How Work Gets Done: Pace, Structure, and Decision Making
Sesame’s full software team is under 15 people, including ML, infrastructure, and everything else, per a 2025 a16z founder interview. That small team ships a voice assistant preview (Maya and Miles, live at app.sesame.com) while hardware roles in Taipei point toward a 2027 eyewear launch promised on the company site. The tension between weekly software iterations and multi-month hardware cycles sets the pace.
Zero G Talent’s board lists six open roles across San Francisco and Taipei: Advanced Manufacturing Program Manager, Global Supply Chain Manager (CapEx), Head of Sales, People Lead, Creative Director, and iOS/Android Automation Engineer. Salary bands run roughly $127k–$280k (median $280k) across 24 salaried positions. The spread signals a founding team that keeps decision rights close.
Glassdoor shows 74 percent of reviewers would recommend the company. Culture and values sit at 4.2 out of 5, work-life balance at 4.0, career opportunities at 3.7. The numbers suggest a place people respect but where growth paths are still forming. The 4.0 balance rating reflects that reality: good for a startup, but not a shield.
Values and Operating Principles in Practice
Sesame’s blog states its mission plainly: “Voice is our most intimate medium as humans. At Sesame, we are creating conversational partners that achieve voice presence, the quality that makes spoken interactions feel real, understood, and valued.” That sentence functions as both product brief and cultural anchor. In a 2025 a16z interview, the same language reappears when the team explains why they built their own speech generation model instead of licensing one: “the naturalness of the voice, getting the voice to kind of generate these sort of human kind of imperfections often that make you sort of feel like you're talking to a human that, you know, you kind of your brain gets tricked for like a second that, oh, maybe this is actually human.” The value dictates the technical roadmap. The team rejected off-the-shelf TTS because it couldn’t deliver the imperfections that create presence. They trained a base model from scratch, fine-tuned it for two personas (Maya and Miles), and open-sourced the base while holding back the demo service. That decision reflects a second stated principle: “We just kind of want to give back and be part of the community from like a research perspective and and open weights open source it's good for the research community. But there's this tension, right? Between giving back and holding on to core parts of valuable technology and we've got to build a real business.”
The tension is operational, not rhetorical. With a full software team under 15 people, resource allocation is a daily expression of values. “We don't have the resources to do everything, right?” the founder said. “We focus on the problems that are most important to achieve the kind of product experience that we want to achieve.” That prioritization shows up in what they don’t do: no frontier-scale pre-training, no developer API, no customer-acquisition open source push. “We are not a frontier model company. We're not pre-training LMs at insane scale and so forth.” Instead, the team spends cycles on “contextual speech”, the difference between reading text aloud and knowing how to say it in this moment, with this interlocutor. “You need more context to tell what is an appropriate way for this moment in the conversation.” The product goal forces every model decision through a product-experience filter. “Good taste in ML... picking the things that you have to do and not doing the things that you don't have to do.”
The interdisciplinary framing appears in the hiring signal. The founder emphasized: “We need great just ML people who are great at ML... But they also would care about the end customer experience, the end product experience, right?” That dual filter (technical depth + product taste) is the operating principle that turns a value statement into a hiring decision. It also shapes the release cadence. The team removed transcription from the inference path, where “the LM takes as input the audio directly and generates the response... it never goes the user's audio never goes through text”, because transcription loses paralinguistic signal. “Humans, of course, convey a lot of information through their speech that is not the words... and transcription misses that entirely.” The next model version will ingest audio natively. “We're moving towards just removing transcription entirely. That is coming and that's like not not, you know, years away or anything. That's coming soon.” The principle drove an architectural rewrite that a pure research lab might defer.
Open source policy makes the same trade-off visible. “We're open sourcing the speech generation base model basically... The base model can generate any any voice. We fine-tuned this model for Maya and Miles separately. We're not open sourcing the demo. We'll hold some things back for sure. We have to build a business and so on. Over time our models will get will get better. We'll open source some things. We're not going to open source everything.” The line is drawn at the product experience layer. Researchers get the foundation; the company keeps the differentiated application. That clarity lets a 15-person team avoid scope creep. Every open-source decision is a business decision, and the team treats it that way.
