The Work Itself
Raindrop, the AI agent monitoring platform that raised a $15M seed led by Lightspeed Venture Partners, as PR Newswire reported, operates with a compact team in San Francisco where its scale shapes how work gets done. The company sits at the intersection of two unsolved problems: agents that run autonomously for hours across tool chains, and legacy observability that stops at latency and token counts. That technical scope — custom models per customer, an A/B testing platform for agent pipelines, millions of events a day — sets the tempo.
The founding trio shapes every decision. Zubin Koticha, CEO, was building a coding agent before Raindrop and ran into the same blind spots his customers now face; he and Alexis Gauba previously sold a company to Coinbase. Ben Hylak spent four years on Apple's Human Interface Design team after engineering roles there. The Zero G Talent board shows six engineering roles open (see table), including product, security, ML, backend, developer experience, and forward-deployed, with a board salary band across 9 salaried roles (see table).
| Role / Category | Salary Range | Median | Source | Notes |
|---|---|---|---|---|
| Six engineering roles (product, security, ML, backend, developer experience, forward-deployed) | $150k–$250k | $250k | Zero G Talent board | 6 open roles |
| All salaried roles (board-wide band) | $110k–$250k | $250k | Zero G Talent board | 9 roles total |
| Forward-deployed engineer (FDE) | $130k–$200k | — | Zero G Talent board | Listed alongside core engineering tracks |
Lightspeed led the $15M seed with participation from the founders of Replit, Cognition, Framer, Speak, Notion, Vercel, and Figma — investors who are also builders of the very agent stacks Raindrop monitors. When Tolan's CTO Evan Goldschmidt says Raindrop has been "invaluable as we've been growing quickly," he describes a feedback loop: a customer hits a new failure mode, Raindrop's background agents surface the pattern, the team ships a model update, and the customer validates it in Experiments. The monitoring runs in production because the only way to detect silent agent failures is to watch real trajectories.
Decision authority mirrors the cap table. Technical calls (model architecture, data pipeline, API design) are made by the engineers closest to the problem, with founder review on direction. Product priorities emerge from the customer signal the founders hear directly. Operational choices (hiring, tooling, security posture) fall to the same small group. The salary bands reflect the expectation that each hire operates at a high level of autonomy. The team does not have the headcount to carry specialists who only function in a narrow lane.
For someone who needs a spec handed to them, the friction is immediate. For someone who has been waiting to own a technical surface end to end, the absence of guardrails is the feature.
Operating Principles: A Survival Manual
Raindrop's operating principles read less like a values poster and more like a survival manual for a team building infrastructure that sits between unpredictable agents and production consequences. The company's stated philosophy centers on a phrase repeated across interviews: "being honest with yourself." It sounds generic until you see how it shapes product decisions, hiring, and internal disagreement.
The principle originated from the founders' own frustration. When Hylak, Koticha, and Gauba built Sidekick — their earlier AI coding agent — they discovered that existing observability platforms counted tokens, scored hallucinations, or tagged sentiment, but none told them whether the agent actually did what the user wanted. The gap wasn't academic. As Koticha put it, "we have started calling it humanity's last problem internally, because agents are being deployed in finance, military, healthcare, and when they fail it is catastrophic." That framing forces a specific honesty: you cannot pretend a dashboard solves the problem if the dashboard doesn't catch the failure mode that gets someone sued — like the Air Canada case where an agent promised a refund the company hadn't authorized, and the customer won in court.
Honesty with yourself translates into dogfooding as a non-negotiable. "Our team would never want to build in a space where we were not also the user," Hylak said. The team uses Raindrop heavily internally, and that usage drove Workshop, a local tracing tool that eliminated the seconds-long latency of waiting for remote dashboards to update. "If I am running my agent locally I should be able to see my traces locally," Hylak said. The decision came from sitting with the product side-by-side with the agent and noticing the friction. That loop — build, use, notice, fix — is the primary product development mechanism.
