with the security company. Kastle Systems operates in building security: "innovative outsourced security services significantly reduce costs and improve the critically important, 24×7 performance of security systems for building owners, developers, tenants and residents," per its own site. It has a careers page touting Culture Excellence awards for Diversity, Equity & Inclusion from Top Workplaces. Its salary data on levels.fyi shows a median yearly total compensation of $123,533 across job families, with a Software Engineering Manager at the high end of $150,750. That company is not the subject here. The Kastle hiring for 17 roles is the AI agent company founded in 2024, Y Combinator S24 batch, 15-person team (per Y Combinator), $26.8M total funding (per Fast AI Jobs), based in San Francisco with in-person roles only. Its co-founders are the same duo. Different founding years, different domains, different capital structures, different compensation profiles. Conflating them is a category error this article avoids.

On displacement: Kastle's product, KastleOS, automates borrower interactions across voice, SMS, email, and chat: payments, escrow questions, loss mitigation, compliance monitoring. New American Funding (5,300+ employees, $73 billion servicing portfolio, 281,000+ customers) and Carrington Mortgage Services (~$200 billion servicing portfolio) have deployed Kastle agents in their contact centers. Kastle reports roughly 70 percent lower cost per call, 40 percent lower handle time, and CSAT above 90 percent. New American Funding's president, Christy Bunce, said the partnership lets them "serve each customer with greater speed and consistency throughout the life of their loan" while freeing operations teams for "complex client situations." Rishi Choudhary, Kastle's CEO, said lenders who "deeply embrace AI without compromising on compliance or control" will win the next chapter. Those are the partners' stated positions. This article does not extrapolate from them to a thesis about net job loss, role transformation across the mortgage industry, or the future of contact center employment. The hiring screen is the subject; the macroeconomic ripple is not.

The screen itself rewards production agent experience and mortgage-domain fluency: voice AI and LLM application work, consumer lending knowledge, regulated financial software background. It penalizes generalist credentials without that specificity. That is the claim. The adjacent narratives, salary surveys, industry displacement forecasts, the other Kastle, are real and documented. They are also outside the frame.

Who Is Watching

Job boards are the first to register a hiring spike, and Kastle's numbers diverge across the aggregators that track early-stage AI companies. Landed.jobs captured 11 open roles as of October 19, 2026, with 9% remote and a third tagged senior. Two days earlier, aillmjobs.com counted 10 roles across five teams and two locations: three in product, three in engineering, nine on-site, one remote. Y Combinator's own jobs portal lists Kastle as actively hiring without a specific count. Its board shows 17 live openings on Kastle as of the latest sync. The variance — 10 to 17 roles depending on the source — reflects posting velocity and sync lag more than disagreement about Kastle's intent. Boards that scrape weekly will undercount a company adding two roles in seven days, as Kastle has done.

Candidate pools are sorting themselves along the lines Kastle's screen establishes. Mortgage servicing veterans, people who have worked loan-level queues at Carrington or large sub-servicers, recognize the operational language in those postings. They see a product that plugs into the servicing system and telephony layer they already know. AI-agent builders from the applied-LLM layer, engineers who have shipped voice agents to production, not just prompt-chained demos, map to the Staff AI Engineer, Forward Deployed Engineer, and Agent Product Manager roles. Fintech generalists without mortgage-domain fluency or production-agent mileage are the cohort self-selecting out; the screen filters them before a recruiter spends time.

Adjacent employers watch for different reasons. Carrington Mortgage Services, Kastle's announced deployment partner, has a direct view of the agent architecture in production: autonomous customer-service and collections agents resolving interactions end-to-end. Other top-20 servicers and sub-servicers, unnamed in public filings but referenced as "largest mortgage lenders in America" across Kastle's own materials, are evaluating whether the 90% call-center cost reduction and 40x usage growth in six months translate to their portfolios. Competing AI-agent startups in the financial-services vertical track Kastle's role mix to infer product direction. A spike in Forward Deployed Engineer and AI Trainer (Part-time) hires signals a push toward customer-side deployment and data-labeling velocity, not core model research.

The $24M Series A led by Insight Partners, closed after the YC S24 batch, amplifies the signal. Investors who passed or stayed on the sidelines now see a 15-person team. They are the talent magnet. Their next hires will indicate whether Kastle doubles down on mortgage servicing or expands into adjacent loan verticals. The job boards will catch that pivot in the role titles before any press release does. The screen, meanwhile, stays narrow: the next Staff AI Engineer still needs to have shipped an agent that a regulator has audited.

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