Ali Ansari is the founder and CEO of micro1, a company that provides expert human data to AI labs for training and improving models. The episode covers micro1’s evolution from an AI recruiting tool into a full-stack human data platform, its philosophy around expert happiness and incentive design, and how the company grew revenue roughly 30x in a single year (from ~$4–5M to ~$150M run rate) while staying lean and profitable.
From recruiting tool to human data platform
Ali started micro1 while at Berkeley, where he ran a software development agency and built an AI screener to vet engineers for projects. That internal tool became the first version of micro1: an AI-powered recruitment engine with a marketplace of pre-vetted engineers and product people.
A data provider became a micro1 customer and hired 700 engineers in three weeks. Ali and his CRO realized this was for training AI models on coding, and that the human data space was the best application of what they’d built.
Ali doesn’t call it a pivot so much as an evolution: the core AI recruiting product was already well-suited to the bottleneck AI labs faced, which was recruiting experts at scale. The company went all-in on the human data space about a year and a half before this interview.
The product now spans the full data pipeline: AI recruitment, a modular data platform, performance management tools, and RL environments. The AI recruiter remains the most important piece.
The human happiness index
micro1 tracks what it calls a “Happiness Index” for the experts on its platform, which Ali says is the number one focus. Happy experts produce better data, which serves customers better.
Experts fill out a form called “Are You Happy” rating their experience 1–5 across multiple dimensions, plus qualitative questions. Project leads use the results to take action (e.g., increasing pay, adding more human data managers).
micro1 is building the “M1 Happiness Model” that predicts how happy an expert will be on a given job and adjusts the AI match score accordingly. The goal is to match people not just on skills but on job satisfaction.
Ali frames this as a broader mission: micro1 is helping create a new job sector of AI trainers (roughly 100,000+ experts globally), and it should be a great one.
Robotics and real-world data
micro1 is building a robotics data vertical. The core challenge is that there’s no “internet” for robotics models to train on, so micro1 is creating one.
The company has ~3,000 people in 50 countries recording egocentric (head-mounted camera) videos of themselves doing everyday household tasks. These are annotated for vision-action models and designed to be mappable to any robotic system.
The focus is on maximizing diversity by capturing people in their natural state doing tasks they’d do anyway. A second category—industrial and manufacturing data—is planned but not yet active.
Ali gives the example of a chef making a dish repeatedly: that kind of structured, repeated real-world task data is what robotics labs need.
Operational intensity and white-glove service
The business is deeply operational. Customers won’t just log into a product; micro1 provides a white-glove service. The limiting factor is hiring exceptional “strategic project leads” (SPLs) who manage pipelines end to end.
If an SPL hire is bad, it materially affects the customer relationship. The company’s products (recruitment engine, data platform, performance management) are designed to reduce dependency on any single hire and enable scalability.
Ali describes hiring hundreds of doctors or lawyers in a week—world-class surgeons in specific countries who speak specific languages. This is only possible because of micro1’s AI recruiting agent (Zara), which can vet niche capabilities at scale. Without it, it’s a chicken-and-egg problem.
Hiring philosophy: agency, risk, and judgment
Ali looks for three things in hires: agency (will they care and take initiative), risk tolerance (will they take bold moves and recover from mistakes), and judgment (can they distinguish between two-way and one-way door decisions).
He tells the story of an SPL who promised an impossible timeline to a major customer. The team tried anyway, recovered, and the customer’s trust actually increased. That SPL now runs the account and was named team member of the month.
Ali references Jeff Bezos’s two-way vs. one-way door framework: high velocity on reversible decisions, much lower velocity on irreversible ones. micro1 tries to limit bureaucracy so team members can get quick opinions on one-way decisions.
Incentive design: long-term equity and short-term outlier bonuses
Ali spends significant time on incentive alignment, which he considers one of his three core responsibilities alongside hiring and product.
Long-term incentives are aligned through equity. As a sole founder, Ali can give meaningful equity to early and later hires alike.
Short-term incentives matter too, because in a space growing this fast, three months can double the run rate. Examples: recruiters get “absurd bonuses” for hiring 1,000 people in two weeks; a key deal lead might be told “if this closes, you’ll double your equity.”
The doubled equity vests on the same schedule (and the vesting restarts for the top-up), so it’s a short-term motivator structured as a long-term instrument. Ali says he hadn’t heard of anyone else doing this until he found a similar approach attributed to Jensen Huang at Nvidia.
Culture: sustainably hardcore
The team works 13–14 hour days, with leadership working even more. Ali’s goal is to reach a state of being “sustainably hardcore.”
micro1 does not enforce working weekends. The theory is that inspired work outperforms forced work—and in practice, the team works more weekends than most because leadership models it and the incentives are in place.
Ali believes the founder’s hours set the ceiling for the company. He stays in the details (CFO runs payroll, GC handles all legal work herself) because being too abstract means losing the context needed to unblock day-to-day decisions.
Losing their biggest customer
A few years ago, a customer that represented nearly 50% of revenue was lost due to mostly external factors. Ali found out via email in the elevator on his way up to pitch investors during an early fundraise.
The company had to lay off people and came close to missing payroll. Ali walked around Palo Alto in shock, leaving his laptop on the sidewalk.
In the immediate aftermath, Ali became risk-averse for a few weeks, questioning whether he should hire more “professional” managers. He spent two days consciously deciding to return to his risk-taking mode, concluding that the founder’s job is to inject as much upside risk as possible into the company because no one else will.
