Interview, Explained
In-depth breakdowns of tech interviews — section-by-section analysis with diagrams, quotes, and insights.
Melisa Tokmak on automating everything except the labor
Netic founder Melisa Tokmak is building an autonomous enterprise for essential-services businesses — HVAC, roofing, pet care — where over 70% of the businesses it serves now go 'Netic-first,' betting that the moat is the last-mile orchestration and proven ROI that raw models skip.
Jesse Zhang & Ashwin Sreenivas on Why the Moat Isn't the Model
Decagon's founders argue that in the agent era the durable moat isn't the model but the software and process around it — fine-tuned 'model factories,' encoded business logic, and the machinery that makes frontier capability deployable inside the enterprise.
Jeff Dean on why inference hardware is AI's next bottleneck
At Y Combinator's Startup School 2026, Jeff Dean argues that as models commoditize the writing of code, the leverage moves down to the systems layer — specialized low-latency inference hardware and the energy cost of moving data — and up to the human skill of taste in choosing what to build.
Sam Altman on why intelligence on tap raises the bar
Altman argues AI agents compress startup execution from months to minutes, raising the ambition bar while making the concentration of power the central safety risk to design against.
Eric Landau on why physical AI is won at the data layer
Encord's Eric Landau explains why physical AI — roughly 80% of the economy — will be won not by cleverer models but by whoever can collect, curate, and evaluate multimodal data at petabyte scale.
Akshay Nathan on One Shared Harness, Two UX Shells
OpenAI merged Codex's agent power into ChatGPT Work behind one shared harness after non-developers quietly adopted the coding agent, betting that as AI blurs job roles the product should route users rather than box them in.
Fei-Fei Li & Yunzhu Li on Why Robots Learn in Simulated Worlds
World Labs and SceniX argue the path to robots that work runs through simulated worlds: because real-world robot data is scarce, a real-to-sim-to-real loop manufactures the counterfactual coverage, reliability, and fast evaluation that real data alone can't.
Damian Borth on Training Models From Models, Not Data
St. Gallen's Damian Borth argues the weights of already-trained neural networks are a new data modality you can learn from and generate — letting you train new models from old models instead of scarce data, and one day sample a network on demand.
Jay V on Why OpenCode Bets the Whole Field, Not One Model
OpenCode's Jay V explains how a model-neutral, open-source coding agent reached ~13M users and ~7T tokens a day by betting the entire open-weight field and serving the world the frontier labs price out.
Boris Cherny on treating the model like a living creature
Boris Cherny, creator of Claude Code, argues that building on frontier models is an empirical craft: delete your scaffolding every generation, unhobble capability the product is hiding, and let the model verify its own work while it runs for days.
Travis Kalanick on why every industry is a computer
Travis Kalanick lays out the framework behind his stealth comeback, Atoms: every physical industry is a computer whose three resources are manufacturing (the CPU), real estate (the storage), and transport (the network), and he is building those computers one industry at a time, starting with food, mining, and freight.
Jensen Huang on why learning is the greatest superpower
At YC Startup School 2026, Jensen Huang argues that Nvidia's founding technology was flat wrong, that the durable edge across 34 years and $5 trillion was never any chip but the willingness to confront what he didn't know and learn it fast.