Uploaded May 2025 | Updated September 2026, 3 weeks ago
[2025 - Day 3 - Lightning Talks] Arvind Prabhakar shares insights from reimagining data integration through Pub/Sub for Tables, exploring how data producers can publish complete, ready-to-use tables with embedded metadata and semantics. This talk offers valuable perspectives on enabling quality checks, data contracts, and observability without pipeline complexity, essential for data teams seeking to move beyond traditional record-moving approaches that strip away context and meaning.
ABOUT THE SPEAKER:
Arvind Prabhakar, Co-founder & CEO, Tabsdata -
ποΈ GET YOUR TICKET TO AI COUNCIL 2026 ποΈ
Meet the world's top AI infrastructure minds where architects of AI share what works. Three days of high-quality technical talks and meaningful interactions.
β aicouncil.com/sf-2026
β‘ FIND US:
X: https://x.com/AICouncilConf
LinkedIn: linkedin.com/company/aicouncilconf
Website: aicouncil.com
[2025 - Day 3 - Lightning Talks] Arvind Prabhakar shares insights from reimagining data integration through Pub/Sub for Tables, exploring how data producers can publish complete, ready-to-use tables with embedded metadata and semantics. This talk offers valuable perspectives on enabling quality checks, data contracts, and observability without pipeline complexity, essential for data teams seeking to move beyond traditional record-moving approaches that strip away context and meaning.
ABOUT THE SPEAKER:
Arvind Prabhakar, Co-founder & CEO, Tabsdata -
ποΈ GET YOUR TICKET TO AI COUNCIL 2026 ποΈ
Meet the world's top AI infrastructure minds where architects of AI share what works. Three days of high-quality technical talks and meaningful interactions.
β aicouncil.com/sf-2026
β‘ FIND US:
X: https://x.com/AICouncilConf
LinkedIn: linkedin.com/company/aicouncilconf
Website: aicouncil.com
![Trinity: Training a 400B MoE from Scratch Without Losing Your Mind | Arcee
[2026 - DAY 3 - MODEL SYSTEMS] Training sparse Mixture-of-Experts models at scale is notoriously unstable. Experts collapse, routers drift, and loss spikes appear out of nowhere. This talk covers how we built Trinity Large, a 400B parameter MoE (13B active), trained on 17 trillion tokens with zero loss spikes.
Well walk through the decisions that actually mattered: why we replaced standard aux-loss-free balancing with a momentum-based approach (SMEBU), how interleaved local/global attention made context extension surprisingly smooth, and what broke when we first tried running Muon at scale.
Ill also cover the less glamorous stuff: our Random Sequential Document Buffer to reduce batch heterogeneity, recovering from B300 GPU faults on brand-new hardware, and the six changes we shipped at once when routing started collapsing mid-run.
Practical lessons for teams training their own MoEs or scaling up sparse architectures
SPEAKER:
Lucas Atkins - CTO, Arcee AI
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf Trinity: Training a 400B MoE from Scratch Without Losing Your Mind | Arcee](https://i.ytimg.com/vi/_GXUlM5DCL4/mqdefault.jpg)
![A Guide to Deploy an Enterprise-Ready ClawdBot in Under 30 Minutes | TextQL
[2026 - DAY 1 - WORKSHOP] The always-on AI agent that manages your calendar, drafts emails, runs analyses, and executes real tasks. Under this promise, Clawdbot took the world by storm, until CISOs (the fun police) quickly realized the drawbacks - plaintext credentials, security holes, auth issues, massive token burn, unrestricted access to your entire system...
The capabilities are real. The risks are disqualifying. What if you didnt have to choose?
In this short session, we deploy TextQL: our enterprise agentic analytics platform with the capabilities your team wants from ClawdBot and the guardrails your CISO demand. SOC 2 and HIPAA compliance, scoped data access through a semantic layer, PII anonymization before any LLM touches your data, and sandboxed compute that eliminates runaway costs. Well also set up a multi-agent architecture, which monitors, analyzes, and surfaces insights across your data. Everything Clawdbot promised, we cover in this talk.
