Join us for another Agentic + AI Coding Night on Thursday, May 7 from 3:30pm - 9:00pm PST in SF for a multi-track deep dive into the architecture of autonomous coding agents. This isn't about simple chat interfaces, it's about building, serving, and monitoring agents that actually ship code.

This night is engineered for backend/infra engineers, AI engineers, and founders who are moving beyond basic wrappers and into the world of complex agentic workflows, repository-level reasoning, and high-scale execution.

You’ll hear from the following leaders in the AI space: Databricks, Anthropic, Open AI, Datafold, 1Password and more!

Topic Overview

Productivity at what cost? AI coding agents are supercharging developer output — but every line of agent-generated code that ships without oversight is a liability waiting to surface. This event confronts the gap between individual velocity and enterprise accountability.
The governance gap is real. As agentic tools proliferate across engineering orgs, the absence of guardrails isn't just a security concern — it's a trust problem. Learn how leading teams are drawing the line between empowering developers and protecting the business.
Agents don't know what they don't know. From hallucinated dependencies to overconfident completions, AI coding agents often report success while quietly introducing risk. We'll examine how to build verification layers that catch what the agent won't.
Enterprise-scale adoption requires enterprise-scale controls. What works for a solo developer in a sandbox breaks down across hundreds of engineers and production systems. This event maps the path from ad-hoc agent use to governed, auditable deployment at scale.

AGENDA

3:30pm: Registration & Mingling
5:00pm: Keynote
5:45pm: Breakout #1 (2 tracks)
6:30pm: Breakout #2 (2 tracks)
7:15pm: Breakout #3 (2 tracks)
7:45pm: Reception
9:00pm: Goodnight

Session Descriptions

Fireside Chat with Rohan Varma from OpenAI on Coding and Agents
- Hear from Rohan Varma from OpenAI, moderated by Ankit Mathur, Databricks on what it really takes to build and ship autonomous coding agents at scale. Rohan will dig into the shift from simple autocomplete to full Agent Development Environments, and how teams can balance developer productivity with the governance, verification, and controls required for enterprise use.
Your Agent Thinks It Did a Great Job by Jason Schwartz, Anthropic
- A year ago early agents would last a few minutes unattended before confidently proclaiming the task was done. Now at Anthropic we run Claude for hours building full apps end to end. The models are obviously getting better every week but this talk will focus on the harness we built to get there. I'll cover the top failure modes plus the loops, adversarial evaluators, and rubrics that made 30 hours possible and which parts we expect future models to make obsolete.
The Coding Agent Multiverse of Madness by Ankit Mathur, Databricks
- As AI coding agents proliferate across the enterprise, the real crisis isn't capability: it's control. Ankit Mathur argues that the explosion of agentic frameworks is quietly creating a "multiverse of madness": agent sprawl, compounding security risks, and hidden costs that "10x engineers" may be accelerating rather than solving.
  
  This talk examines the technical architecture powering modern coding agents, LLMs as reasoning engines inside plan-act-observe-refine loops, alongside the organizational reality of deploying them at scale. Ankit makes the case for a gateway layer to govern agent access before enterprises lose visibility entirely, because as the human role shifts from coder to system architect and reviewer, the question isn't just whether agents can do the work, but whether anyone is still in charge.
When Good Creds Go Bad: The New Failure Mode of AI Agents by Nancy Wang (CTO, 1Password), Jeff Malnick (VP of Engineering, 1Password), and Richard Liu (Member of Technical Staff, Anthropic)
- AI agents operate as authenticated principals within existing access control systems. Using valid credentials, they can perform actions that are policy-compliant but semantically incorrect, such as retrieving sensitive data into the wrong context or executing unintended changes. The failure mode is not unauthorized access, but misaligned execution under valid authorization.
- Traditional IAM systems evaluate identity and permissions, not intent or correctness at runtime. As agents scale across development and production workflows, this creates a new risk surface. This talk examines that shift and explores approaches like contextual policy evaluation and just-in-time credentialing to better constrain agent behavior at runtime.
The 5 Levels of Agentic Data Engineering by Gleb Mezhanskiy, DataFold
- Every data practitioner will tell you they're using AI. But the leverage they're getting varies by an order of magnitude, and it depends almost entirely on where their workflow sits on the agentic maturity curve. The best data engineers used to be 10x; the ones operating at the top of this curve now are closer to 30–50x. This talk walks through five levels of AI automation in the data engineering workflow, from tab completion to autonomous agent teams. We'll look at what defines each level, why data work has a harder context and harness problem than software engineering, and how teams can climb the curve without sacrificing data quality or security.

More coming soon!

Agentic + AI Coding Night SF