AI Engineer · Tenex
October 13 · Remarks
Opening Remarks
Lia McBrideAI EngineerGeneral Manager
Alex LiebermanTenexManaging Partner
AI engineering meets Wall Street: a room of builders comparing what already works, and what might become normal next.
← AI Engineer NYC 2026Download all transcripts ↓ 
October 13–14, 2026 · New York City
The talks, in writing.
The people building AI across finance, frontier labs, and infrastructure. Explore their illustrated reading editions, recordings, and transcripts.
53 of 53 sessions · Recording order · timestamps are positions in the livestream
Tuesday, October 13
Full livestream ↗Bridgewater
October 13 · Main stage
Vault: Making Unstructured Data Work for Investing
Aaron LinskyBridgewaterHead of Product Engineering
Sam GreenBridgewaterTeam Lead, Applied AI - AIA Labs
Investment agents need the research an investor would read—and a way to remember what the firm learns. Bridgewater’s Vault puts both into the same research loop.
Anthropic
October 13 · Main stage
Anthropic: Accelerating frontier AI in Financial Services.
Belinda NealAnthropicManaging Director, Financial Services
Keon ShahabAnthropicApplied AI Architect
A trade-break agent can investigate a discrepancy without having permission to change the order. Belinda Neal and Keon Shahab show how to reuse that separation across financial workflows.
Two Sigma
October 13 · Main stage
Balancing Acceleration and Safety for AI Agents
John FeminellaTwo SigmaSVP Architecture
John Feminella uses a catastrophic payment mistake to explain five controls that make agents visible, constrain their access, and reserve human attention for decisions that need it.
Capital One
October 13 · Main stage
Does Code Quality Even Matter?
Max Kanat-AlexanderCapital OnePrincipal Software Engineer
Cheap code makes some programs disposable. For the systems people will keep changing, Max Kanat-Alexander argues that AI should make engineering rigor cheaper too.
Arize AI
October 13 · Main stage
What does an AI engineer do?
Aparna DhinakaranArize AICo-Founder & Chief Product Officer
Arize Signal reads traces, finds recurring problems, and can propose fixes. Building that loop requires decisions about memory, temporary access, recovery, and what counts as a good fix.
Wells Fargo
October 13 · Main stage
Wells Fargo Foundation Model: Application to Fraud Detection at Scale
Freddy LecueWells FargoHead of Frontier AI Models, Managing Director
Freddy Lecue explains why Wells Fargo trained a transformer on transactions rather than text, how it tokenizes heterogeneous banking data, and how the resulting representation supports fraud detection.
Arca
October 13 · Main stage
Inside Arca’s long-horizon agents that run the business
Aman SinghArcaLead AI Engineer
Aman Singh follows one client’s tax payment across days of analysis, approval, trading, and settlement to show how Arca’s agents keep work moving between human decisions.
AAuth · JPMC · Keycard
October 13 · Main stage
Intent Based Auth
Allie HoweKeycardMember of Technical Staff
Ian LivingstoneKeycardCEO and Co-Founder
Michael DavisJPMCGlobal Chief Security Architect
Dick HardtAAuthFounder
Permission to call a payment API cannot tell an agent whether it should pay this invoice again. This panel explores how mission history, budgets, and changing conditions can inform each authorization decision.
Range
October 13 · Online track
We Let Our Financial Planners Write the Agents
Matt RagoneRangeFinancial Planner
At Range, financial planners write and maintain the skills that draft their work. Matt Ragone explains how to turn professional judgment into a repeatable process without handing client communication to an agent.
Keycard
October 13 · Online track
Your Agent Is a Lethal Trifecta (and What to Do About It)
Kim MaidaKeycardFounding GTM Engineer
An on-call agent can be useful and dangerously overprivileged at the same time. Kim Maida follows the same incident queue before and after adding identity, token exchange, and task-scoped policy.
Stealthium
October 13 · Online track
Your GPU Neighbor Can Hear Your Model
Branislav BrzakStealthiumCybersecurity Researcher
Branislav Brzak demonstrates how timing on shared GPUs can reveal model behavior even when memory and compute are partitioned, then explains what operators can control.