Who this is for:
Data and technical leaders who felt the pull of ClawdBot but knew the risks were unacceptable. This is the version of autonomous AI your security team will actually approve.
SPEAKER:
Ethan Ding - CEO, TextQL
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf A Guide to Deploy an Enterprise-Ready ClawdBot in Under 30 Minutes | TextQL](https://i.ytimg.com/vi/_LXYwdFAFCE/mqdefault.jpg)
![How to Unlock Enterprise Value by Training Your Own Language Models | Snowflake
[2026 - DAY 3 - APPLIED AI] At Snowflake I have been straddling between product engineering and model training. I have been involved with training snowflakes own embedding model series arctic embed v1 and arctic embed v2, and snowflakes text2sql model. So the talk will be about how at Snowflake we decided when to train a model vs when to take advantage of open source models / partner with frontier model providers. I could then go on to describe briefly the models we have trained and the technical learnings we discovered along the way.
SPEAKER:
Gaurav Nuti - Software Engineer, Snowflake
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf How to Unlock Enterprise Value by Training Your Own Language Models | Snowflake](https://i.ytimg.com/vi/_pk8IhWq9zw/mqdefault.jpg)
![How Notion Cut Millions from Their Vector DB Bill Without Sacrificing Search Quality
[2025 - Day 1 - Data Eng & Infra] Simon HΓΈrup Eskildsen and Mickey Liu share insights from powering one of the worlds largest vector search use cases, exploring the database and pipeline engineering behind Notions semantic search across billions of vectors. This collaboration between turbopuffer and Notion offers valuable guidance on managing massive vector workloads and object-storage native indexing, essential for teams connecting LLMs to data or building RAG systems at scale.
ABOUT THE SPEAKERS:
Simon HΓΈrup Eskildsen, Co-Founder, turbopuffer
Mickey Liu, Software Engineer, Notion
ποΈ GET YOUR TICKET TO AI COUNCIL 2026 ποΈ
Meet the worlds top AI infrastructure minds where architects of AI share what works. Three days of high-quality technical talks and meaningful interactions.
β https://aicouncil.com/sf-2026
β‘ FIND US:
X: https://x.com/AICouncilConf
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
Website: https://aicouncil.com/ How Notion Cut Millions from Their Vector DB Bill Without Sacrificing Search Quality](https://i.ytimg.com/vi/_yb6Nw21QxA/mqdefault.jpg)
![Five years of OpenLineage: How we built an industry standard and why agents need it | Datadog
[2026 - DAY 3 - LIGHTNING TALK] Over the past five years, OpenLineage has become the de facto standard for data lineage metadata, adopted across the industry by leading platforms and enterprises. In this talk, well trace the journey of building an open standard. Youll learn what changed in the ecosystem that made standardization possible, the critical features that drove adoption (column-level lineage, streaming support, unified facets), and where OpenLineage stands today - five years since its initial release. Most importantly, well explore why this matters now: as AI agents increasingly make decisions about data - where to read from, what to trust, how fresh it is - they need a shared understanding of data context. Lineage metadata is the knowledge graph that transforms agents from black boxes into informed decision-makers. The talk covers the standards perspective, the pragmatic integration challenges, and a forward-looking vision for how great metadata enables intelligent data systems.
SPEAKER:
Harel Shein - Senior Engineering Manager, Datadog
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf Five years of OpenLineage: How we built an industry standard and why agents need it | Datadog](https://i.ytimg.com/vi/a9S0SoXFXcQ/mqdefault.jpg)
![Building Durable, Long-Running Autonomous Agents | RedScope AI
[2026 - DAY 1 - AGENT INFRASTRUCTURE] Most AI agents work in demos. Few survive in production. LLMs are stateless. Infrastructure fails. Context windows reset. Real-world objectives span hours or days. Building long-running autonomous agents requires durability engineered across the entire system.
This talk compares and contrasts dominant approaches to durability for agents and presents three pillars of durable agentic systems.
1. Durable Execution
Agents must survive crashes, retries, and partial task completion. Durable execution engines like Temporal persist workflow state and enable deterministic replay. Graph-based orchestrators such as LangGraph model control flow as explicit state machines. These approaches reflect different assumptions about recovery, replayability, and operational resilience, and directly shape how agents behave under failure.
