See how engineers who've already shipped agentic systems are rethinking the architecture — from the people who built fleets, watched them fail, and rebuilt around a different model.

​You've got agents running. The question now is whether one well-trained agent could outperform all of them.

In collaboration with Lazy 8, this is for AI engineers who've moved past deployment and are working on the harder problems like:

• ​what actually breaks when you give a single agent broad access and real autonomy

• ​how to train, evaluate, and optimize an agent that understands how your whole company runs

• ​what it takes to push reliable autonomous time horizon past the point where you'd trust it overnight

• what non-technical team members actually adopt in mature organizations

​SPEAKERS

• ​Arnaud Ferreri, CTO at Headway
  Healthcare AI infrastructure

• ​Mike Taylor, Head of AI Technology Consulting at Every
  Leading AI consulting engagements with technology teams

​TOPICS

​When Compliance Forces You to Build

• ​Arnaud Ferreri on what it takes to deploy agentic systems inside a regulated environment. Why standard tooling failed their security requirements, what they had to build instead, and how that custom infrastructure enabled roughly a quarter of their non-technical staff to ship production code.

​Against the AI Org Chart

• ​Mike Taylor on why his team built a multi-agent system, watched their people ignore it in favor of a single unified agent, and rewrote their entire internal AI strategy as a result. He'll cover how to train and optimize a super-employee using structured evals and programmatic prompt refinement, and where this architecture is heading.

​Fireside: Where Control Breaks Down

• ​Arnaud and Mike on the real failure modes — maintaining control over an agent with broad access, what cost per successful task looks like in a unified model versus a fleet, and the architectural decisions that determine whether your agent runs reliably for 40 minutes or four hours.

​WHAT YOU'LL GET

• ​A framework for building custom connectors and tooling when off-the-shelf options fail your compliance requirements

• ​A mental model for the super-employee architecture — how a single orchestration layer with unified context outperforms fragmented specialist agents for most real-world tasks

• ​A working approach to training and optimizing agents using DSPy, structured evals, and programmatic prompt refinement

• ​Two contrasting case studies: one inside a regulated healthcare environment, one inside an AI consulting practice — and what the failure modes look like from both sides

• ​Happy Hour specials: enjoy cheap drinks and food all night at the newest hip spot on Union Square

​🚨 This event is for engineers and technical leads already shipping agents in production who are ready to rethink how they're structured.

​AGENDA

• ​6:00 — Arrive & Mingle

• ​6:40 — Arnaud Ferreri: When Compliance Forces You to Build

• ​7:00 — Mike Taylor: Against the AI Org Chart

• ​7:20 — Fireside chat + Q&A

• ​7:40 — Meet the Speakers