Hetz Ventures

Everyone's racing to pick the right model, but the teams actually winning in production? They obsess over something else entirely. Guy Fighel made the case at LangTalks that across every layer of AI (frontier models, open weights, APIs, frameworks) what separates the solutions that win comes down to one thing: context. Context as a discipline, not a buzzword. The way Guy frames it: AI models are statistical, not deterministic. They don't carry the knowledge that lives inside your organization or your users' heads. The teams that figure out how to bridge that gap of how to bring model output to the level of knowledge a human expert would have, are the ones seeing real adoption and real results. Everything else is implementation detail.

Great energy at the Hetz office yesterday as we proudly hosted YWVC 🚀 Three fabulous investors brought it, TED style: Aya Neeman (UpWest) - 'The art of being a vulnerable investor' Zemer Shwartz-Noy (ex-Blumberg Capital) - 'VC to startup to corporate' Tamar Shoshan (The Garage) - 'Behind the scenes of the pitch'. Thank you to Ziv Idan (Citi Ventures) and Roni Kahn (The Garage) for building and leading this community of Israeli women in venture. And big thanks to Chen Zvi for hosting and representing from our side 🙌

The clearest signal that agentic commerce is real: an entire layer of new protocols, built on the assumption that no human is watching. 👉 MCP for agent-to-tool communication. 👉 UCP for commerce interoperability, with a Tech Council that now spans Google, Amazon, Meta, Microsoft, Salesforce, and Stripe. 👉 AP2 for agent payments, backed by 100+ partners and a "Human-Not-Present" mode where you approve the policy and the agent transacts within it. The internet was built for humans. The next era is being built for everything else acting on their behalf. Guy Fighel on what this means for anyone building: https://lnkd.in/d5bscv6G

Hiring in TLV 📣: Fundamental just announced its engineering hub and is hiring across infrastructure and applied AI. NEXUS, their Large Tabular Model, is built for the structured data that runs the enterprise: the rows and columns where trillions of $$ in decisions actually live. Fortune 100 customers, AWS, SAP, already on board. Looking forward to top tier innovation, Gabriel Suissa, Jeremy Fraenkel. We're proud to be investors. Fundamental is hosting a technical evening at co-investor Battery Ventures in TLV - RSVP: https://luma.com/5esok18x

It's unusual - but beneficial - for two Israeli cyber funds to co-lead the same seed round. That was Pavel Livshiz's observation at Cybertech in January. We've now done it twice: Opti with YL Ventures, BAND with Team8. These kinds of rounds give founders two locally rooted partners who know the Israeli cyber playbook from the inside, and set up a cleaner Series A conversation with US leads.

Two lines that shouldn't be moving in opposite directions. Cost to produce an AI token has fallen ~99% in three years. Value per token keeps climbing. For almost any other commodity, that gap closes fast. Tokens are different and the 100s of billions in capex coming out of Microsoft, Google, Amazon, and Meta is proof. The race isn't about whether AI works anymore... It's about who owns the infrastructure underneath the commodity everything else runs on. Judah Taub unpacks the gold rush economics in his latest (link in comments).

"Stop planning for your massive hire P&L." Guy Fighel argues that the seed-stage budget most founders are still pitching looks like it was written in 2016: a long line for headcount, a polite line for SaaS, and AI tooling tucked in near office snacks. The founders he's watching most closely are running three or four humans alongside a layered set of agents handling work that a year ago would have required a team of fifteen. This isn't a cost-cutting exercise, it's a different theory of the company. The full piece, including the three questions Guy uses to route work between Cowork, Routines, and Managed Agents, is on the blog (link in comments).

Congrats to Majestic Labs ai on making the CB Insights AI 100. Prometheus tackles the problem most of the AI stack is about to run into: memory, not compute, is what's bottlenecking the next generation of models. Ofer Shacham and the team saw it early and built for it. We're proud to be early backers of a category-defining company in a category-defining year.

LangTalks 2026 is a wrap, and the AI engineering scene in Israel showed up in force. What the day made clear: in the year since the last LangTalks, the AI engineering ecosystem here has grown up. The people writing the evals, debugging the agents, owning the data pipelines, leading the teams - and starting the companies. It's why we wanted to help bring this day together. We were proud to co-produce the day along with Langtalks founders Lee Twito and Gal Peretz. Thanks to everyone who came through, spoke, and stuck around for the hallway conversations. More to come...

Live from LangTalks 🦜 🎙️ : we're surrounded by 1k AI engineers, CTOs, startup founders, discussing and demoing what it takes to build with AI. Guy Fighel with a peek into Langtalks AI Engineering Conference 2026: Lee Twito Gal Peretz TV10 Israel