LLMday NYC at the Microsoft Experience Center
At LLMday NYC, the question was not what AI can do. It was what happens when it goes to work inside a business that has rules.
The day
On October 29, fifteen sessions ran back to back on a single track at the Microsoft Experience Center at 677 Fifth Avenue. LLMday is a worldwide series of community events for engineers building with large language models and AI agents. The New York edition was run by the team behind SREday, LLMday and PLATFORMday, with Microsoft providing the space and Stigian Consulting, Reliaburger and Hockeystick supporting the day as sponsors. Our thanks to all of them.
The room belonged to practitioners: engineers and platform teams whose job is to keep AI running after the demo ends. The speakers came from companies doing that work at scale, among them AWS, Databricks, dbt Labs, Google and USAA. Little of the day was about capability. Nearly all of it was about application.
Midway through the program, Josh Lee of Altinity put the AI adoption curve on screen: delight, then tokenmaxxing, then despair, then, eventually, productivity. He circled the dip and wrote beside it, “You are probably somewhere in here.” It captured the tone of the whole day: candid about where adoption actually stands inside working companies.
At the afternoon break, attendees took over the Experience Center itself, from a Forza racing rig to a walk-through history of Xbox. The day closed with LinkedIn profiles swapped and a walk to a nearby pub.
Three conversations worth carrying out of the room
An agent is not an employee. Sai Varma Dantuluri of VotalAI made the case that identity systems were built for people, not for software that spawns sub-agents and chains tool calls on its own. Lending an agent a human’s login is a shortcut that will not hold. For the banks and insurers headquartered here, the stakes are concrete: every action on a client account has to trace back to someone accountable. An agent that cannot be identified cannot be trusted with that work, which makes access the first design decision, not the last.
The bill reaches the CFO. Yiwen Zhang of AWS walked through the token economics of custom-built agents. Prices per token keep falling, but agents consume tokens faster than those prices drop. For the New York businesses that run on margin, the implication is plain: AI stops being an experiment and becomes an operating cost, and operating costs are managed by people who were never in the pilot.
Review is evidence. Chris Harding of groundcover brought the day’s sharpest provocation: when AI writes most of the code, the traditional pull-request review stops catching what matters. For New York’s regulated industries, that matters more than it sounds. Review there is more than an engineering habit. It is part of the record a firm shows its auditors. If the ritual is hollowing out, something has to replace it, and the businesses that answer first will set the standard.
Why it matters for New York
Technology is worth nothing until an industry uses it. The questions raised on Fifth Avenue (who an agent answers to, what it costs to run, how its work gets checked) are the questions any capable model meets the moment it enters a serious business. The businesses where those questions carry real weight, in banking, insurance, finance and real estate, are headquartered here. This is where the value of AI gets decided.
Capability is no longer scarce. The people who can carry it through a compliance review, a budget cycle and an audit are still hard to find. LLMday put a room of them in one building for a day. That network is what we follow.