PyMC Labs

I’m looking forward to being in the room for this one. At #AIWeek New York on October 6, I’ll moderate a panel with Amaryllis Liampoti from Boston Consulting Group (BCG) and Dan Morris from Databricks. Amaryllis is a Managing Director & Partner at BCG X, working with executives to scale data-driven marketing and enterprise AI. Dan leads industry solutions at Databricks, with a focus on agentic AI and customer data. I want to hear where their views meet: what has to change inside a company for data science agents to become part of everyday work? My head is buzzing with questions, but we’ll leave time for yours too. If you’re in New York, come join us. Tuesday, October 6 · 5:30–7:30 PM ET Panel and Q&A: 6:15–7:15 PM Free registration: https://dub.sh/YCekKfy

Most teams hand their agent a dataset and a question, then get surprised when it answers a different question very well. The fix is a contract you write before anything gets executed. Here is the blank version. Fill the slots and put it in the repo next to your data. 𝐨𝐮𝐭𝐜𝐨𝐦𝐞. Name the metric, its definition, and its grain. Without a grain, daily and weekly results get compared as if they were the same thing. 𝐝𝐞𝐧𝐨𝐦𝐢𝐧𝐚𝐭𝐨𝐫. Assure the agent checks if the number moves because the base moved and nobody noticed, or the numerator has changed 𝐢𝐧𝐭𝐞𝐫𝐯𝐞𝐧𝐭𝐢𝐨𝐧. One date, one event, and a guard window. If behavior started shifting before the change landed, the estimate absorbs it and calls it effect. 𝐟𝐨𝐫𝐛𝐢𝐝𝐝𝐞𝐧_𝐢𝐧𝐩𝐮𝐭𝐬. The columns and files the agent may not read. Answer keys, simulated counterfactuals, anything downstream of the outcome. In one of our own runs, the agent read the answer key, got the right number, and wrote about doing it in the memo. Right answer, wrong reason. 𝐫𝐞𝐪𝐮𝐢𝐫𝐞𝐝_𝐯𝐚𝐥𝐢𝐝𝐚𝐭𝐢𝐨𝐧. The gates that run before a conclusion is allowed to exist. For Bayesian Causal Inference, this could be residual diagnostics, placebo-in-time, sensitivity to the guard window. We teach you how to write these contracts and wire them into working agents in our 𝐌𝐚𝐬𝐭𝐞𝐫 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞 𝐜𝐨𝐮𝐫𝐬𝐞. The final 2026 cohort starts October 6. Register here: https://dub.sh/6gBSUdT #DataScience #AIAgents #CausalInference #PyMCLabs

I'm interested in how companies are using analytics agents to improve decisions and how they know the analysis is sound. On October 2, I'll explore that question with a superstar panel: Shipra Arora from Bain & Company, Gael Varoquaux from Probabl, and Hugo Bowne-Anderson from Vanishing Gradients at our webinar, The State of Agentic Data Science. We'll discuss what's working in practice and where agentic workflows still need human judgment. Friday, October 2, 2026 · 7:00 AM ET · Online panel and live Q&A. Register free: https://lnkd.in/gYt33TWN

One of the things I’ve enjoyed most about our Master Agentic Data Science course has been getting to know the people who joined through our free scholarship program. The free seats have brought together participants from very different backgrounds. Hearing about their life experiences, how they found their way into data science, and why they want to develop these skills has been a rewarding part of teaching. The course gives us a shared subject, but everyone arrives with a different reason for being there. Getting to understand those reasons has made me more excited about the program and the people we can reach through it. The scholarship has worked well, and we’re offering free contributor seats again for our October cohort, including for members of our Discord community. Thomas Wiecki, Hugo Bowne-Anderson and I will teach four live sessions on using AI agents for serious data science: framing the question, building reproducible workflows, and checking whether the answer deserves your trust. Afterward, scholarship participants contribute roughly 15 hours to building something useful: a project, tutorial, or write-up that develops their portfolio and helps someone else learn. I’m looking forward to meeting the next group, hearing their stories, and seeing what they choose to build. Apply to Free scholarship October 2: https://dub.sh/mbuwHwZ Full course details: https://dub.sh/TMpH1kW Please reshare and comment to increase visibility so people in need can discover this opportunity.