2. Durable Autonomy
Autonomous systems inevitably encounter ambiguity and incomplete information. Durable autonomy means designing agents that recognize uncertainty, escalate intelligently to humans when necessary, and resume coherently without losing progress. Weβll examine architectural patterns for human-in-the-loop integration that preserve control while maintaining forward momentum.
3. Durable Statefulness
Long-running agents cannot rely on ever-growing prompts. Some systems serialize state into resumable bursts using patterns like Anthropicβs Git-Commit approach. Others externalize cognition into layered memory architectures - separating working, episodic, semantic, or procedural memory through memory virtualization. Different workloads and time horizons demand different state strategies.
Viewers will leave with a deeper understanding of agent durability and a practical architectural framework for building resilient agents, systems designed not just to respond, but to endure.
SPEAKER:
Parminder Singh - Co-founder and CEO, Rescope AI
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf Building Durable, Long-Running Autonomous Agents | RedScope AI](https://i.ytimg.com/vi/aYSl1hbuPfs/mqdefault.jpg)

![AI Needs a New Kind of OLTP: Lakebase & Serverless Postgres in the Agent Era | Databricks
[2026 - DAY 1 - DATA ENG & DATABASES] AI agents are driving a new category of operational databases, creating workloads that look nothing like traditional SaaS traffic. In Lakebase today, over 80 percent of new databases are created programmatically by agents rather than humans, resulting in extreme burstiness, highly ephemeral environments, and large volumes of short lived databases. These patterns push classic OLTP assumptions, including always on instances, steady traffic, and tightly coupled storage and compute, beyond their limits. In this talk, we will explain why existing databases struggle with agent workloads and how a new OLTP design emerges from separating storage and compute, enabling fast autoscaling, true scale to zero, and database branching that allows agents to run experiments and roll back state instantly. Using Lakebase and serverless Postgres as a concrete example, we will share practical design lessons for anyone building data infrastructure in the agent era.
SPEAKER:
Stas Kelvich - Principal Software Engineer, Databricks & Neon.com co-founder
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf AI Needs a New Kind of OLTP: Lakebase & Serverless Postgres in the Agent Era | Databricks](https://i.ytimg.com/vi/b1ND5zxrmWU/mqdefault.jpg)


![Your Database Wasnt Built for This | CockroachDB
[2026 - DAY 2 - WORKSHOP] Databases were built for a world where humans were the primary users. That world is changing fast. As AI moves from copilots to autonomous systems, agents are starting to create environments, test queries, assist migrations, coordinate workflows, and make operational decisions at a speed and scale that humans cannot match. One useful way to think about this shift is that traditional software is crystallized intelligence: fast, reliable, and optimized for known paths. Agents are fluid intelligence: less optimized step by step, but far better at adapting to new situations and conditions. Production systems will increasingly need both, which means databases will need to serve not only human developers and operators, but also agents acting on their behalf.
This talk explores what that shift means for the data layer. As agents create more databases, generate machine scale traffic, run constant experiments, and operate continuously across systems and regions, old assumptions begin to break down. Databases must work efficiently across many scales, from tiny agent created databases to very large production systems. They must support safe sandboxes for experimentation, whether through instant cloning, isolated environments, or strong resource governance within shared systems. They must provide fine grained, on demand scoped permissions, strong governance, and full auditability for autonomous actions. And they must stay available through outages and failures, while providing the resource accounting and automatic cleanup needed for a world of temporary, agent driven infrastructure. The core argument is simple: production AI systems need more than a familiar transactional database. They need truth, governance, and control at machine scale, with the resilience to handle concurrency, growth, experimentation, and failure without losing correctness. Attendees will leave with a practical framework for thinking about database architecture in the agent era and what it takes to support agents and workflows in production.
SPEAKER:
Andy Kimball - Fellow, CockroachDB
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf Your Database Wasnt Built for This | CockroachDB](https://i.ytimg.com/vi/c0TUB9uez8k/mqdefault.jpg)