Netic's Agents Answer the Phone When the Heat Breaks

On *No Priors*, Elad Gil introduces Melisa Tokmak as the founder and CEO of Netic, a company that builds AI for real-world services — HVAC, pet care, roofing, and the rest of the work that keeps daily life running. Before Netic she was a director of engineering at Scale AI, where she also worked go-to-market, and she spent time at Meta. Gil, who is an investor in the company, wants to talk about what it means to put AI into the physical world rather than into another chatbot for other startups.

Netic, Tokmak says, builds AI to run millions of real-world businesses in essential services. The first customers are large enterprises: billion-dollar home-services firms in HVAC, plumbing, and electric; consumer wellness and hospitality companies; automotive; pet services. Netic sits between the company and its customers. Understanding the need, matching it to the business's operational rules, and deploying the service or the labor all happen on the platform. If you call, text, or use the online scheduler of a Netic customer, it is Netic's agents that talk to you.

She asks you to imagine minus twenty, heat dead, a child or an elderly parent in the house. You pick the provider you trust and can actually get. On the other end, a voice agent asks what kind of home you live in, whether the company has your records, and whether someone should come today or tomorrow. That decision is not "Elon's heat broke, send Melisa." It depends on the unit, the need, whether the company can service it, lifetime value, and whether the one free technician today only knows boilers. Every step is optimized for customer delight and for revenue.

Until Netic, the incumbent was people — hundreds of them — because these are EBITDA businesses, often private-equity owned, that cannot grow without hiring. The labor that actually matters is the blue-collar tech who does the job. Support teams are unreliable. Tokmak walks through a day at a billion-dollar firm: the business starts at 4 or 5 a.m., technicians at 6, and before that the office is empty. A manager arrives to find three people quit and five did not show. Customers pile in at 6 because the country just set a heat-wave record. Many of these businesses are cyclical; winter, summer, and holidays have to pay for the year. Early customers used Netic for overflow. Today more than 70 percent are "Netic first": the end customer's first interaction with the company is an agent.

Gil has also invested in AI roll-ups — firms such as Long Lake that buy traditional companies and optimize them with AI — and the target profiles overlap with Netic's customers, many of them already PE-owned. Why build a service instead of buying the assets? Tokmak gives three reasons: what she wants to build, skill set, and scale. At Scale she spent about four years standing up real-world businesses, including the government unit and large accounts in logistics, manufacturing, financial services, and healthcare. Gil notes she left before Scale's Meta deal, which he puts around $28 billion and she rounds to a little 30.

Scale showed her how much tangible impact those industries hold. She grew up with nothing in a small town in Turkey, arrived in the United States only for college, and did not own a computer before a full scholarship to Stanford. Her whole town works in businesses like the ones Netic serves. Tech talks constantly about impact, she says, and then ships products that mostly serve other startups. She wanted the opposite, fused with a technical bet: copilots for consumers are solved enough; the open problem is AI inside mission-critical, fully executing workflows.

She loved Scale, fit the culture, still counts friends from there, and calls Alexandr Wang an amazing founder to work under. The job was missing product. Building infrastructure for AI is operationally heavy, even if people do not like to say so, and she wanted something that could scale and compound. Roll-ups, in her view, live or die on M&A. She is an engineer and a product person, not an M&A person, and she does not want a company where her contribution is not the main event. Products built for the companies you just bought rarely transfer. She is interested in the other question: if owners could focus on labor and service quality, how could Netic run the rest?

Gil asks whether Netic, robotics, and self-driving converge, or whether the robotics timeline is so far out it does not matter. Tokmak treats them as chapters of the same book. There is a robotics chapter in the future. In the industries she serves, it is decades away. Every Netic engineer has to visit customers on site before writing product in San Francisco — that is how she learned the work — and if you look out a window at the buildings, robotics would need every one of them 3D-printed or fully standardized. She does not see that happening.

Today's robots still struggle with dexterity on different screws, desk heights, and the tiny spaces where a repair actually lives. Often you have to open the walls before you know what is broken. On top of that, these jobs carry a human element. People call on the worst day of their lives — a flooded house — or they want to drop the stress and play tennis at the Bay Club. Either way, they are going through something. Robotics, she thinks, is a long way from that chapter.

Can the labs just do this — OpenAI, Anthropic, Google, Meta? She chuckles. Ten years ago the question was whether Google would do it. Some core competencies they could take on; they are not investing in the rest. For Netic she does not see them as a competitive risk. The leading labs are impressive businesses and good partners. The startups actually in the blast radius are the ones whose products look like a lab demo: a chatbot or a coding agent. Enterprises in her industries will not bet critical workflows on OpenAI's habit of shipping and killing products quickly. Silicon Valley credits Anthropic's focus for Claude; she sees the opposite in its enterprise line, something like twenty products and no clear center.

Researchers want the most general solution, which in this domain becomes "when we have AGI, we will ask it how to do essential services." She thinks that is operationally and intellectually lazy. The hard problem is the last mile: millions of people, different worries, accents, contexts, and the work of making them come back. That does not come from models alone. It comes from harnesses, orchestration, software, and product on top. Companies like Netic have to be good at all three layers.