Railway
October 13 · Online track
Please… Enough with the Sloppy, One-Shot SaaS Replacements
Mahmoud AbdelwahabRailway
A working internal app can still be unpleasant to use. Mahmoud Abdelwahab walks through the product details, reusable patterns, and verification loop behind Railway’s agent-built marketing tools.
Southbridge
October 13 · Online track
Just-in-Time Ontologies
Hrishi OlickelSouthbridge
The meaning of a customer depends on the decision being made. Hrishi Olickel explains why data pipelines need contextual schemas, distinct model roles, and bounded experiments that can trade cost against quality.
Perpetual
October 13 · Online track
Building Reliable, yet Smart Workflows for Finance, Wealth and Insurance
Karthic RaoPerpetualCTO
A request to complete one form can become a process lasting days. Karthic Rao follows that request through Perpetual’s compiler, approval gates, and typed model calls.
Flywire
October 13 · Online track · Excerpt
Agents in Orbit: Workflows, Not Autopilot
Eduardo SimonFlywire
Javier CasalFlywire
Flywire’s Orbit brings engineering agents into an existing GitLab and AWS environment. This opening excerpt covers the company’s infrastructure and privacy constraints before the livestream returns to the live stage.
Modal
October 13 · Main stage
Great Agents Start with Great Infrastructure
Akshat BubnaModalCTO
A spreadsheet with an agent behind every cell needs fast compute, shared storage, and deliberate isolation. Akshat Bubna explains how those requirements change the sandbox beneath an agent.
Vals AI
October 13 · Main stage
Beyond the Model Card: The Jagged Frontier of AI in Finance
Rayan KrishnanVals AICo-Founder & CEO
A model can earn strong partial credit and still produce work that cannot be sent to a client. Rayan Krishnan shows how evaluations built from a firm’s own work can guide model choice, spending, and deployment.
Kepler
October 13 · Main stage
AI and the Future of Fundamental Investing
Jack WoodruffKeplerSenior Advisor
Vinoo GaneshKeplerCEO
Jack Woodruff sees AI changing research preparation and analysts’ roles while leaving a central investment problem intact: deciding where limited human attention can uncover a divergent view.
Revolut
October 13 · Main stage
Shipping like a startup, regulated like a bank
David AbitbolRevolutOps & Infrastructure Leader
David Abitbol describes Revolut’s move toward self-hosted coding models, the inference tuning that made them practical, and the operational controls used around a regulated engineering organization.
Sid Pednekar
October 13 · Main stage
Everyone Has AI. Who Has the Advantage?
Sid PednekarHead of Data and AI | Asset Management
Sid Pednekar follows two hypothetical investment firms to show why widespread access to AI can still produce very different results: the scarce resource is the judgment that shapes the work.
Tenex
October 13 · Remarks
Welcome Back - Featured Keynotes
Alex LiebermanTenexManaging Partner
Alex Lieberman returns to the adoption gap: distributing an LLM is much more common than putting agents to work across a business.
Coatue
October 13 · Main stage
A Hedge Fund as a Function
Frank LongCoatueHead of AI
Frank Long models an investment firm as a function that turns external and internal context into selection, size, and timing. The engineering challenge appears when that function is decomposed into work that code, models, and people can each perform well.
General Intuition
October 13 · Main stage
Frontier Models for Acting in Space and Time
Pim de WitteGeneral IntuitionCo-Founder & CEO
Pim de Witte explains General Intuition’s bet on action-labeled game data: models that observe continuously and select actions while the world keeps moving, with language models available for longer-term planning.
Nous Research
October 13 · Main stage
Hermes gives you wings: Build on open foundations.
Dillon RolnickNous ResearchCEO
Dillon Rolnick makes the case for owning the agent layer as AI becomes part of everyday business operations. His argument connects model economics to a practical architecture: interchangeable models beneath software and workflows a company controls.
Tenex
October 13 · Remarks
Closing Remarks
Alex LiebermanTenexManaging Partner
The closing invitation is simple: turn the day’s ideas into conversations with the people who are building them.
Wednesday, October 14
Full livestream ↗Turing Post
October 14 · Remarks
Welcome Remarks
Ksenia SeTuring PostAI Inferencer
Ksenia Se opens the final day of AI Engineer NYC with a question about how agent capabilities change the work inside companies.
