How work actually gets done
Kavak has restructured around AI agents rather than layering tools onto existing workflows. Alejandro Maza Ayala, chief product and AI officer, described the shift as a radical transformation into an AI-native organization in a 2024 a16z podcast and Tokenless newsletter deep-dive. The company instantiates 100,000 to 200,000 customer-specific agents daily, each with its own virtual machine and persistent memory. Those agents convert 2.1 times more customers than human-only teams and handle 96 percent of all interactions without human involvement.
Kavak's 8,000-plus employees work across six countries: Mexico, Brazil, Argentina, Chile, Colombia, and the UAE. The software-heavy workforce organizes in small teams. The roughly 800 mechanics in Mexico use agents as sidekicks, surfacing information while technicians work.
The "Rules of the Game" onboarding session simulates real experiences where new hires discuss how scenarios reflect Kavak's ethics. Vault by Diligent provides a multilingual reporting channel with anonymous and GoTogether pattern-matching options.
Glassdoor data shows culture ratings climbing 11 percent over 12 months, suggesting the speak-up investment and operational changes are registering with employees.
The principles that show up in practice
Kavak publishes three principles on its careers site that "define how we work, how we decide and how we treat our customers and our team." The clearest translation of principle into practice is the speak-up culture built to solve a concrete problem: 8,000-plus employees across six countries needed a reporting mechanism they would actually use. Kavak selected Diligent Vault and ran a comprehensive roll-out plan to drive adoption.
A second principle appears in the AI-native reorganization led by Maza Ayala, who describes a deliberate redesign of systems, teams, and customer experience around AI agents. The decision to measure success by autonomous resolution rates signals a culture that treats velocity and measurable autonomy as operating defaults.
Founder and CEO Carlos García Ottati, a Venezuelan entrepreneur who built Mexico's first unicorn, leads an organization where small teams ship decisions. The ClavePrep hiring guide notes that Kavak interviews "still reward clarity, role fit, and a crisp project story," and warns candidates against "memorizing trivia about Kavak instead of practicing spoken problem-solving." That emphasis on demonstrated problem-solving over brand loyalty aligns with the autonomy the company claims to value.
Employee sentiment data reflects tension between stated values and lived intensity. Glassdoor ratings show culture at 3.4 out of 5, below the IT benchmark, but an 11 percent climb over the prior 12 months. The upward trend matters: a company that tracks its own cultural health quarterly is already practicing a form of transparency.
What the hiring bar selects for
Kavak's hiring funnel — application screen, OA or take-home, technical deep-dive, hiring-manager/bar round, offer — filters for engineers and product people who can operate at speed in a high-autonomy environment. The company runs roughly 3,500 people and processes around 10,000 transactions a month at ~$0.5B revenue. Recruiters and automated screens first check JD-keyword alignment, credible ownership bullets, and eligibility (location, level, work authorization). A résumé that buries impact under tool lists rarely clears the volume filter.
The OA or take-home stage rewards timed accuracy and clear communication of approach. Interviewers listen for fundamentals, debugging habits, and how candidates handle follow-ups; they want the plan narrated before the code, complexity explained in plain English, and trade-offs stated when stuck. At the technical deep-dive, the signal shifts to system or domain depth and recovery from ambiguity.
The hiring-manager/bar round calibrates level and team fit. The bar expects two STAR stories with metrics: ownership framed as "I owned X; partners handled Y," not "we decided." Interviewers probe for conflict/recovery narratives and a crisp "Why Kavak?" that references the specific business unit and Internet context, not generic brand praise. ClavePrep's mock rubric scores six dimensions: clarity (states plan, then executes), ownership (clear personal contribution), fundamentals (correct basics even when stuck), recovery (names trade-offs, asks clarifying questions), time use (hits a complete answer in the box), and fit signal (business-unit/Internet reason).
Track differences sharpen the selection. Campus and early-career funnels compress to days or ~2 weeks with OA-heavy, compressed interview days. Lateral and mid-level loops stretch 2–6+ weeks, adding deeper ownership stories. Senior and leadership tracks run longer calendars with multi-panel strategy and org-impact narratives and fewer timed OAs. Across tracks, the consistent selectors are: demonstrable ownership with numbers, structured problem-solving under time pressure, fundamentals that hold under follow-up, and a narrative tying the candidate's experience to the specific Kavak unit they'd join.
