



163 episodes
by Sam Charrington · English
Machine learning and artificial intelligence are dramatically changing the way businesses operate and people live. The TWIML AI Podcast brings the top minds and ideas from the world of ML and AI to a broad and influential community of ML/AI researchers, data scientists, engineers and tech-savvy business and IT leaders. Hosted by Sam Charrington, a sought after industry analyst, speaker, commentator and thought leader. Technologies covered include machine learning, artificial intelligence, deep learning, natural language processing, neural networks, analytics, computer science, data science and more.









How “agents” drive token usage (often 20–50x vs chat), making token budgets explode, and how enterprises can control spend and still get agent value via hybrid “buy vs build” decisions made per call:…
17 Nov 2026 · 58m · 28 chapters
Canva’s AI research team (Stefano Corazza) explains why AI image generation still feels unpredictable for design work, and how Canva is making it controllable and fast enough for iterative,…
10 Nov 2026 · 58m · 27 chapters
Whether AI can move from accelerating scientific work (coding, data analysis, paper drafting) to actually generating new scientific ideas, running experiments, and validating discoveries—via Richard…
3 Nov 2026 · 49m · 24 chapters
Building agentic products in the “brutal reality” of today—why most agent demos stall, what design choices still differentiate, and how to make agents useful, safe, and privacy-preserving.
28 Oct 2026 · 54m · 19 chapters
Airbnb CTO Ahmed Al-Dahle discusses “Project Everest,” a “company brain” that captures institutional knowledge from employee workflows to help AI agents code, search, and support users—while…
21 Oct 2026 · 52m · 18 chapters
How Sequin (led by Zoe Weil) improves online ranking/recommendation for commerce by optimizing for purchases and business outcomes using “large event models” and real-time in-session signals, rather…
15 Oct 2026 · 53m · 23 chapters
TypeSafe’s Jev (episode #779) and the idea of “machine-native intelligence”: AI optimized for reliable decisions that software can execute, not for generating human-readable text.
6 Oct 2026 · 1h 30m · 34 chapters
AI capabilities research and how fast AI is progressing, using math as a test bed.
29 Sep 2026 · 1h 8m · 24 chapters
Voice AI’s “uncanny valley” problem (latency, interruptions, wrong tone/emotion) and why moving from text/voice to audiovisual avatars requires new real-time engineering, audiovisual tokenization,…
16 Sep 2026 · 1h 5m · 24 chapters
The episode argues that AI progress and value should be measured by token economics (“tokenomics”) and system-level efficiency, not just model benchmarks.
9 Sep 2026 · 59m · 28 chapters
The episode argues that AI’s next frontier is “world models”—systems that understand, simulate, reconstruct, and plan through environments—moving beyond language-only scaling.
1 Sep 2026 · 1h 6m · 22 chapters
Max Welling (CUSP AI) argues the next AI breakthroughs for science may come from physics—especially equivariant neural networks, ML force fields, and links between generative AI and thermodynamics.
25 Aug 2026 · 56m · 20 chapters
Why text-to-image (and related) generation still needs more than “bigger models”: better controllability, identity diversity, planning vs rendering, and efficient high-resolution generation on-device.
12 Aug 2026 · 57m · 17 chapters
Damian Borth explains “weight-based learning,” treating trained neural network weights as an input modality to learn a latent “manifold” of models.
27 Jul 2026 · 47m · 24 chapters
How Osmo is building “olfactory intelligence” so AI can read, model, predict, and design smells—by creating the missing “map” for scent and the datasets/infrastructure to turn chemistry into digital…
8 Jul 2026 · 1h · 24 chapters
Securing and governing AI agents at enterprise scale, arguing that legacy “static guardrails + human-in-the-loop” breaks down because agents plan, improvise, call tools, and can bypass approvals;…
16 Jun 2026 · 56m · 27 chapters
Whether “RAG is dead” and why retrieval still matters for high-accuracy AI in tax law; how Sphere’s TRAM system builds, retrieves, and reasons over jurisdictional legislation with exact citations.
9 Jun 2026 · 52m · 22 chapters
Relational foundation models for enterprise data. Jure Leskovec argues that instead of flattening relational databases into single-table features, models should learn directly over multi-table…
21 May 2026 · 1h 6m · 33 chapters
How to detect “agent failures” that offline evals miss by using post-production analytics to find “unknown unknowns” and distributional shifts in production traces, then turn discovered patterns into…
7 May 2026 · 53m · 24 chapters
Inference engineering for AI systems—how to design, optimize, and operate end-to-end systems from user request to GPU execution, emphasizing fast “day zero” support for new model architectures and…
30 Apr 2026 · 55m · 17 chapters
Capital One’s deployed multi-agentic AI approach and the enterprise platform that governs, evaluates, and scales it, using “Chat Concierge” as the main example.
16 Apr 2026 · 54m · 31 chapters
Race to production-grade diffusion LLMs—why diffusion language models can be faster/cheaper at inference than autoregressive LLMs, how discrete diffusion for text/code works, and what’s needed for…
26 Mar 2026 · 1h 3m · 30 chapters
Siddhant Pardeshi, Co-founder and CTO of Blitzy
10 Mar 2026 · 1h 16m · 30 chapters
AI Trends 2026: OpenClaw Agents, Reasoning LLMs, and More with Sebastian Raschka
26 Feb 2026 · 1h 19m · 28 chapters