Building Agents 101: Design Patterns, Evals and Optimization (with Sinan Ozdemir)

Want to be building and deploying agentic A.I., but don’t know where to start? Distilling the best bits of Sinan Ozdemir's latest book, today’s episode is an intro “101” course on everything you need to know!

Sinan’s latest book is his TENTH book and it’s called “Building Agentic AI”. It’s a hands-on book (in Python) and one of the first books in Pearson's "Jon Krohn A.I. Signature Series". It covers everything you need to know to design, fine-tune, optimize and deploy agentic systems effectively… and today’s episode distills all of the most valuable tips and tricks from the book into a fun, hour-long conversation, including:

• What exactly is an A.I. agent?

• Should you use an agent or a workflow to automate a given task?

• What LLM should you select for your agentic task?

• How can you evaluate your agent's performance?

In case you don't already know Sinan:

• As mentioned above, a TEN-TIME author of bestselling technical books.

• A top A.I. educator and content creator with Pearson (often within the O'Reilly platform), transforming the way people learn about and engage with LLMs.

• Provides advisory services on A.I. to VCs, publicly-traded companies and startups alike.

• Holds Master's in Pure Mathematics from The Johns Hopkins University where he also lectured on A.I.

• Founded one of the first Generative A.I. startups to go through Y Combinator, patenting the concept of tool use with agents in 2018.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Without Trusted Context, Agents are Stupid (featuring Salesforce’s Rahul Auradkar)

If A.I. models are so intelligent, why do they keep doing such stupid things? For my guest today, the answer is simple: They lack the context they need. Meet Rahul Auradkar, the EVP and general manager at Salesforce responsible for the firm's $7B data foundation. hashtag#ad

More on Rahul:

• Responsible for Salesforce's data foundation, which includes Tableau, MuleSoft, Data360... and now, the data-management solution Informatica!

• Leads Salesforce's push to integrate data and A.I. across all products.

• Previously spent 12 years at Microsoft and served as Chief Product Officer at several A.I. and analytics startups.

In this episode, Rahul covers:

• How effective A.I. grounding requires not just context but *trusted* context—meaning governance, lineage, consent management and real-time signals all working together.

• The components of Salesforce’s unified data engine and how the acquisition of Informatica brings enterprise data catalog, enterprise MDM, data lineage and data quality tools that complement the unified data engine’s ability to provide trusted context to A.I. models andagents.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

How AI Agents Are Automating Enterprise Data Operations, with Ashwin Rajeeva

Acceldata has raised over $100m to bring you self-healing data pipelines. Today, brilliant co-founder/CTO Ashwin Rajeeva explains how they detect errors, rewrite code and redeploy... all fully automatically!

Ashwin’s outstanding communication on this meaty technical topic, involving autonomous ("agentic") scouring over petabytes of enterprise data, makes for a tremendous episode that will probably appeal most to hands-on practitioners like software engineers and data scientists.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

From Agent Demo to Enterprise Product (with Ease!) feat. Salesforce’s Tyler Carlson

Building AI agents is easy... but only for demos. Getting agents from prototype to enterprise-grade production, that's tricky. Or, it was! ...Until today's guest, Tyler Carlson, helped open up Salesforce's new Agentforce 360 platform for entrepreneurs to build upon, customize and monetize. hashtag#ad

Tyler:

• SVP, Head of Product for AppExchange & Ecosystem at Salesforce, where he's been for nearly a decade.

• Focused on delivering a modern AI marketplace that allows folks to discover and install apps, integrations, agents and agent components from Salesforce’s open ecosystem.

• Holds a degree in math from UC Santa Barbara.

We partnered with Salesforce on this episode to announce the launch of the Agentforce 360 platform for businesses, which provides an easy way to white-label commercial AI agents and applications.

This episode will be of interest to hands-on AI practitioners (e.g., data scientists, software developers, AI engineers) as well as anyone who'd like to understand the challenges of deploying enterprise-grade agents and solutions.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Jon Krohn's AI Signature Series: First Two Books are Bestsellers

Two books have been published so far in Pearson's "Jon Krohn A.I. Signature Series"... and now both are near the top of Amazon's "Artificial Intelligence" bestsellers list!

Sinan Ozdemir's "Building Agentic A.I." (circled in purple) is in 9th while Sadie St Lawrence's "Becoming an A.I. Orchestrator" (circled in red) is in 11th.