First-party board data shows active requisitions for Junior Software Engineer (Mexico City), Senior Product Manager (Mexico City), Senior Software Engineer (Mexico City), and Senior to Staff Software Engineer (Mexico City | Buenos Aires), roles that map to the same funnel with level-appropriate depth gates. Candidates who treat every req as identical, skip timed practice, or surprise HR late with location or authorization constraints tend to stall. The ones who clear the bar arrive with a funnel map, two rehearsed STAR stories, a tailored "Why this business unit?" answer, and logistics already resolved.
What current and former employees say
Glassdoor's main U.S. site shows 1,096 anonymously submitted reviews for Kavak as of September 2026. The Singapore and Australia mirrors report smaller samples — 81 and 83 reviews respectively — suggesting the bulk of English-language feedback comes from the company's Mexican and U.S. operations. Across all three portals the overall rating clusters between 3.2 and 3.5 out of 5, placing Kavak within one standard deviation of the information-technology industry average of 3.8.
The culture-and-values score sits at 3.4 on the U.S. site, 8 percent below the IT benchmark, while the Australian mirror logs 3.2. Work-life balance reads 3.3 in the U.S. and 3.2 in Australia. Career opportunities are consistently rated 3.1.
Recommendation rates tell a similar story. Sixty-three percent of U.S. reviewers would recommend the company to a friend; the review-only page puts it at 61 percent, and the Australian portal at 60 percent. That narrow spread across geographies suggests a genuine consensus. Roughly four in ten people who have worked at Kavak would not actively encourage a peer to join.
Reviews repeatedly surface two themes. Engineers and product managers praise autonomy, fast decision cycles, and the ability to ship code without layers of approval. The flip side, voiced by people in operations, support, and some mid-level management roles, is sustained intensity, ambiguous prioritization, and limited process guardrails.
No LinkedIn aggregate sentiment data or comparable public employer-review platforms (Blind, Levels.fyi, Comparably) were captured in the research corpus, so the Glassdoor picture — imperfect, self-selected, and English-language heavy — is the only quantified external view available. The consistency across three regional Glassdoor instances reduces the likelihood that a single brigade or review campaign is skewing the signal.
The data reinforces a profile of a company that rewards self-starters comfortable with ambiguity and penalizes those who need structured onboarding, predictable rhythms, or explicit career ladders. The 60-something percent recommendation rate is the market's verdict: Kavak works for a specific slice of talent, and the rest leave.
Who thrives here and who burns out
The operating model is the filter. Decisions move through small teams with minimal approval layers. For engineers and product managers who measure progress by shipped code and measurable metrics, that autonomy functions as acceleration. The board's live postings signal a hiring profile weighted toward individual contributors who can own a problem end to end: Senior Software Engineer roles in Mexico City and Buenos Aires, Senior Product Manager positions in Mexico City, and a Junior Software Engineer slot in Mexico City.
Negative reviews cluster around ambiguous prioritization, on-call expectations that bleed into evenings, and the cognitive load of context-switching across multiple initiatives. The work-life balance score is not a catastrophe, but it is a signal, especially when paired with 2026 neurological research on remote-work burnout identifying blurred boundaries, asynchronous overload, and the loss of informal recovery moments as primary drivers of exhaustion.
Who burns out? The pattern points to operators who thrive on clear handoffs, defined ceremonies, and predictable cadence: project managers accustomed to RACI matrices, engineers who want a sprint plan locked two weeks out, and people who recharge by physically leaving a building. The remote-first environment removes the natural circuit breakers those profiles rely on. Who thrives? Self-starters who treat ambiguity as design space, who document their own decisions because no one else will, and who calibrate their own sustainable pace without waiting for a manager to notice. The year-over-year culture improvement suggests leadership is aware of the attrition risk, but the fundamental trade-off remains: autonomy without guardrails rewards the disciplined and penalizes the dependent.
The agent fleet spins up 100,000 instances a day. The people who stay are the ones who can operate with similar independence.
Working in frontier tech? Zero G Talent tracks the openings: see every open Kavak role, browse frontier tech jobs, the companies hiring, and the people building the field.