Browse our notes from both livestream days. Matching Day 1 sessions also link to their official reading editions.
Explore topic coverage · Find a lesson by topic · Day 1 digest · Day 2 digest
48 session summaries across both days
Day 1 · Oct 13
Bridgewater built Vault, a governed 100-million-document platform for external and internal unstructured data, because frontier models fail on the open web — and the real prize is an always-on agent that captures what the firm learns back into the corpus.
Day 1 · Oct 13
Anthropic's Belinda Neal and Keon Shahab argue that model intelligence now runs far ahead of adoption in financial services, and they lay out a five-layer agent foundation (MCP connectors, skills, identity and permissions, approval, evals) plus Claude for Financial Advisors as the proof that the same pattern scales fro…
Day 1 · Oct 13
Two Sigma's John Feminella retells Citibank's $900 million Revlon wire error to show why agent safety and speed are not a tradeoff, then lays out five composable controls — an agent registry, risk-scored review, declared-purpose tool access, shared evals, and classifier-governed runtime — that let a regulated firm run…
Day 1 · Oct 13
The author of Code Simplicity reduces all software success to understanding and execution, splits software into durable versus disposable, and answers the title directly: quality matters for durable software, not for disposable software — and rigor does not have to be expensive.
Day 1 · Oct 13
Aparna Dhinakaran (Arize AI) argues the AI engineer's job has shifted from writing AI features to building the always-on loop — agents, evals, and telemetry — that writes, reviews, and improves those features, illustrated by Arize's Signal agent.
Day 1 · Oct 13
Wells Fargo pre-trained its own 1-billion-parameter transaction foundation model from scratch on billions of monthly customer transactions, and it beat the production XGBoost fraud model on detection rate, false positives, and latency.
Day 1 · Oct 13
Arca, an "AI native" wealth-management services firm, gives agents four primitives (signals, operational context, a harness with persistent memory, and human collaboration) so they own 10-day client goals end to end — and the firm steers the business by steering its agents.
Day 1 · Oct 13
A Keycard-moderated panel on runtime authorization for AI agents: approve each action against the agent's mission rather than standing OAuth scopes, using AAuth missions and budgets, supervising agents, egress proxies, and tamper-proof audit logs.
Day 1 · Oct 13
A certified financial planner at Range explains how non-coding planners wrote the Claude skills they use daily, and gives a five-step method (pick, teach, pack, test, ship) for getting domain experts to author their own agents.
Day 1 · Oct 13
A Keycard demo talk that first lets an overprivileged on-call triage agent leak customer PII and drop a production database, then rebuilds the same agent with OAuth 2.0 token exchange and intent-based authorization so every tool call gets an ephemeral, task-scoped, attributable credential governed by policy.
Day 1 · Oct 13
Branislav Brzak demonstrates how timing on shared GPUs can reveal model and workload information without reading another tenant’s memory. Reported results refer to distinct experimental setups.
Day 1 · Oct 13
Mahmoud Abdelwahab of Railway replaced his company's marketing SaaS with one agent-built internal tool.
Day 1 · Oct 13
Southbridge AI co-founder Hrishi Olickel argues that no universal schema exists — only four schemas that align solely at decision time — so build just-in-time ontologies, split intelligence into three pipeline roles, and let self-optimizing "Pareto surfing" pipelines drive an Ohio campaign-finance entity-resolution job…
Day 1 · Oct 13
An extra lunch talk from Perpetual's CTO arguing that reliable AI in regulated finance comes from deliberately placing the boundary between deterministic compiled workflows and minimal, typed model judgment - not from choosing "more agency" or "less agency".
Day 1 · Oct 13
Two Flywire engineers open their lunch talk on structured agentic workflow automation — Flywire's compliance context, the Victoria PaaS, and the Orbit platform — before the captured livestream audio cuts off after less than three minutes.
Day 1 · Oct 13
Modal CTO Akshat Bubna shows what agents actually need from infrastructure — sandboxes holding multiple containers and sidecars, shared volumes, snapshots, path-level network control, and agent-facing observability — backing workloads like Hebbia's agent-per-spreadsheet-cell at up to 10,000 sandboxes per second.
