Gian Segato works on research at Anthropic. He was early at Replit, where he led growth and helped build coding agents, after founding the acquired education app Uniwhere. — Gian Segato.

Part 1: Building in the Probabilistic Era
- Managing uncertainty: Segato argues that AI-product builders should choose an acceptable level of unpredictability for their market rather than demand perfect reliability. — Building AI Products In The Probabilistic Era.
- Data as a lever: In AI products, Segato puts data both upstream and downstream of the model at the center of understanding whether users succeed. — Building AI Products In The Probabilistic Era.
- From funnels to fields: Because AI products accept open-ended inputs and can yield unforeseen outcomes, Segato says builders must move beyond fixed conversion funnels toward a wider field of possible user paths. — Building AI Products In The Probabilistic Era.
- The limits of fixed tests: A small, hand-picked test suite cannot cover an open-ended AI input space; Segato recommends keeping evaluations close to real user behavior. — Building AI Products In The Probabilistic Era.
- Statistical software: Segato contrasts deterministic software, which maps known inputs to expected outputs, with AI systems whose outputs are sampled from distributions. — Building AI Products In The Probabilistic Era.
- Beyond deterministic dashboards: Segato warns that conventional dashboards can miss how people explore emergent AI features; he favors analyzing user trajectories and outcomes. — Building AI Products In The Probabilistic Era.
- Latent capabilities: Segato says changing a model or prompt at Replit revealed unplanned use cases, including game development, which evaluations need to catch. — Building AI Products In The Probabilistic Era.
- Measure trajectories: Segato proposes grouping users by task and tracking milestones along their journeys, rather than relying only on static funnel events. — Building AI Products In The Probabilistic Era.
- Sample real cases: To test AI products well, Segato calls for continually sampling relevant cases from actual usage as model behavior changes. — Building AI Products In The Probabilistic Era.
Part 2: Agency and the New Economic Reality
- Human agency: Segato defines agency as a willingness to make things happen without waiting for instruction, validation or permission. — Agency is Eating the World.
- Agency as a dividing line: Segato argues that as AI lowers some barriers to specialized work, initiative becomes more important than credentials alone. — Agency is Eating the World.
- AI amplifies initiative: Segato describes AI as expanding what a determined individual can do, while acknowledging that mistakes and risk still matter. — Agency is Eating the World.
- Smaller teams, greater leverage: Segato points to solo founders and very small teams using AI to pursue businesses that previously required many specialists; he presents the one-person billion-dollar company as a possibility, not an inevitability. — Agency is Eating the World.
- Institutional resistance: Segato expects credential-focused institutions and middle management to resist a shift that values outcomes over formal specialization or headcount. — Agency is Eating the World.
- The chaos cost of solo work: He cautions that one-person companies lack team redundancy, so small AI-assisted mistakes can cascade into operational problems. — Agency is Eating the World.
- Leverage rather than org charts: Segato suggests high-agency individuals can carry work once spread across several teams, challenging headcount as a proxy for impact. — Agency is Eating the World.
- Human versus software agency: He distinguishes human initiative from AI tools that may be capable but are designed to respond predictably to user instructions. — Agency is Eating the World.
Part 3: The Challenges of EdTech and Scaling
- Geographical fragmentation: Segato argues that education demand differs by region in curriculum, culture and law, complicating venture-style expansion. — Why Ed-tech Startups Don't Scale.
- The localization trade-off: He describes the choice between adapting an edtech product for new markets and remaining in a region where it already fits users well. — Why Ed-tech Startups Don't Scale.
- Buyer and learner incentives: Segato warns that products bought by parents or institutions can be optimized for purchasers rather than the students who use them. — Why Ed-tech Startups Don't Scale.
- Learning takes effort: In Segato’s view, learning that builds durable skills requires effort, creating tension with engagement mechanics that remove friction. — Why Ed-tech Startups Don't Scale.