The nonprofit Sesame Workshop lists five values (integrity, innovation, collaboration, accountability, and a mission to help children grow “smarter, stronger, and kinder”). The AI startup Sesame shares a name and a heritage, with alumni from the Workshop founding Overplay, which went from “Sesame Street to Shark Tank”, but its operating values are distinct: voice presence over scale, product taste over benchmark chasing, selective openness over full transparency. In practice, those values mean a small team ships a voice model that leapfrogs better-funded labs on naturalness, then immediately identifies the next gap (audio-native understanding) and starts rebuilding the stack. The pace is set by the product bar, not the publication cycle.
The Hiring Bar: Traits and Signals That Get You In
Sesame runs a three-round interview loop that typically spans three to five weeks: a screening call, a take-home assignment, and technical sessions. Reported difficulty sits at 4.0 out of 10 across ten interview reports, split 50 percent easy, 40 percent medium, 10 percent hard. Dataford’s breakdown shows Sesame tests for behavioral interviewing and ML infrastructure at 100 percent each, system design at 96 percent, work ethics at 95 percent, and internal tooling at 92 percent. Background storytelling, data management, and conversational communication skills all clear 85 percent. Machine learning research collaboration, stakeholder interaction, and data modeling follow close behind. The signal is clear: they are hiring engineers who can operate inside a research-heavy product, not just ship features.
The must-have list is short and specific. At least three years of industrial software engineering experience. Demonstrated ability to work in high-ambiguity environments. Proficiency with database management and internal tooling. Experience working alongside ML or research-focused teams. Nice-to-haves include prior audio or text data processing, scaling consumer-facing applications, and a computer science degree.
| Role | Location | Salary Band |
|---|---|---|
| Advanced Manufacturing Program Manager | Taipei | 4.0–5.5M TWD/year |
| Global Supply Chain Manager (CapEx) | Taipei | 2.5–2.9M TWD/year |
| Head of Sales | San Francisco | $230k–$300k |
| People Lead | San Francisco | $220k–$280k |
| Creative Director | San Francisco | $200k–$280k |
| iOS/Android Automation Engineer | San Francisco | $175k–$280k |
Board-wide median: $280k across 24 salaried listings
Interview guides emphasize that the “why” behind technical decisions carries more weight than the code itself. Candidates are expected to have researched the product deeply, having tried the voice assistant if possible, understood the user experience, and formed opinions about the engineering challenges. “If you cannot access the product in your region, be prepared to discuss the company's mission and explain how you have researched their public-facing technology,” one guide states. Candidates who fail to show genuine interest or ask insightful questions about culture and engineering challenges are frequently rejected.
The take-home assignment functions as a portfolio piece. Evaluators look for clean Git history, professional commit hygiene, thorough documentation, and testing. Instructions are treated as a compliance check: ignore them and you signal you won’t follow team standards. The assignment also probes separation of concerns, state management, and how candidates handle large streams of audio or text data. Quality is prioritized over speed.
System design conversations go beyond architecture diagrams. Interviewers want to see how you structure data, manage dependencies, and ensure long-term reliability of the tools you build. Because Sesame is a growth-stage startup, projects often lack clear specifications. The ability to self-start and define requirements is weighed equally with technical output. Candidates should come ready with examples of defining project scope themselves and bridging research requirements into production-ready engineering.
Behavioral interviewing carries full weight. The company values engineers who have strong, reasoned opinions about architecture and can articulate them. Asking questions about the tech stack and team culture is expected, signaling that you are evaluating them as much as they are evaluating you. Experience sentiment across reports splits 50 percent positive, 30 percent neutral, 20 percent negative, suggesting the bar is real but not performative.