Day 1 · Oct 13
Rayan Krishnan of Vals AI argues that model cards hide a jagged, task-dependent frontier, so firms should mine their own work history into rigorous evals to pick models, cut spend, and codify their competitive advantage.
Day 1 · Oct 13
In a Kepler-hosted fireside, former Citadel PM and Candlestick Capital founder Jack Woodruff argues that AI will not change long-term stock picking or divergent-view alpha, but will flatten analyst prep work and turn investing into a time-management game where humans stay the bottleneck.
Day 1 · Oct 13
Revolut's David Abitbol explains how self-hosting GLM on 56 dedicated B200 GPUs cut the bank's token bill to one-tenth, lifted engineering lead time 70%, and still passed EU bank compliance through sandboxing and human-in-the-loop controls.
Day 1 · Oct 13
With AI access now universal, Sid Pednekar argues that advantage comes from judgment plus three pillars: design the right problem, build shared capabilities, and compound the advantage.
Day 1 · Oct 13
Coatue Head of AI Frank Long models a hedge fund as a function that compresses external and internal context into one tuple — selection, size, timing — and decomposes it into evaluable steps, each run by the best of four runner types: code, statistics, LLMs, or people.
Day 1 · Oct 13
General Intuition CEO Pim de Witte presents one model family that converts pixels to actions in real time — trained on Medal's game-play action data, it plays video games at the highest level and controls over 10 robot types, on the theory that scaling pixel space repeats Anthropic's bet on coding space.
Day 1 · Oct 13
In a 98-second closing Q&A, Pim de Witte distinguished General Intuition's action-label-trained foundation models from JEPA, calling JEPA "the more correct approach" in the limit but far less practical short-term.
Day 1 · Oct 13
Nous Research CEO Dillon Rolnick argues that commoditized models leave frontier labs two ways to recoup their trillion-dollar bets — trap business users at the agent layer or "build God" and aim it at lucrative problems like finance — so companies should build agents on open foundations they own, such as the MIT-licens…
Day 2 · Oct 14
Ramp Labs' Alexander Shevchenko argues that an application-layer company should not wait for the next frontier model, and shows three Ramp cases where applied research closed the gap: a post-trained Qwen 3.5 specialist, KV-cache cartridges, and 10,000-agent pen-test swarms.
Day 2 · Oct 14
Sonar CEO Tariq Shaukat argues that agentic coding makes code abundant and confidence scarce, and presents a four-part code confidence formula (intent, guardrails, verification, evidence) for engineering trust into agentic software factories.
Day 2 · Oct 14
Coinbase engineering manager Evan Kormos maps policy-driven workflows as deterministic DAGs on the outside, lets agents loop adaptively inside each node, and turns autonomy on only when measured evaluation quality clears a policy threshold.
Day 2 · Oct 14
Joseph Kim, head of ChatGPT for financial services at OpenAI, presents the month-old product built with Morgan Stanley and Evercore, frames it as a way to close the "capability overhang", and predicts models will outperform the best human experts on financial judgment tasks by 2027.
Day 2 · Oct 14
Bloomberg's Andrey Rybka walks through a catalog of unnamed real agent incidents — runaway agents, a July 26 sandbox escape with 17,600 actions in 4.
Day 2 · Oct 14
A former Coatue VP and Compound AI co-founder explains how Compound rebuilt Excel, PowerPoint, and Word from scratch with coding agents so multiplayer, long-running agents can move buy-side finance work from copilot to autopilot.
Day 2 · Oct 14
A SemiAnalysis research analyst argues that identical GPUs deliver very different amounts of useful intelligence depending on who operates them, so compute buyers, benchmarks, and futures markets must grade the operator, not just the chip.
Day 2 · Oct 14
Evan Conrad of San Francisco Compute explains why a compute market only works if it is physically settled, why credit (not power) is the number-one AI bottleneck, and what happens to startups if nobody builds one.
Day 2 · Oct 14
Jump Trading's compute trading leads map the unlit, chat-brokered over-the-counter market for physical GPU compute, from Jevons-paradox demand and the prepayment squeeze on middle-market buyers to the microstructure and policy that could turn compute into a real market.
