Adarsh Hiremath on Mercor, AI Recruiting and the Future of Work

Mercor co-founder and CTO Adarsh Hiremath joins Harry Stebbings to discuss building an AI recruiting marketplace, supplying expert labor to AI labs and the future of work. Hiremath has a direct commercial interest in the platform and its growth. The interview does not independently verify its revenue, hiring-quality or labor-market claims.

Hiremath says he met co-founder Surya as a child through competitive debate and later met Brendan at high school. He compares a debate partnership to a startup: both partners share an outcome, receive frequent feedback and must take ownership. Choosing a partner, he says, matters enormously in both settings.

The founders first operated a software development shop and recruited engineers in India. Hiremath says finding exceptional people became more interesting than the software itself, prompting them to automate candidate discovery and eventually parts of the company-facing side. That experience became the basis for Mercor's marketplace.

He recalls being physically at Harvard while spending much of his attention on the company. The decision to leave school was not obvious: the business had some revenue but no seed round or Thiel Fellowship at that moment, and his parents objected. He describes choosing to work with his friends as an emotional decision, not a general recommendation to drop out.

The founders moved to New York before raising seed funding; Hiremath says changing their salaries to $500 a month felt like an early milestone. He recalls raising more than $3 million in a seed round led by General Catalyst. These are his recollections of the company's early financing.

Stebbings cites $50 million of annualized recurring revenue in November and suggests the company was doing more by the interview, with 30 people at the earlier point. He later mentions a rumored 50% month-on-month growth rate; Hiremath discusses fast growth without providing complete definitions or independently audited figures. A run rate or ARR is not the same as realized annual revenue, and the host's claims are not verified here.

Asked about a 9 a.m. to 9 p.m., six-day schedule, Hiremath says the founders mentioned those hours partly to keep the team from working Sundays. He describes long hours as a byproduct of selecting people who care deeply about the mission and says momentum is energizing. This is a founder's characterization; the interview does not establish the experience or sustainability for every employee.

He says rapid growth constantly tests processes and forces people to outgrow their roles. Scaling culture can be harder than scaling software, he argues, because adding people quickly makes the initial team's norms harder to maintain. His emphasis on caring deeply about the mission reflects the company's hiring philosophy.

Stebbings asks whether Mercor is primarily labeling data for foundation models. Hiremath argues that advanced AI work increasingly requires finding experts in specific domains, making human data and talent assessment closely related. He says AI labs hire people through Mercor to help improve models after initial training; this is a company description, not independent evidence of placement quality.

Asked for a measure of hiring success, Hiremath says customer net retention is well above 100% and interprets expansion as a sign Mercor finds good people. The interview does not provide the period or calculation, nor does customer expansion establish individual hiring accuracy, worker outcomes or fairness.

He says the platform places people across engineering, law, medicine, finance and consulting, and describes an automated interviewer that tailors questions to a candidate and role. He claims an interview can be created in under ten seconds. The interview provides no independent validation of the system's accuracy, bias or suitability for specific hiring decisions.

Mercor uses multiple models, including OpenAI's, and Hiremath expects better underlying models to improve its interviewer and wider product. He predicts stronger domain-specific reasoning models and many specialized use cases, while expecting only a few firms to build broadly used foundation models. These are forecasts from a company that depends on such models.

On training data, Hiremath argues expert human data will remain important for evaluations, supervised fine-tuning and reinforcement-learning environments. Stebbings challenges him with the case for high-quality synthetic data; Hiremath says both can matter, but he sees qualified experts as a bottleneck. His commercial interest in supplying experts is relevant, and the interview does not establish which approach will dominate.

He expects humans to remain useful on a long tail of specialized tasks that models cannot complete. As an illustration he imagines AI doing 60% to 80% of a task and a person completing the remainder. Those percentages describe a hypothetical division of work, not measured system performance.

Stebbings asks whether fewer people will be needed as models improve. Hiremath predicts work will move toward greater specialization and sophistication, but does not quantify future employment or resolve the displacement question. His scenario for a far future with automated work and basic income is speculation, not an established timeline.

Hiremath says customers are impressed when the first candidates begin working and claims Mercor can find exceptional people at software-like costs. He says the founders handle sales and that most customer interest is inbound. These are company claims; the conversation does not supply controlled price, quality or sales-efficiency comparisons.

He says Mercor's side of the process—from a job listing and candidate profile through a personalized interview and payment—is automated. Asked about its take rate, he says it varies by customer, exceeding 30% in some cases and being lower in others. The interview does not provide an overall average or establish that every customer's own hiring decisions are automated.

The founders initially looked for talent in India, drawing on family connections and their parents' schools. Hiremath recalls manually interviewing an engineer who did not pass but sent a thoughtful explanation of his mistakes, leading them to hire him. He says the US later became the largest source of workers placed through the platform; he does not give a precise US share.

Asked whether students should still study programming, Hiremath says yes, but expects the work to move to a higher level of abstraction. He imagines people orchestrating many coding agents and says he uses Cursor for tasks such as generating tests and refactoring. His productivity and capability claims are examples, not a measured comparison.

He predicts coding agents will make software cheaper to build and increase the importance of network effects. Rather than say SaaS disappears, he expects products to replace entire services, including recruiting workflows. This is a forecast aligned with Mercor's business, not proof that software costs will approach zero or that all services can be automated.

Hiremath describes two proposed advantages for Mercor: a marketplace connecting more employers and candidates, and a data feedback loop using job outcomes to improve matching. He says quality drives customer retention and a fast revenue ramp. The interview does not independently establish the size of those effects or show that its predictions are accurate and fair across groups.

He would like Mercor's own hiring process to become fully automated before the founders meet a new colleague, though he says they are still working toward that. He favors in-person work in San Francisco and says proximity helps generate ideas and energy. These are his preferences and ambitions, not evidence that a particular schedule or hiring process works for everyone.

Hiremath recalls a product mistake: the founders made chat the sole interface for hiring because they expected all software interaction to shift to chat. Mercor later combined chat with other interfaces. He still considers conversational interfaces possible in the longer run, but says their timing was off.

He says several financing rounds were not initially planned and recalls Benchmark's Victor Lazarte pursuing a meeting with Brendan, followed by a helicopter ride with Peter Fenton. The founders decided they wanted Benchmark as a partner. About six months later, he says, Mercor raised another round while its revenue was in eight figures, without specifying an accounting period for that figure.

He confirms a $100 million round led by Felicis at a $2 billion valuation, with General Catalyst and Benchmark also participating. Stebbings characterizes the financing as approximately 5% dilution; the interview does not specify whether the valuation is pre- or post-money or detail all transaction terms. Hiremath says the company did not intend to spend the proceeds immediately, but wanted capital for a long-term marketplace effort.

In quick fire, Hiremath calls recruiting an underappreciated and prestigious function because talent shapes a company. He argues better matching supports efficiency even when organizations aim to do more with fewer people. That argument is consistent with Mercor's commercial objective and is not independent proof of its product's effectiveness.

He says he has become convinced that the next generation of SaaS will automate entire services, and reflects that building Mercor has been harder than he anticipated when the founders left school. Asked about advanced AI, he associates it with automating more economically valuable work, including research; he offers a broad view rather than a precise definition or timeline.

Looking to 2035, Hiremath imagines a global marketplace that helps create 100 billion jobs over time by connecting people with roles and tasks. This is an ambitious hypothetical derived from his assumptions about job seekers and multiple jobs per person, not a projection of verified placements or a guarantee. He says he would like Mercor to become public one day.