He reframed risk assessment: the worst-case scenario of almost any business decision is bankruptcy, so focusing on worst-case isn’t useful. Instead, he thinks about the probability distribution and expected value.
The experience also made him more focused. He references Brian Chesky’s approach at Airbnb post-COVID: one product roadmap that the CEO personally approves every module for. Ali has adopted a similar discipline.
Intuition over KPIs
Ali describes himself as not particularly analytical, despite what newer team members assume. He resists reducing people’s work to a few quarterly KPIs, arguing that for a fast-growing company, KPIs become stale within a week and optimizing for them can lead people away from what actually matters.
For revenue-adjacent roles, micro1 uses a “revenue override”: if the company hits its ambitious quarterly goals, individual KPIs don’t matter.
Ali thinks of intuition as a large neural net considering many variables at once, while a few KPIs drastically reduce the feature set. He acknowledges KPIs will be necessary long-term as roles specialize, but for now, qualitative assessment works better.
30x revenue growth in a year
In 2025, micro1 grew from roughly $4–5M to ~$150M run rate. Ali describes the experience as intensely stressful and grateful.
His day-to-day changes every few months. He constantly questions whether he’s spending time on the right things. Threads he reads on Slack for 10 minutes often already have good decisions made by the time he catches up.
He still acts as an account executive because the customer base is small and the relationships are with exceptional researchers. Sales is non-transactional: dinners where nothing is pitched. Customers sometimes message afterward asking why micro1 didn’t try to sell them anything, and that approach drives expansion.
His current focus areas: product roadmap, fundraising, customer calls, incentive alignment, and staying in the weeds with the team.
Staying focused vs. experimenting
Ali admits he’s prone to starting side experiments (his CMO pushes back on this). The company tries to limit the number of new engineering teams and subdomains, keeping everything on one platform where possible.
When entering new verticals, Ali relies on intuition. He guessed 4–5 months ago that egocentric human demonstration data would be key for robotics. When Physical Intelligence published a paper confirming this, micro1 scaled the pipeline from 100 to 3,000 people immediately.
Not all proactive pipelines work out—some data becomes stale. But the ones that hit can be transformative.
Lean hiring
micro1 grew its internal team from ~35 to ~80 people in 2025, far outpaced by revenue growth. The philosophy is to hire only as a last resort.
Before hiring, the company first questions whether the requirement can be deleted entirely (e.g., automating a contract-signing process that was consuming a team’s bandwidth). Then it looks at whether existing team members are constrained by coordination overhead rather than the core function.
In late 2025, Ali loosened the hiring constraint slightly, adding more people to sustain 20–30% month-over-month growth. As long as headcount growth isn’t proportional to revenue growth, he considers it disciplined.
Profitability and the case for raising
micro1 has been profitable for most of 2025 and is net profitable historically—it hasn’t touched its raised capital and has added to it.
Despite profitability, the company plans to raise. The reason: proactively building pipelines requires high cost bases. A pipeline that could generate tens of millions in revenue needs tens of millions in upfront spend on data creation.
Ali sees the cash cushion as enabling bold bets. If the human data market doesn’t materialize as expected, the company has time to apply its product to something else.
Why human data will be a trillion-dollar market
There is no “last mile” in AI. As current functions get automated, humans will create new, more creative functions, which will in turn need human judgment to automate. This loop continues indefinitely. Ali’s example: micro1’s recruiters no longer do interviews (the agent does), so they’ve invented creative sourcing strategies that are essentially marketing campaigns—a net new function.
Compute buildouts demand new capabilities. Labs and companies are spending hundreds of billions to trillions on compute. For that investment to pay off, models need vastly more capabilities unlocked, which requires structured human judgment in each domain.
Synthetic data makes human data more valuable, not less. Synthetic data extrapolates human judgment by orders of magnitude (1000x, potentially 1,000,000x). If models need fewer human data points because synthetic data is better, each human data point becomes more valuable, increasing total spend.
A percentage of the entire labor market will become human data creation. Ali notes that lawyers at micro1 are paid ~20% more than at their law firms because the structured judgment they create is more valuable than unstructured legal work. If even 5% of the ~$50 trillion global labor market becomes human data spend, that’s $2.5 trillion/year—and discounting heavily still gets to $1 trillion/year.
Long-horizon tasks
Models are good at answering individual questions (~90% accuracy) but struggle with multi-step tasks because errors compound: 0.9^20 ≈ 0.12. A 20-step task done at 90% per-step accuracy succeeds only ~12% of the time.
micro1 is building RL environments around long-horizon tasks that simulate end-to-end workflows. Example: filing someone’s taxes involves gathering incomplete information through multiple rounds of back-and-forth, having a conversation about tax optimization, and then filling out the final form. Each intermediate state needs partial rewards.
This is why “the year of agents” (2025) didn’t fully materialize: foundational models still struggle with long-horizon tasks, and enterprises haven’t yet built contextual evaluations into their workflows. Ali believes agent adoption will accelerate when enterprises treat evaluation as core to engineering.
Personal drive
The hardest thing Ali has overcome is retrospective: realizing how much his parents gave up to leave Iran and restart their life in the US. They lived in a single bedroom as a family of four and struggled for years.
Ali’s drive comes largely from wanting to make their sacrifice worthwhile—to give them a great outcome for the move and contribute to that through what he builds.