Day 2 · Oct 14
Carmen Li, CEO of both Compute Exchange and Silicon Data, argues for a neutral GPU exchange that owns no hardware, verifies both sides of every trade, and publishes the term-structure curves and residual values that now feed CME futures, NVIDIA's investor deck, and bank appraisals.
Day 2 · Oct 14
A recorded, off-agenda talk in which a senior staff engineer from Intuit Credit Karma shows how to move AI agents from experiment to enterprise: an MCP gateway with JWT auth and Redis sessions, security-reviewed FastMCP servers, OpenTelemetry-based monitoring, LangGraph with human-in-the-loop, memory, eval tables, a PI…
Day 2 · Oct 14
Michele Grant, creator of Leading at Scale, uses a $15 billion M&A integration case to argue that a working agent earns enterprise trust only through a written value case with baselines, assigned ownership under the three lines model, and engineering evidence gated by a four-stage autonomy scorecard.
Day 2 · Oct 14
Jellyfish research on the industry's largest engineering-workflow-plus-AI-telemetry dataset shows AI is only about 2.5% of the cost of a merged PR, so token bills are still cheap relative to salary - track cost per merged PR, not the token bill.
Day 2 · Oct 14
Nori Agentic's CEO Amol Kapoor and CTO Clifford Ressel, joined on screen by their agent Jiro, explain in a podcast-format online-track talk how cloud agents can maintain a securities master whose every cell change today sends a person hunting for provenance in Google Drive, email, and contract PDFs.
Day 2 · Oct 14
A fireside chat in which Scott Worland (CTO, Norm Ai) interviews John Stecher (CTO, Blackstone) on keeping a technology roadmap durable while frontier AI jumps every few months: fix the data strategy first, treat evals as the quality bar, build throwaway POCs fast, and leave conduits open for frontier intelligence inst…
Day 2 · Oct 14
Kepler CEO Vinoo Ganesh argues that Wall Street's distrust of AI is structural, not a model-quality problem, and shows how making code write every number through deterministic pointers yields citations that are verified, corrected at search time, or refused.
Day 2 · Oct 14
Philipp Schmid of Google DeepMind builds "Bug", an autonomous Wall Street co-worker, live on Gemini managed agents — with skills, MCP connectors, write-only credentials behind a network proxy, hooks, cron triggers, and a Gemini 4.0 recap.
Day 2 · Oct 14
Citadel Securities' Joao Wenzel argues that the one person who "seems to know everything" is a company's only world model, and shows the state-estimator loop (update, age, observe, grade, correct) his team uses to bootstrap one from decks, Slack, meetings, code changes, and agent traces — a job RAG cannot do.
Day 2 · Oct 14
Will Brown of Prime Intellect argues that after chatbots, reasoners, agents, and innovators, the next stage of AI progress is the AI-native enterprise: an organization that compounds its implicit institutional knowledge by continually training its own open-source models, instead of re-teaching agents in context on ever…
Day 2 · Oct 14
Jemima Abu (CAIS) argues that letting AI agents ship inaccessible products wastes tokens, time, and money — her experiment saw a 58% cost reduction and one-third the build time when the agent ran with an accessibility skill.
Day 2 · Oct 14
Coder staff product manager Zhao Chenjiang opens a talk about three real incidents of AI usage gone wrong at real companies by arguing they were all preventable with infrastructure controls you mostly already own — not with a smarter, more expensive frontier model — but the stream cut to a music interstitial before the…
Day 2 · Oct 14
Rowspace co-founder Michael Manapat interviews Jack McDonald, COO of Apollo's Redding Ridge Capital, about turning AI from a nice-to-have into mission-critical infrastructure for structured credit and CLO management.
Day 2 · Oct 14
Basis co-founder Mitchell Troyanovsky argues that agents only earn end-to-end production use when they are simple and reliable, and shows three tools — behavior specs, forking, and simulations — that supervise the process of long-horizon work, not just the outcome.
Day 2 · Oct 14
Runway co-founder Cristóbal Valenzuela argues that next-frame prediction, scaled far enough, turns video models into universal simulation engines, and demos Continuum, a video-model operating system that Runway plans to release within 8 to 12 months.
No sessions match. Try another search or choose both days.
The Intuit Credit Karma recording has two source records. It appears once in the session list; both records remain in the archive.