Almost Human

Shai Wininger Takes on AI-first Product Organizations Pt. 2

What happens when AI actually runs the company? In this episode, Shai Wininger (co-founder of Lemonade) lays out a concrete framework for AI-native organizations: where agents are the default operators, humans are summoned only when needed, and companies evolve into self-improving systems. We explore what it actually means for “humans to become tools,” how AI systems decide when to escalate to humans, and how uncertainty is measured in production environments. Shai breaks down Lemonade’s real-world approach—from claims automation to risk guardrails—and shares how they think about confidence scoring, hierarchical decision systems, and human-in-the-loop design. The conversation goes deeper into the future of product development: self-healing code, self-implementing systems, and eventually self-driving innovation – where agents generate, test, deploy, and kill features autonomously. We also cover why traditional metrics and benchmarks are breaking down, how to think about continuous improvement instead, and where founders are wasting time with AI today. Shai argues that software is rapidly commoditizing and that durable value is shifting toward hard problems: distribution, regulation, and real-world complexity. This is a tactical and forward-looking conversation about how companies will actually operate in an AI-first world. Please rate this episode 5 stars wherever you stream your podcasts!

Shai Wininger Takes on AI-first Product Organizations

MIT famously claims that 95% of AI projects fail, not because the models don’t work, but because organizations aren’t built for AI. In this episode of Almost Human, Shai Wininger, Co-Founder & President of Lemonade, explains how one of the most AI-native consumer companies rebuilt its product org, workflows, and technology stack to make AI work in production in one of the most regulated industries in the world. We unpack Lemonade’s internal LoCo platform (an LLM-first, no-code insurance application builder), why “engineers writing code” is being replaced by engineers writing text configuration, and how specs are evolving from static Google Docs into tests that define when an AI agent is done. Shai shares: * Why 1 engineer + AI tools can now replace traditional teams * How Lemonade iterates on pricing, underwriting, and claims with AI at scale * Why tests act as guardrails and reward functions for AI agents * How product specs, workflows, and artifacts are changing * What an AI-native product organization will look like 12 months from now * How to build AI systems that self-heal, self-improve, and eventually pursue business goals This episode is a tactical playbook for founders and product leaders who want AI to be a durable capability, not a perpetual experiment. Please rate this episode 5 stars wherever you stream your podcasts!

Assaf Elovic Takes on the SMB AI Revolution

What does “Head of AI” actually mean inside a company like Monday.com and why are SMBs such an insane AI opportunity? In this episode, Eden talks with Assaf Elovic, Head of AI at Monday.com, about bringing a massive SaaS company from “almost no AI” to AI products that ship real value across millions of users. Assaf shares how his role evolved from a 0→1 “startup inside Monday” to building an internal AI platform and culture the whole company can build on. They break down why AI is 10% model, 90% workflows and culture, and why SMBs don’t need more magical chatbots - they need AI that is transparent, constrained, and deeply tied to the way they already work. Assaf walks through Monday’s first failed AI co-pilot, why users didn’t know what to ask, and how they reversed course by killing the chat, embedding AI Blocks directly into boards, and then adding explainability and feedback to rebuild trust. From there, they zoom out to how Monday runs AI internally: AI Champions across departments, an “AI month” where everyone paused roadmaps to build AI, and a bottom-up culture where people share new papers and tools because they’re genuinely afraid to miss the next unlock. For founders, Assaf and Eden get tactical: * Why risk analysis with generic LLMs backfired and what it taught them about subjective risk and domain-specific models * How to think “business problem → research task → product,” instead of doing cool research and hoping it lands * A simple framework for interviewing SMBs and finding a first workflow worth automating * Why a two-person startup today should be obsessed with voice agents and narrow, high-value workflows If you’re building AI for real businesses, especially SMBs, this episode is a playbook on turning research into product, closing the trust loop with users, and using AI to actually do the work, not just manage it. Please rate this episode 5 stars wherever you stream your podcasts!

Zeev Farbman Takes on Real-Time Diffusion Models

What if video wasn’t a file you export, but a service that responds to every viewer in real time? In this episode, Eden Shochat talks with Lightricks co-founder & CEO Zeev Farbman about LTX 2, their real-time video diffusion model that runs on edge devices (phones and gaming GPUs) instead of giant cloud data centers. They break down: * Why “every pixel is programmable” is the real story behind diffusion models * How we got from Facetune to a real-time video engine * Why edge AI might beat cloud APIs for the next wave of products * China’s open-weight strategy vs the West’s closed, API-first approach * What happens when you can personalize every ad impression with generated video * Why foundation models are a fast-depreciating asset and where the real moats will be * How Lightricks thinks about being both a product company and a platform/API * The coming shift from blank 3D scenes to “never start from scratch” creative workflows * The sci-fi idea from Neal Stephenson’s Diamond Age that Zeev would build as a startup today If you’re a founder, engineer, or creator thinking about the future of video, gaming, or adtech, this episode is a playbook for what’s now possible when real-time, per-user video runs on everyday hardware. Please rate this episode 5 stars wherever you stream your podcasts!

Nir Hemed Takes on Semi-Autonomous Agents

Property management is a brutal business: low margins, endless paperwork, and human bottlenecks everywhere. So what happens when you give AI agents the keys? In this episode of Almost Human, Eden Shochat sits with Nir Hemed, co-founder and CTO of Daisy, a New York–based startup reimagining property management through AI. Nir reveals how Daisy built a team of ten autonomous agents - each with names, roles, and Slack access, and how one of them, “Steven,” now handles complex onboarding processes once run entirely by humans. Eden and Nir unpack what it means to design trust between humans and agents, how to build eval frameworks that become a company’s secret IP, and why the future of work isn’t just automation - it’s orchestration. From Temporal workflows to generative UIs and “AI coworkers,” this episode dives into the scaffolding era of AI: where humans teach machines how to manage, reason, and evolve. Key takeaway: The companies that win in this era won’t be the ones that collect the most data, they’ll be the ones that build the best feedback loops. Please rate this episode 5 stars wherever you stream your podcasts!