Total compensation estimates from self-reported data show a median of $467k with a wide band from $41k to $893k, though the first-party board data clusters more tightly around $127k–$280k for salaried roles. The discrepancy likely reflects level, team, and the small sample size (two data points for the higher figure). Either way, the hiring bar selects for people who can navigate ambiguity, collaborate with researchers, and treat internal tooling as a first-class product, exactly the profile that thrives in Sesame’s flat, fast-moving structure.
What the Structure Selects For
Zero G Talent’s board data shows 24 salaried roles posted with a median salary band of $280k, spanning hardware, software, supply chain, sales, and people operations, but hiring signals are not retention signals. What follows is an inference from the roles Sesame is funding and the product it’s building, not a summary of documented employee sentiment.
The hiring pattern points to a company betting on ambient voice agents that run all day on wearable hardware. The board lists an Advanced Manufacturing Program Manager and a Global Supply Chain Manager, CapEx, both based in Taipei, with compensation in the 4–5.5 million TWD/year range. Those roles exist because Sesame is moving toward physical product, with the site promising “all-day comfort with high-quality audio” and a 2027 hardware launch. An iOS/Android Automation Engineer in San Francisco ($175k–$280k) signals mobile client work. A Creative Director ($200k–$280k) and a Head of Sales ($230k–$300k) suggest the company is staffing for brand and go-to-market in parallel with engineering. A People Lead ($220k–$280k) at this stage usually means the headcount plan is aggressive enough to need dedicated recruiting and operations support.
The roles share a requirement the flat model makes explicit: operate without a spec. Candidates who need a detailed requirements document before writing code or cutting steel will stall. The ones who move fast are comfortable writing the spec themselves, validating it with a prototype, and throwing it away when the demo reveals a better path.
They also tolerate hardware’s unforgiving cadence. A software rollback takes minutes; a tooling change takes weeks and six-figure capital. The Taipei roles exist because the cost of a missed tape-out is measured in months, not sprints. Engineers who have shipped a consumer device through EVT/DVT/PVT, especially audio wearables, know the rhythm: long stretches of incremental bring-up punctuated by crisis moments when a yield issue or regulatory blocker appears. The people who stay motivated through that cycle are the ones who treat the crisis as the job, not an interruption.
Cross-functional fluency is non-negotiable. The voice agents (Maya and Miles) are the product, but the hardware is the delivery mechanism. A firmware engineer who can’t explain latency constraints to the LLM team, or a designer who doesn’t understand battery budget implications for always-on listening, creates friction that a flat org structure amplifies, there’s no manager to escalate to. The board’s salary bands ($127k–$280k median $280k) reflect a premium for that fluency: the market pays more for a mobile automation engineer who also understands BLE audio codecs than for one who doesn’t.
Specialists who expect deep focus in a narrow lane will struggle. The Creative Director role asks for “brand, product, and marketing” — three disciplines that usually sit in separate departments. The People Lead will likely run recruiting, onboarding, compensation, and culture with zero HRBP support. In a flat, fast cycle, the cost of coordination is pushed onto the individual contributor. Someone who thrives on clear handoffs and defined swim lanes will find the ambiguity exhausting.
The 2027 hardware target is another filter. That date is public on the site. It implies a fixed launch window for a first-gen consumer device with novel AI integration, a combination that historically slips. People who need predictable roadmaps to feel grounded will burn out when the schedule compresses. The ones who last treat the date as a forcing function, not a promise.
There is no public attrition data, no named former employees on record, and no review corpus to cite. If you’re evaluating Sesame, ask the hiring manager: “What’s the last decision the team made without founder input?” and “How did the last hardware delay get resolved?” The answers will tell you more than any salary band.
The 2027 eyewear launch will be the first real stress test of a model that has so far replaced management layers with a shared roadmap. If the voice agent in your glasses needs to understand a whisper in a crowded room, the engineer who hears the gap will be the one who closes it, no approval chain required. That’s the bet Sesame placed when it chose pace over hierarchy.
Working in frontier tech? Zero G Talent tracks the openings: see every open Sesame role, browse frontier tech jobs, the companies hiring, and the people building the field.