It also translates into vocabulary discipline. Raindrop invented its own taxonomy: stumbles (every user friction event), issues (actionable regressions affecting multiple users), and experiments (behavioral changes you A/B test rather than fix). "The most important thing is the vocabulary," Hylak said. "Imagine if we did not call issues issues, if we called them alerts. A user comes and asks what issues exist, and you have to say there are no issues. Weird things like that. Creating and defining your own vocabulary, because this is an entirely new problem set, is very important. The hardest thing is keeping that vocabulary clear and consistent, especially because the space is changing constantly." The precision isn't branding; it's a coordination mechanism for a small team moving fast in undefined territory.
The contrarian streak is deliberate. When Raindrop launched, "the predominant wisdom was that every team needs to write a ton of evals," Koticha said. "We do not love writing evals, the same way we do not write that many unit tests except for the core parts of the app. What we love is that we have Sentry, and when something goes wrong we get an alert, and that lets us ship really fast on the things that are less critical." Koticha got into a public Twitter fight with a major eval company over it. "Nobody really thought it mattered, which is crazy. There was a lot of cognitive dissonance. You have something that is working and you do not want to change it." The team bet on monitoring over pre-deployment evaluation, a bet that only pays off if you're honest about what monitoring actually catches.
That honesty requirement extends to internal disagreement. Raindrop's three-founder structure (Hylak and Koticha were best friends and roommates; Koticha and Gauba met in college, co-founded a company, took it through Series B, and sold it to Coinbase) creates a high-trust container for conflict. But the principle applies beyond the founding team. The hiring bar selects for curiosity and the ability to "unravel fundamental rules about the universe," in Hylak's phrasing, people who treat design and engineering as truth-seeking rather than execution.
The principle also governs what Raindrop won't build. "A lot of people would rather cram it into their app somehow, and we would rather build the thing that should exist," Hylak said. That restraint shows up in the product's integration philosophy: first-party support for Cognition's Devin to run auto self-healing loops, but no attempt to become the agent framework itself. The team stays in the observability layer because that's where the unsolved problem lives, and because pretending otherwise would violate the honesty rule.
Speed is the other operating constant. "The reason our company is doing well right now is that we are constantly at the bleeding edge of what is possible and how agents are changing, building the tools people will need a month or so out," Koticha said. That pace demands a team that can decide without consensus theater. In a small team, every person carries enough surface area that a wrong call shows up in the product immediately. The cost of pretense is visible in days, not quarters.
The principles cohere around a single constraint: the stakes of the problem space. When agents fail in healthcare or finance, the failure isn't a bad user experience, it's liability, regulation, harm. Raindrop's values are the ones that survive contact with that reality. Everything else gets discarded.
The Hiring Bar
The same honesty that shapes product shapes hiring. Hylak describes the target as people with "that quality", engineers who treat design and code as truth-seeking, not execution. The six open roles on the Zero G Talent board each require the candidate to demonstrate they can own a surface area end-to-end.
The forward-deployed role is the clearest signal. At its posted range (see table), it sits alongside core engineering tracks, not below them. The FDE ships custom models for customer agent pipelines, debugs in production, and translates field feedback into product changes. Raindrop does not hire solutions engineers who hand off to "real" engineers. The FDE is the real engineer, and the customer call is the spec review. Candidates who view customer interaction as a distraction, or who prefer to stay several layers removed from deployment reality, self-select out before the offer stage.
The bar also selects for calibrated tolerance of founder-mode decision-making. The median posted salary across roles that span individual contribution and high-leverage technical leadership, per Zero G Talent's figures, buys autonomy. Candidates who need consensus-driven culture, formal career ladders, or insulated scope will chafe. The ones who stay are those who have reported directly to a technical founder before, or who have been a technical founder themselves, and who can disagree productively without needing a process to mediate.