Both books are excellent (as the Amazon reviews quantify) and they are complementary — I (of course!) highly recommend them both.

Nested Learning, Spatial Intelligence and the AI Trends of 2026, with Sadie St. Lawrence

The same level of A.I. capability that cost you $100 a year ago now costs ONE dollar. Such radically cheap intelligence changes everything about work and society; today, Sadie St Lawrence explains how!

For the fifth year in a row, we’re kicking the year off by welcoming the inimitable Sadie to the show to:

• Predict the five biggest trends in A.I. for 2026, including nested learning, spatial intelligence and AiOps.

• Recap how she did on her predictions for 2025.

• Bestow four awards for 2025 (biggest “wow” moment, comeback of the year, disappointment of the year, and overall winner).

• Get a glimpse at the year ahead, in which intelligence will be vastly cheaper than ever before... transforming work and play for all of us.

📚 Her brand-new book, "Becoming an AI Orchestrator", guides readers to work creatively and confidently alongside intelligent machines. It is part "Jon Krohn's A.I. Signature Series" published by Pearson.

More on Sadie:

• Founder and CEO of HMCI.AI, optimizing human-A.I. collaboration in the knowledge economy (amplified by an ecosystem partnership with NVIDIA).

• Founded Women In Data™️, a global non-profit spanning 55 countries and empowering 70,000+ data professionals; it earned Top 50 Non-Profit status and was named the premier Women in A.I. & Tech community in 2021.

• Named among DataIQ’s Top 100 Most Influential People in Data & A.I. and Dataleum’s Top 30 Women in A.I.

• Has educated over 700,000 learners through courses with University of California, Davis, Coursera, and LinkedIn Learning.

This episode can be enjoyed by technical and non-technical folks alike.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Recap of 2025 and Wishing You a Wonderful 2026

We had 104 new episodes in 2025. For the first one of 2026, I reflect on what my favorite episodes of the past year were and what we can look forward to in data science in the year ahead :)

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Beyond “Agent Washing”: AI Systems That Actually Deliver ROI, with Dell’s Global CTO John Roese

Today's blockbuster episode (on MCP, agent-to-agent protocols, getting an ROI on A.I. investment, and more) stars a blockbuster guest: Dell's global CTO and Chief A.I. Officer John Roese.

Last year, John helped Dell Technologies' revenue grow by $10 billion dollars while A.I. helped costs go DOWN. That decoupling of revenue and cost had not happened in Dell's 41-year history but now, thanks to A.I., it has and John tells us exactly how they did it... so you can too!

This is an exceptional episode that can be enjoyed by technical and non-technical folks alike.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

How to Avoid Burnout and Get Promoted, with “The Fit Data Scientist” Penelope Lafeuille

Feeling burnt out? Or looking to get promoted? Today’s episode (in time for New Year's resolutions!) with the "Fit Data Scientist" Penelope Lafeuille is packed with fitness and nutrition tips for thriving at work.

Penelope was a burnt-out data scientist who then worked to develop a more balanced body and mind, earning promotions and a job paying over $180k in annual salary. Hear her journey in today's episode as well as her key insights, which include:

• Working out every day without proper recovery is overtraining, not stress relief... four structured sessions per week with adequate rest is more effective.

• Eating one gram of protein per pound of body weight supports muscle recovery, and concentrating carbs around your workout prevents energy crashes during work hours.

• Quality sleep requires more than eight hours in bed, reading instead of scrolling before sleep, and brief walks throughout the day to help your nervous system downregulate.

• Writing down everything on your mind before bed tells your brain it can let go until morning... and gives you a ready-made task list when you wake up.

Previously of Citi and Crédit Agricole CIB, Penelope now works as a Senior Data Scientist at Medidata Solutions in San Diego. She holds a graduate degree in engineering from CentraleSupélec and a graduate degree in management science and engineering from Columbia University.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Context Engineering, Multiplayer AI and Effective Search, with Dropbox’s Josh Clemm

Context engineering, "multiplayer" A.I. and building effective A.I.-search products are some of the cutting-edge topics covered by Josh Clemm, VP of Engineering at Dropbox, in today's episode!

Think about how many search bars you use at work... Slack has one. Google Drive has another... Your email, your project management tool, your file storage. Josh reckons knowledge workers juggle as many as 20 search tools daily… and now he’s built the fix.