Gil's side comment is that founders today flinch more than they used to. Four years ago he was funding Harvey, Perplexity, later Decagon and Abridge — vertical applications, Netic included. Now many people stay out of a vertical because they think it sits on a lab roadmap. Tokmak thinks a lot of current building is aimed at an immediate exit. Being a founder, she says, used to feel more honorable: you knew you were dedicating your life. Before Netic she actually looked for a job, asked whether there was anything else she wanted to build, and whether there was someone whose mission she wanted to further. Founding is decades, not something to take lightly.

The people she wants at Netic are customer-obsessed for the long term, with agency and urgency plus the rigor and patience to carry work forward. In hiring, especially among Gen Z, she keeps meeting a "permanent underclass" mindset: if I do not make my money in eighteen months, or learn everything in six, I am forever poor. Gil translates it as an AGI-pilled story in which skill value collapses. She calls it dangerous. Building good things takes a long time. The important lessons in her Scale years, and now in building a company, came from staying with a thing past the first version. She avoids shiny-object seekers.

A quote she attributes to Martin Luther has become her founder ethic: the Christian shoemaker does not honor God by putting little crosses on the shoes, but by making the best shoe. Craft is the reverence. That is what she wants the company to be — the best product for people she cares about, more done with less, a small group trying to take down a large problem.

How do you screen for agency? Not a single heroic weekend. If someone has agency, it shows up continuously. For a new grad she will dig into college, how they got there, a project they kept. The question she likes is what the hardest thing they have ever done is, then she pulls the story apart. Follow-through is the point. Sarah Guo says she used to ask the same question and got such bad answers she stopped. Tokmak's reply is dry: maybe that person is not a fit. She is not looking for a work story. A recent hire said the hardest thing was a simple life — work, a strict health regimen, the same hour every day — kept for more than fifteen years without getting bored. Tokmak's own answer is half company-building and half leaving a tiny town where her parents had nothing. Everybody's story is different. She wants to know how you wrote yours.

The five-year north star is an autonomous enterprise: everything in these companies handled by Netic except the actual service and labor, so owners can put all their energy into customer delight. Every product on the intelligence layer is supposed to connect and compound toward that. The second goal is to keep working only with remarkable people as the company grows. Palantir, SpaceX, and Notion — firms that cared about craft for years — have all struggled with talent density. She wants Netic to take that problem seriously.

Gil asks whether selling agents into industries people call slow has been hard. Tokmak calls the old-school label a misconception. Some of the most tech-forward, business-focused owners she has met are in these trades. They are large enterprises and extremely value-focused, which forces them to be tech-forward. One deal was a half-million-dollar contract that went from first contact to close in fourteen days — not because she or AI is a magic potion, but because the buyer checked for value and then moved. She describes the industries as extremely primal and tech-forward at the same time.

A large roofing company still sends door-knockers through neighborhoods to talk about reroofs and solar. The same company now runs inbound, outbound, analytics, and labor delivery on Netic. The platform pulls satellite data on how hurricanes hit different roofs and which materials to use, then feeds that context to agents so they handle the inbound call and also spot which blocks to knock. The interest in that data is not new; people used to sift it by hand across tools. One platform now assists the ad-driven inbound and reaches out for net-new revenue. Zoom out to a PE owner with twenty or thirty companies on Netic, and the context travels: a dog lover shopping a wellness membership can be steered toward the portfolio company that also does pet care.

The PE playbook itself has changed, she says. The old move was find an undiscovered gem, swap the team, sell the multiple. Those gems are gone, so the work is creating tangible value. Some funds lean in too hard, trying every AI product as if week one were a software trial. She wants them to judge the year: results should exist across the calendar and keep getting better, not worse. That is how you separate vaporware from ROI. Funds now hire AI operating partners or a few AI-forward engineers to educate the portfolio. Netic does not lead with demos. They pull up a live deployment. Gil asked for her north star; she says Netic has generated over $600 million for customers from AI-handled interactions. First PE conversations still default to cost cutting. She keeps turning them toward net-new revenue. "It would be pretty sad if we use AI only for cost cutting."

Outside Netic, what excites her about AI is education. When she had no access to anything, the technology that helped was finding one or two people from her high school who had made it to American colleges, then Facebook-Messaging them to read essays. She thinks education will change the lives of people who would not otherwise know this world exists. Everything you want to learn is in your pocket. The world has been made more about agency, and she cannot wait to see that unfold.

Gil's lesson is that you can just go do things; you do not need permission. The dystopian version is watching people still not do them. Tokmak agrees it stays a choice. We keep making the choice easier, and she still guarantees that the majority of the world will not make it. At least resourcing — whether you have access — is being taken off the table.

The mainstream American conversation about AI, she says, is very negative. The use cases she works on are often the worst day of someone's life, when they really need help. That is also true of education, and of being able to read your own health records. She tried a health feature to understand how to sleep better and got no useful answer — the joke being you have to stop being a founder. What she wants is a public conversation in which a person's first interaction with AI is not about jobs going away, but about help arriving when the pipe bursts or the lab results make no sense.