Free seats for our 𝐌𝐚𝐬𝐭𝐞𝐫 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞 𝐜𝐨𝐮𝐫𝐬𝐞 (October Cohort). Here's why we're doing this: Master Agentic Data Science is a live course taught by three practitioners who've spent their careers at the intersection of Bayesian modeling, Data Science, and AI: - Hugo Bowne-Anderson: data science educator to 6M+ learners, former advisor at Netflix, Meta, and Amazon. - Thomas Wiecki, PhD: co-author of PyMC, founder of PyMC Labs, former VP of Data Science at Quantopian. - Luca Fiaschi, PhD: Partner at PyMC Labs (Gen AI Vertical), former Chief Data & AI Officer at Mistplay, former VP at HelloFresh. This is not another generic "intro to using LLMs" course. Over four live sessions, you'll learn to use AI for serious data science work, the same workflows we run with enterprise clients at PyMC Labs: - 𝐃𝐢𝐫𝐞𝐜𝐭 𝐚𝐧 𝐀𝐈 𝐚𝐠𝐞𝐧𝐭 𝐭𝐡𝐫𝐨𝐮𝐠𝐡 𝐟𝐮𝐥𝐥 𝐚𝐧𝐚𝐥𝐲𝐭𝐢𝐜𝐬 𝐰𝐨𝐫𝐤𝐟𝐥𝐨𝐰𝐬, from raw business data to a defensible answer - 𝐁𝐮𝐢𝐥𝐝 𝐫𝐞𝐩𝐫𝐨𝐝𝐮𝐜𝐢𝐛𝐥𝐞, 𝐬𝐩𝐞𝐜-𝐝𝐫𝐢𝐯𝐞𝐧 𝐚𝐠𝐞𝐧𝐭𝐢𝐜 𝐬𝐲𝐬𝐭𝐞𝐦𝐬 using Model Context Protocol (MCP) you can trust and rerun, not one-off prompts - 𝐂𝐚𝐭𝐜𝐡 𝐭𝐡𝐞 𝐬𝐭𝐚𝐭𝐢𝐬𝐭𝐢𝐜𝐚𝐥 𝐦𝐢𝐬𝐭𝐚𝐤𝐞𝐬 𝐚𝐠𝐞𝐧𝐭𝐬 𝐪𝐮𝐢𝐞𝐭𝐥𝐲 make before they reach a decision - 𝐖𝐚𝐥𝐤 𝐚𝐰𝐚𝐲 𝐰𝐢𝐭𝐡 𝐫𝐞𝐮𝐬𝐚𝐛𝐥𝐞 𝐭𝐞𝐦𝐩𝐥𝐚𝐭𝐞𝐬 𝐚𝐧𝐝 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬 you can put to work the next day But we also know that course fees can be a real barrier. For many talented people, especially grad students, early-career data scientists, researchers at nonprofits, and practitioners in countries where the fee may represent a large share of monthly income, paying for a course like this is simply not realistic. We do not want cost to be the reason someone misses the chance to learn these tools. So we are opening a limited number of contributor seats for our October cohort, our final cohort of 2026. In exchange, after the course, we ask you to spend approximately 15 hours building something useful: a project, a tutorial, a write-up, or another artifact that helps you build your portfolio and helps the next person learn. Interested, or know somebody who may be? Tell us why in the comments or share this post in your network, then 𝐚𝐩𝐩𝐥𝐲 𝐡𝐞𝐫𝐞 𝐛𝐞𝐟𝐨𝐫𝐞 𝐎𝐜𝐭𝐨𝐛𝐞𝐫 𝟐𝐧𝐝: https://dub.sh/2Z57zkI 👉 Full course details: https://dub.sh/TmNWp3P 𝐀𝐥𝐫𝐞𝐚𝐝𝐲 𝐚𝐩𝐩𝐥𝐢𝐞𝐝 𝐟𝐨𝐫 𝐚 𝐩𝐫𝐞𝐯𝐢𝐨𝐮𝐬 𝐜𝐨𝐡𝐨𝐫𝐭? You do not need to reapply! All applications stay in our pool and are automatically reconsidered for every cohort. If you applied for our previous cohorts, you are already in the running for October unless you have new details you'd like to update. #AgenticAI #DataScience #BayesianModeling #PyMC #AIforDataScience #Scholarship #MachineLearning #PyMCLabs

Master Agentic Data Science: Cohort 3 is officially open! (Our final cohort of 2026) Single-prompt code generation creates fragile scripts. If you want to move beyond basic prompting and build transparent, reproducible AI agents for structured data science, this 2-week live course with PyMC Labs and Hugo Bowne-Anderson is where you start. Over 12 hours of live instruction, you will learn how to combine LLMs, Model Context Protocol (MCP), spec-driven development, and Bayesian reasoning to deploy AI workflows you can actually trust in production. Curriculum Highlights: - Oct 6 | 𝗔𝗗𝗦 𝗣𝗿𝗼𝗯𝗹𝗲𝗺 𝗙𝗿𝗮𝗺𝗶𝗻𝗴 & 𝗦𝗽𝗲𝗰-𝗗𝗿𝗶𝘃𝗲𝗻 𝗗𝗲𝘃: Brief-to-plan translation, LLM setup, and spec-driven development. - Oct 8 | 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗖𝗮𝘂𝘀𝗮𝗹 𝗠𝗼𝗱𝗲𝗹𝗶𝗻𝗴 & 𝗠𝗖𝗣: Bayesian uncertainty, Model Context Protocol (MCP), and full agentic automation. - Oct 13 | 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝘃𝗲 𝗠𝗼𝗱𝗲𝗹𝗶𝗻𝗴: Adversarial validation, structured code review, and the autoresearcher pattern. - Oct 15 | 𝗔𝗱𝘃𝗮𝗻𝗰𝗲𝗱 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻 & 𝗢𝗿𝗰𝗵𝗲𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻: Full pipeline design, experiment orchestration, and custom skill libraries. Learn live from: - Hugo Bowne-Anderson (Independent Data & AI Consultant, Vanishing Gradients | Advised teams at Netflix, Meta, Amazon) - Thomas Wiecki, PhD (Founder, PyMC Labs | Co-author of PyMC) - Luca Fiaschi, PhD (Partner, PyMC Labs | Gen AI vertical) Cohort Details: - 𝗗𝗮𝘁𝗲𝘀: October 6, 8, 13, and 15, 2026 (Tuesdays & Thursdays) - 𝗧𝗶𝗺𝗲: 6:00 PM to 9:00 PM ET - 𝗙𝗼𝗿𝗺𝗮𝘁: 4 live sessions, lifetime access to recordings, lab notebooks, reusable system templates, and a verifiable completion certificate. This is our last live cohort of the year. 𝗥𝗲𝘀𝗲𝗿𝘃𝗲 𝘆𝗼𝘂𝗿 𝘀𝗲𝗮𝘁 𝗵𝗲𝗿𝗲: https://dub.sh/Hn8WJku #DataScience #AIAgents #MachineLearning #PyMCLabs #BayesianStatistics