At Dropbox, Josh is focused on building Dash, an A.I.-powered search across every application you use for work. Based on his extensive experience (including 8 years leading 400 engineers at Uber), this is an enlightening episode across A.I., building effective products and cultivating productive teams.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Happy Holidays from All of Us at the SuperDataScience Podcast

Behind the scenes at the SuperDataScience Podcast, nine humans work to produce each of the 104 episodes we release annually. For today's episode, all nine of us recorded holiday messages for you 🌲

Thanks for listening all year 'round! The team wishing you gratitude for supporting us and wishing you peace this holiday season are:

• Kirill Eremenko: founder and original host

• Sonja Brajovic: operations manager

• Natalie Ziajski: partnerships manager

• Mario Pombo: media editor

• Dr. Zara Karschay: writer

• Serg Masís: researcher

• Anthony Gagnon: animator

• Boldiszár Mészáros: graphics

• Yours truly!

Listen to today's episode to get a glimpse behind the curtain at each of these humans, who are all exceptional at what they do. It's an honor to work alongside such clever and steadfastly reliable folks.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Why AI Keeps Failing Society, with Stanford professor Alex “Sandy” Pentland

The Soviet Union collapsed because of A.I. and today's guest, Stanford/MIT professor Alex 'Sandy' Pentland, argues society is on the same crash course today!

Prof. Pentland is a rockstar:

• Distinguished professor at both Stanford University and the Massachusetts Institute of Technology.

• Pioneered the field of computational social science.

• Many-time best-selling author... in today's episode, we focus on his latest book, "Shared Wisdom: Cultural Evolution in the Age of A.I."

We discuss:

• How A.I. systems fail when they treat humans as logical engines rather than social foragers who learn by watching others.

• The parallels between Soviet central planning through early A.I. and today's systems.

• Why Washington DC residents wanted A.I. to defend them against city bureaucracy rather than optimize city services, revealing what real people actually need from A.I.

• "Data unions" as the next evolution after labor unions and agricultural co-ops.

• Living laboratories and digital twins as tools for testing A.I. systems in real-world contexts before deployment.

• ...and much more!

Today's episode is relatively high-level so can be enjoyed by any interested listener.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

In Case You Missed It in November 2025

I learned a lot from these brilliant guests last month. ICYMI, today's episode features the best parts of my conversations with them:

1. Tyler Cox and Shirish Gupta explain what state-space LLMs (like those incorporating Mamba layers) are and what advantages they have over transformer-only architectures.

2. Dr. Vijoy Pandey, who leads development of Cisco's open-source platform for the coming "internet of agents", explains a paradigm that allows us to trust agents (from a privacy and security perspective).

3. Fabi.ai co-founder Marc Dupuis details how to navigate several key organizational metrics simultaneously so that teams don’t lose sight of their goals and, crucially, so that AI models stay aligned with us too.

4. Santa Clara University professor Maya Ackerman and I discuss the importance of keeping human interests and welfare at the centre of the AI conversation.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

How to Get Hired at Top Firms like Netflix and Spotify, with Jeff Li

Many folks dream of doing AI/ML at top firms like Netflix and Spotify. In today's episode, Jeffrey Li spills the beans on how he got hired at these companies... and how you can too!

More on Jeff:

• Data Scientist at Netflix, focused on ads forecasting and optimization.

• Spent the last eight years building ML systems for Netflix, DoorDash and Spotify, specializing in time-series forecasting.

• Outside of work, he's sports-obsessed, incl. surfing, snowboarding, running, cycling and BJJ.

Today's episode gets into the technical weeds on time-series analysis on massive scales so it will appeal especially to hands-on practitioners like data scientists, AI/ML engineers and statisticians. Lots of it, however, will be fascinating to any listener, including Jeff's general tips on getting hired at top firms and his experience launching an AI startup.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

AI is a Joke, with Joel Beasley

This is definitely the funniest episode of my podcast: Software-engineer-turned-comedian Joel Beasley explains how he uses A.I. to quantify and optimize his comedy, with a lot of laughs along the way.

Joel:

• Two decades of experience as a software developer, including several CTO positions.

• Hosts the popular "Modern CTO Podcast", a top half-percent podcast globally.

• Most recently, a comedian who took off professionally as the opening act for comedy big-wig Jeff Allen on a US-wide tour.

Today's episode should be accessible — and hilarious! — to any interested listener.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.