AI Infra Nights x Autonomous Engineering: SF Edition Sep 24
AI Infrastructure Nights and Autnomous Engineering are coming to SF! We're bringing together founders, engineers, and FDEs building the systems underneath production AI.
Join us for three focused discussions on the hard infrastructure problems behind secure, reliable AI products with the people solving them in practice every single day.
🎟️ Space is limited, tokens are inifinte.
Speakers
Adam Gold, Founder & CPO, Islo
When software factories don't work
We wanted to build a software factory - so we built agents that push PRs... but that wasn't enough as no one could trust the code. We then continued our journey in improving the factory - and learned what doesn't work along the way.
Now that we have fully autonomous agents - they implement, verify the fix actually works, review the change in a live environment, and publish a PR preview, omnipresent - we want to talk about the pitfalls and how to avoid them.
In this session we'll walk through where we started, what broke along the way, and the infrastructure that makes it possible.
Dhruv Gupta, Software Engineer, Databricks
Omnigent: A Meta Harness for AI Agents
Omnigent is an open-source meta-harness that allows developers to use multiple coding agent harnesses of their choice together through a single UI. In this talk, we will share the merits and features of using a meta-harness: how to orchestrate different harnesses, collaborate effectively in the era of vibe-coding, and exercise control while doing so. We will walk through a demo showcasing its developer-friendly workflow and features, including setting up coding workflows that use multiple harnesses in a governed way
Alon Gubkin, Founder & CEO, Alien
Giving Customers a Cryptographic Kill Switch for Their SaaS Data
Some enterprises want more than “we encrypt your data”—they want the ability to revoke access without trusting the vendor to cooperate. This talk explores customer-managed keys, envelope encryption, per-tenant key hierarchies, key rotation, revocation, caching, and what actually breaks when a key disappears.
Devanshi Vyas, Cofounder/CTO, headroom labs
AI agents spend an increasing share of their context budget rereading verbose tool outputs, logs, files, and conversation history. This talk explores how Headroom, a popular Github repository(72K+ GitHub stars), compresses that context locally while preserving the information agents need to solve problems. We’ll dive into content-aware compression, prompt-cache stability, and CCR, Headroom’s Compress-Cache-Retrieve architecture for recovering original data on demand.
We will walk through a demo showcasing its developer-friendly workflow and features, and see how we can start saving cost today.