Research engineer joins Mistral after 18-month strategic push

In early September 2025, the author signed as a research engineer with Mistral, one of the few foundation model labs with more than $1 billion（約1600億円） in funding. This outcome was the result of an 18-month effort he describes as a deliberate interplay of strategic and tactical work, framed in a detailed retrospective on how to navigate hiring at major AI labs.

The journey began in April 2024 when the author decided he wanted to advance his career. He first sought a development conversation with his manager to explore growth within his current role, but found no short- or mid-term path forward. This prompted him to reach out to friends and contacts at major tech companies, startups, and frontier labs to gather intelligence: what skills were necessary, what portfolio projects impressed, what was the day-to-day like, and what were the growth trajectories? This exploration phase drove the first wave of strategic skill-building. He invested heavily in LeetCode preparation and obtained textbooks covering distributed systems, data structures, and algorithms. He sent his first application in August 2024 and progressed through six rounds before being rejected around November—a blow he attributes partly to nervousness and avoidable mistakes in interviews.

Unable to balance applications, interview prep, takehome exercises, and coding interview study while working full-time, he resigned his position in late 2024. Beginning applications full-time in January 2025 yielded disappointing results: many positions did not invite him for interviews, and those that did saw him stumble due to nerves and errors. At this point, he shifted back to strategy. He joined Recurse Center, a three-month, cohort-based programming retreat in New York, which he credits with providing both structure (alleviating fear of appearing as "a slacker" on his CV) and community. From February through mid-May 2025, he learned Rust, contributed 15 pull requests to scrutinized open-source projects (ruff and uv), and collaborated on a research paper with his master's thesis supervisor at the AI Safety Institute of the German Aerospace Center. He then returned home and entered a tactical application phase in June 2025.

The application strategy he employed—developed with advice from a friend at a large Silicon Valley company early in his journey but not executed until May 2025—proceeded in batches. He compiled a long list of target companies (ultimately about 40) and attempted to reach existing contacts or cold-outreach to gather insider information and referrals. For cold contacts, he wrote something like: "I'm Max and really excited about xyz and strongly considering applying to role abc. Is there anything you can share to help me make the best possible application?" He did not apply to all companies simultaneously; instead, each batch included one of his top choices (referred or highly vetted) alongside other companies he would consider. This sequencing allowed him to schedule lower-stakes interviews first, gaining routine and making mistakes in lower-risk settings before interviewing with top choices. Between June and August 2025, he accumulated approximately 60 touchpoints with 40 different companies, received a verbal offer from Mistral in mid-August, and formally signed in early September.

Underlying this execution was a clearly defined goal: he sought a "career-inflecting role" where he would build rare and valuable technical skills (software and ML engineering) in work he enjoyed, with ownership and impact, support from senior peers for growth into technical leadership, while remaining an individual contributor and living near people he cares about. This goal was general enough to apply to multiple companies but concrete enough to enable him to decline compromises—roles in large bureaucratic organizations or very early startups optimizing for MVP velocity rather than system quality. The goal effectively narrowed his search to roughly ten frontier labs, all highly competitive. He structured his goal-setting using jobsearch.dev and recognized that criteria like early ownership and growth opportunity are common at startups, while access to senior peers, brand prestige, and resources for niche skill-building are more common at established companies, making scale-ups an attractive middle ground. This framework gave him both motivation—reminding him why he had chosen this path during difficult moments—and strategic clarity for allocating effort.

The author's journey reflects a deliberate distinction between strategic and tactical effort in career advancement. Strategic actions—learning technologies deeply through substantial projects, maintaining tenure at reputable organizations, building networks, and developing a personal brand—require sustained effort over months but compound over time. Tactical actions—polishing a CV, preparing for interviews, reading company news—are lower effort but high-impact in the immediate moment. The author underestimated the value of tactics early on; he sent his first application in August 2024 after months of strategic prep, advanced six rounds, but did not land an offer. A critical turning point came when he realized that balancing both while employed full-time was unsustainable, prompting him to resign and then join Recurse Center—a structured three-month environment that allowed him to pursue ambitious upskilling (Rust, open-source contributions, research collaboration) without the career penalty of appearing directionless. Only after this phase did his tactical application efforts—batch processing, network referrals, insider information gathering—yield results.

The author's framework for goal-setting also shaped resource allocation and risk tolerance. Rather than pursue any well-paying role, he defined criteria that limited his search to roughly ten companies: frontier model labs with billion-dollar-plus funding. This constraint sacrificed breadth for depth, allowing him to invest selectively in network-building and insider intelligence for each target. His criteria also permitted him to decline roles that felt like compromises—positions in large bureaucratic machines or very early startups optimizing for speed over system quality—because his goal was clear enough to support such judgment calls. The application playbook he eventually followed (gathering intelligence, proceeding through batches, scheduling lower-stakes interviews before top choices) operationalized this strategy, turning abstract criteria into concrete actions. The result was not luck but the compounding effect of months of directed effort, coupled with disciplined execution in the final months.

No comments yet. Be the first to share your thoughts!