Ramp
October 14 · Main stage
Chasing capabilities while waiting for the frontier
Alexander ShevchenkoRampHead of Ramp Labs and Applied Research
Ramp keeps adopting better frontier models. Alexander Shevchenko shows where specialist training, prepared context, and parallel agents can still save users time before the next model arrives.
Sonar
October 14 · Main stage
The Machine that Changed the World, vAI
Tariq ShaukatSonarCEO
Agents can produce code much faster than a company can ship it. Tariq Shaukat uses factory design to explain where guidance, independent verification, and evidence belong in an agentic development process.
Coinbase
October 14 · Main stage
Trusting Agents in Policy-Driven Workflows
Evan KormosCoinbaseEngineering Manager
A regulated workflow can move forward through fixed approval stages while agents iterate inside each stage. Evan Kormos explains how to choose those stages, evaluate the work, and reduce supervision only when measured confidence supports it.
OpenAI
October 14 · Main stage
Building the Next Chapter of AI in Finance
Joseph KimOpenAIHead of ChatGPT for Financial Services
A capable model still needs financial data, familiar outputs, and a way to check its work. Joseph Kim explains the product choices behind ChatGPT for Financial Services and the thresholds he expects finance agents to cross next.
Bloomberg
October 14 · Main stage
Security in the Age of AI: Recent Attacks, Incidents & Defenses
Andrey RybkaBloombergHead of Architecture & Emerging Technologies in the Office of the CTO
Andrey Rybka follows failures from stolen credentials to poisoned memory. His defense starts with what an agent can actually do: short-lived identities, checks outside the model, and limits that stop execution.
Compound AI
October 14 · Main stage
Towards Autonomous Analysts in Finance
Yijia LiangCompound AICo-Founder
An investment team needs more than a chatbot that produces a spreadsheet. Yijia Liang explains why Compound built shared agents, its own Office engines, and tools for navigating large data rooms before asking analysts to delegate longer pieces of work.
SemiAnalysis
October 14 · Main stage
Not All GPU's Are Deployed The Same
Oliver KennonSemiAnalysisResearch Analyst, Core Research
Two deployments of the same GPU can produce different amounts of useful output. Oliver Kennon separates chip rental prices from production costs, then explains why operators, service definitions, and benchmarks matter to the compute market.
San Francisco Compute Company
October 14 · Main stage
How Compute Markets Work
Evan ConradSan Francisco Compute CompanyCEO
Evan Conrad traces compute markets back to a startup’s inability to buy the right amount of reliable capacity—and explains why his company ended up operating both the exchange and the hardware.
Jump Trading
October 14 · Main stage
The Dark Pools of Compute
Affaan MustafaJump TradingCompute Trading Lead
Alejandro Soumah FernandezJump TradingCo-Lead Compute Trading
Affaan Mustafa and Alejandro Soumah Fernandez map the financing, delivery, and price risks hidden inside a GPU-hour quote—and the market structures needed to separate them.
Compute Exchange / Silicon Data
October 14 · Main stage
Compute Exchange
Carmen LiCompute Exchange / Silicon DataCEO
Carmen Li demonstrates a neutral market for reserved GPUs, forward contracts, and used hardware, then connects procurement terms to forward curves and residual asset value.
Intuit Credit Karma
October 14 · Online track
Paving the Road to Production: Governed Enterprise AI Agents
Harish GaggarIntuit Credit KarmaSenior Staff Software Engineer
Harish Gaggar follows an enterprise agent from registration and deployment through authenticated tool calls, checkpoints, evaluation records, privacy controls, and context optimization.
Leading at Scale
October 14 · Online track
The Agent Works. Does Anyone Trust It?
Michele GrantLeading at ScaleCreator
Michele Grant follows an acquisition-reporting agent from a successful pilot to the evidence, ownership, and review practices needed for greater autonomy.
Jellyfish
October 14 · Online track
Tokens Are the Cheap Part
Steve JamisonJellyfishAI Research and Engineering Leader
A rising AI bill can accompany a falling cost of delivery. Steve Jamison combines salary and token spending to examine the cost of a merged pull request—and checks whether the apparent gains hide smaller changes or worse quality.