- Edutainment trade-off: Segato argues that entertainment-oriented products can monetize education demand, but pleasant consumption is not the same as durable skill acquisition. — Why Ed-tech Startups Don't Scale.
- The content-monetization trap: Segato says selling reskilling content to employers is difficult: large firms may create their own materials, while smaller potential buyers have limited budgets. — Why Ed-tech Startups Don't Scale.
- Product-market fit can be local: Segato says Uniwhere worked well in Italy but its model did not transfer easily to Europe and North America. — Why Ed-tech Startups Don't Scale.
- Growth can stall after local fit: Segato uses Uniwhere to illustrate that strong local fit need not lead to venture-scale international growth. — Why Ed-tech Startups Don't Scale.
Part 4: Growth Engineering and Product Strategy
- Growth engineering as a hybrid: Segato describes growth engineering at Replit as a combination of coding, marketing and product work. — From Edtech to Dev Tools.
- A shift toward product-led growth: Segato says his Replit group shifted from a marketing-led motion to a product-led team that tested its own hypotheses; his account does not establish that this shift caused Replit’s later revenue growth. — Embracing Risk and Growth.
- Support product engineers: Segato says the cross-functional growth team helped product engineers scope MVPs, choose lead metrics, go to market and experiment. — Embracing Risk and Growth.
- Agent-assisted software creation: Segato describes Replit’s coding agent as writing, running and deploying software in response to users, while noting that such tools remain instruction-responsive. — Agency is Eating the World.
- Approachable complexity: Segato argues for bundling development tools while letting complexity emerge gradually, so beginners can start without losing a path to advanced work. — Replit: the Bundling of Software Development.
- Ship and learn quickly: In his account of Replit, Segato says visible shipping, self-direction and experimentation shaped the way its growth team worked. — Embracing Risk and Growth.
- Education as an institutional entry point: Replit reported roughly 1,000 paying Teams for Education customers during its beta, showing meaningful adoption by schools and other education organizations; the source does not establish that this adoption was organic. — Replit Teams for Education.
- Bundle the workflow: Segato uses a small gym-data scraper to show how Replit combined editor, server, dependencies and persistent storage in one environment. — Replit: the Bundling of Software Development.
- Test growth hypotheses: Segato says Replit’s product-led growth team was formed to test its own hypotheses and later supported metric-led experiments across product work. — Embracing Risk and Growth.
- Lower the barrier to making software: Segato describes non-programmers building practical tools with AI, shifting more of the work from implementation toward expressing the desired outcome. — Agency is Eating the World.
Part 5: Career Evolution and The Future of Startups
- Technical risk in the next startup era: Segato argues that future outlier companies may take more technical risk, particularly in infrastructure, full-stack AI and physical-world applications. — The Dawn of a New Startup Era.
- A bet on developer tools: When he joined Replit, Segato said he expected developer tools to be a booming category in the following decade; this was his forecast, not an established outcome. — From Edtech to Dev Tools.
- Trade-offs of remote development: Segato describes remote IDEs as offering hardware flexibility while raising privacy, control and user-experience questions. — What's the Future of IDEs?.
- From edtech to developer tools: Segato saw Replit as combining learning and building, making his move from founding Uniwhere to growth engineering there a continuation of his earlier work. — From Edtech to Dev Tools.
- Capital-efficient possibilities: Segato points to small AI-enabled teams generating substantial revenue as an emerging pattern, while cautioning that solo operations can be fragile. — Agency is Eating the World.
- Use your own workflow as a test: Segato’s gym-scraper project exposed the friction of stitching many development tools together and helped him recognize the appeal of Replit’s bundled approach. — Replit: the Bundling of Software Development.
- Intent over syntax for some builders: Segato argues that AI tools let people without programming credentials create software by describing what they want, though reliability remains uneven. — Agency is Eating the World.
- A connected career transition: Segato connects building Uniwhere with his later Replit growth-engineering role: both involved helping people learn, build and use software. — From Edtech to Dev Tools.