What the Reviews Don't Tell You
Public review sites show multiple entities named "Raindrop," and the feedback attached to them does not map to the San Francisco frontier-tech team hiring across six engineering specializations. Glassdoor lists "Raindrop Marketing" at 4.0 out of 5 stars across 40 reviews, placing it within one standard deviation of the Media & Communication industry average of 3.7. A separate Glassdoor entry shows only two reviews. Indeed hosts reviews for "Raindrops" in film production and for "Raindrop Car Wash." None of these listings reference the technical roles, San Francisco location, or salary bands that appear on that board for this Raindrop.
The absence of a substantial, attributable review corpus for the specific company profiled here is itself a signal. Early-stage, founder-led teams often operate below the radar of aggregate review platforms, either because headcount is too small to generate statistically meaningful samples or because employees are bound by the same high-autonomy, low-bureaucracy culture that discourages performative feedback. Candidates should treat the lack of public reviews as a data point: it suggests a private, tight-knit environment where reputation spreads through networks rather than forums.
What can be inferred from the board data and the hiring pattern? The company is actively recruiting across six engineering specializations simultaneously, with a median posted salary and top-of-band (see table). The Raindrop Marketing profile (40 reviews, 4.0 stars, media industry) is almost certainly a different organization; its industry tag and review volume don't align with a team posting six simultaneous engineering roles at $150k–$250k in San Francisco.
For a candidate, the practical takeaway is to bypass aggregate scores and seek direct signal: probe how the team handles a product pivot when the only stakeholders in the room are the people who wrote the code.
Who Stays, Who Leaves
The role mix on Zero G Talent's board paints a clear picture: a San Francisco–based, founder-led frontier-tech team hiring the six engineering roles mentioned earlier at the posted salary bands (see table). That spread (spanning core infrastructure, applied ML, and customer-facing deployment) signals a company that ships technical product end to end, with a small team carrying the full stack. In that context, the traits that predict success follow directly from the workload and the organizational shape.
People who thrive tend to share three characteristics. First, they operate without a safety net of process. When the board shows nine salaried roles covering the aforementioned roles, the implication is unavoidable: each engineer owns a surface area that at a larger company would be split across multiple teams. Candidates who have only ever worked in environments where tickets are groomed, specs are handed down, and on-call rotations are shared across dozens of people often struggle to adjust. The ones who ramp fast are those who have previously been the "only person who knows this system" (whether at an early-stage startup, a skunkworks project, or a small team inside a larger org) and who treat ambiguity as a design constraint rather than a blocker.
Second, they treat the forward-deployed engineer role as a signal, not an outlier. The existence of an FDE role alongside core engineering tracks means Raindrop expects engineers to sit with customers, debug in production, and translate field feedback into product changes. Engineers who do will burn out. The ones who last are the ones who have done solutions engineering, field engineering, or founder-led sales engineering before and know how to close the loop without losing technical depth.
Third, the same hiring criteria apply: candidates needing it or formal career ladders will chafe, while those who stay have done so and can disagree productively without process.
The burnout profile is the mirror image. It includes engineers who optimize for predictability (fixed sprint cadences, clear handoffs, dedicated platform teams) and who feel anxiety when the on-call pager is also the product pager. It includes specialists who want to go deep on one layer (say, model training) without touching the serving stack, the data pipeline, or the customer integration. And it includes operators who expect HR-owned performance cycles, structured mentorship programs, and a manager whose primary job is people management. At Raindrop's scale, the manager is also the tech lead, the architect, and sometimes the recruiter.
No employee reviews specific to this San Francisco frontier-tech team appear in the research. That absence is itself a data point: at nine roles, the team is small enough that public reviews are scarce, and the signal lives in the roles themselves.
The question isn't whether the culture is intense. The question is whether you want to be the one holding the pager when the agent fails in production, and whether you trust the other people in the room to hold it with you.
Working in frontier tech? Zero G Talent tracks the openings: see every open Raindrop role, browse frontier tech jobs, the companies hiring, and the people building the field.