Nori Agentic
October 14 · Online track · Excerpt
The Automated Security Master: keeping your master database clean with agents
Amol KapoorNori AgenticCEO
Clifford ResselNori AgenticCTO
In this livestream excerpt, Amol Kapoor and Clifford Ressel explain how cloud agents can trace a changed security-master cell back to its evidence—and why the person who understands the discrepancy still matters.
Blackstone · Norm.ai
October 14 · Main stage
Built to Scale: Future-Proofing Your Technology Roadmap
John StecherBlackstoneCTO
Scott WorlandNorm.aiCTO
John Stecher and Scott Worland explain where to invest as models improve: accessible data, evaluations, and production controls—and where to stay willing to replace yesterday’s code.
Kepler
October 14 · Main stage
Why Wall Street Still Doesn’t Trust AI
Vinoo GaneshKeplerCEO
Vinoo Ganesh separates numerical trust into source selection, exact retrieval, arithmetic, access control, and output verification—and explains why a citation added at the end cannot do all those jobs.
Google DeepMind
October 14 · Main stage
Coding an Autonomous Wall Street Coworker: Custom Skills, Connectors, and Event Hooks
Philipp SchmidGoogle DeepMindMTS
A useful background agent needs a workspace, a playbook, credentials, and rules for when to act. Philipp Schmid builds a portfolio-auditing coworker around Gemini Managed Agents, showing how those pieces fit together.
Citadel Securities
October 14 · Main stage
Your Company Has No Model of Itself: Bootstrapping a World Model from Slack, Decks and Meetings
Joao WenzelCitadel SecuritiesCOO
A green status slide can conceal a project already slipping. Joao Wenzel describes a system that tracks what a company believes, how old the evidence is, and what happens next—then asks people to correct the beliefs that matter.
Prime Intellect
October 14 · Main stage
Self-Improving Organizations
Will BrownPrime IntellectApplied Research
Will Brown describes how an organization’s corrections, decisions, and repeated workflows could become training data. Start with evaluations and routing, then give specialized models a way to improve alongside the people using them.
CAIS
October 14 · Online track
The Cost of Accessibility in AI
Jemima AbuCAISSenior Product Engineer
Jemima Abu uses a small coding-agent experiment to show how missing accessibility requirements create extra code, extra repair work, and products people cannot use.
Coder
October 14 · Online track · Excerpt
Your AI Isn’t the Problem. Your Infrastructure Is.
Jiachen JiangCoderStaff Product Manager
This livestream excerpt captures Jiachen Jiang’s opening argument: an instruction to avoid destructive work is weaker than infrastructure that prevents the action.
Turing Post
October 14 · Remarks
Welcome Back — Featured Keynotes
Ksenia SeTuring PostAI Inferencer
Ksenia Se returns for the final keynotes with a practical conference project: use an agent to research potential employers, then make the introductions yourself.
Apollo Global Management · Rowspace
October 14 · Main stage
AI in Practice: An Inside Look at Apollo's Approach
Michael ManapatRowspaceCo-Founder
Jack McDonaldApollo Global ManagementPrincipal at Apollo Global Management, COO of Redding Ridge
Michael Manapat and Jack McDonald explain how agents turn fragmented credit data into usable workflows: prepare a portfolio, reconcile financing activity, and leave investment judgment with the people responsible for it.
Basis
October 14 · Main stage
Process Supervision for Agents That Do Long-Horizon Work
Mitchell TroyanovskyBasisCo-Founder
Mitchell Troyanovsky shows how behavior specs, trajectory forks, and inexpensive simulations help diagnose agents that work for hours. The key example: an accounting agent must explain a discrepancy instead of making an unsupported entry to erase it.
Runway
October 14 · Main stage
Everything will be Simulated
Cristóbal ValenzuelaRunwayCo-Founder and Co-CEO
Runway’s early films generated images to watch. Cristóbal Valenzuela shows research aimed at generating images people and agents can act on, moving from video clips toward interactive interfaces and shared visual reasoning.
AI Engineer · Turing Post
October 14 · Remarks
Closing Remarks
Ksenia SeTuring PostAI Inferencer
swyxAI Engineer
Ksenia Se and swyx close the conference by returning to the conversations, engineering work, and community that continue after the talks.