Agents Unleashed

Welcome to Episode 16 of the Agents Unleashed Podcast, the show that helps you sort the signal from the noise in the fast-paced world of agentic AI.Hosted by Thomas Maybrier (Digital Media Producer, Valory), this week’s episode dives into three cutting-edge developments: the mainstream rollout of ChatGPT’s Agent Mode, a startling subliminal learning paper, and Bidhan Roy of Bagel AI on why true open source needs verifiable training data.In This Episode:ChatGPT Agent Mode Goes Mainstream: In-session memory, sandboxed computer tools, human-in-the-loop controls, and real-world task testsSubliminal Learning Paper: How a preference for “owls” can transfer via pure number sequences and what it means for hidden model biasesOpen Verification with Bagel AI: Bidhan’s take on balancing verification and monetization to build truly open AIResources & Links:ChatGPT Agent Mode announcement → https://openai.com/index/introducing-chatgpt-agent/“Subliminal Learning Language Models Transmit Behavioral Traits Via Hidden Signals” → https://arxiv.org/abs/2507.14805Bagel AI → https://bagel.aiBidhan's talk → https://www.youtube.com/watch?v=8VFfxKrw1DIZkLLora paper → https://arxiv.org/abs/2501.13965https://olas.network/blog/pod-surveyFollow Thomas on X: https://x.com/thomasmaybrier🎵 Theme music: “Forward” by Grand Project on Pixabay → https://pixabay.com/users/grand_project-19033897/Sponsored by Sigil Fund → https://www.sigilfund.com

Welcome to Episode 15 of the Agents Unleashed Podcast, the show that helps you find signal in the noisy world of agentic AI.Hosted by Thomas Maybrier (Digital Media Producer, Valory), this week’s episode explores the shifting landscape of open and closed AI ecosystems, with dispatches from Europe and Silicon Valley.This episode begins with a look at the Swiss AI Initiative and their upcoming sovereign open-source LLM, which boasts carbon-neutral training and support for over 1,000 languages. Then, a turn to Mistral’s Le Chat, a sleek, privacy-forward assistant with features rivaling GPT, built for European data compliance. Finally, checking in on Reflection’s "Asimov", a new multi-agent coding tool from ex-DeepMind team.In This Episode:- Switzerland’s sovereign open LLM (and why it matters for agent infra)- Mistral’s Le Chat: open-core speed, enterprise control, and EU-native compliance- Asimov: a coding-first multi-agent system built for team memory and code comprehensionQuick hits: OpenAI’s new agent features, Replit deletes itself, and LLM smart contract exploit generatorResources & Links:

Swiss AI Initiative announcement → https://ethz.ch/en/news-and-events/eth-news/news/2025/07/a-language-model-built-for-the-public-good.html

Mistral Le Chat → https://mistral.ai/news/le-chat-dives-deep

Reflection AI’s Asimov → https://reflection.ai/

Replit AI bug → https://x.com/jasonlk/status/1946069562723897802

OpenAI’s new agent features → https://openai.com/index/introducing-chatgpt-agent/

Smart Contract Exploit Paper → https://arxiv.org/abs/2507.05558Follow Thomas on X: https://x.com/thomasmaybrier🎵 Theme music: “Forward” by Grand Project on Pixabay: https://pixabay.com/users/grand_project-19033897/Sponsored by Sigil Fund → https://www.sigilfund.com

Welcome to Episode 14 of the Agents Unleashed Podcast, the show that helps you find signal in the noisy world of agentic AI.Hosted by Thomas Maybrier (Digital Media Producer, Valory), this episode dives into three recent developments from the frontlines of agentic AI: Grok’s controversial model behavior, Microsoft’s leap into medical diagnostics with agentic AI, and a timely discussion from Agents Unleashed Cannes on why protocols, not just frameworks, will define the next era of AI agent design.

If you’re curious about how these systems are evolving, what risks we need to manage, and how to stay informed in the chaos of AI hype, this episode has what you’re looking for.

In This Episode:Transparency gaps and political bias in Grok-4 by xAI

Microsoft’s MAI‑DxO outperforms doctors using multi-agent systems

Ron Bodkin on why protocols are foundational for agentic interoperability

What 50M+ health-related queries a day tell us about AI’s role in care

A few quick hits from the frontlines of open source and infrastructureResources & Links:Boaz Barak on Grok 4 → https://x.com/boazbaraktcs/status/1945165577154175288

Simon Willison on X → https://x.com/simonw/status/1945108916309746077

Read Matteo Wong in The Atlantic → https://www.theatlantic.com/technology/archive/2025/07/new-grok-racism-elon-musk/683515/

Read “The Path to Medical Superintelligence” → https://microsoft.ai/new/the-path-to-medical-superintelligence

Watch Agents Unleashed Cannes panels → https://olas.network/agents-unleashed

Mistral Voxtral → https://mistral.ai/news/voxtral

AWS goes in on agents → https://www.techradar.com/pro/its-a-tectonic-change-aws-ai-head-calls-agents-the-most-impactful-change-weve-seen-since-the-dawn-of-the-internet

GameNGen - DOOM as a neural diffusion model → https://gamengen.github.io/

Try Pearl → https://olas.network/operate

Follow Thomas on X: https://x.com/thomasmaybrier🎵 Theme music: “Forward” by Grand Project on Pixabay: https://pixabay.com/users/grand_project-19033897/Sponsored by Sigil Fund: https://www.sigilfund.com — performance-focused crypto investing since 2018.

Welcome to Episode 13 of the Agents Unleashed Podcast — the show that helps you find signal in the noisy world of agentic AI.Hosted by Thomas Maybrier (Digital Media Producer, Valory), this episode recaps standout moments from the Agents Unleashed event held in Cannes.Thomas highlights key presentations that embody the themes of ownership, action, and personalization in agentic AI. You'll hear insights from top builders at Olas, NEAR Foundation, and Giza on how agents are already transforming finance and how you can own and operate one today.In This Episode:- David Minarsch on Pearl's ownable agents and the Olas vision- Michael Broderick on NEAR’s Mindshare agent outperforming top tokens- Michael Beaudoin on the personalized Optimus agent inside Pearl- Renç Korzay on ARMA: a warden agent managing hundreds of millions in DeFi- Why personalization is the real alpha in agent designResources & Links:Watch the full Cannes playlist here

Try Pearl: https://olas.network/operateFollow Thomas on X: ⁠https://x.com/thomasmaybrier⁠🎵 Theme music: “Forward” by Grand Project on Pixabay: https://pixabay.com/users/grand_project-19033897/Sponsored by Olas, the best way to own and monetize your AI agents: https://www.olas.network

Can AI make crypto usable for everyone?

In Episode 12 of Agents Unleashed, Thomas Maybrier sits down with Sid Shekhar, co-founder of Ask Gina and former Coinbase Research lead, to explore how AI agents are transforming the crypto user experience.

Ask Gina is an intelligent wallet with a built-in agent that helps users navigate DeFi, execute transactions, and stay on top of market data. In this episode, Sid shares how the product uses a multi-agent architecture to support a wide range of users, from beginners to veteran degens.

You’ll learn:

Welcome to Episode 11 of the Agents Unleashed Podcast — the show that helps you find signal in the noisy world of agentic AI.Hosted by Thomas Maybrier (Digital Media Producer, Valory), this episode covers the latest frontier in AI agents — multi-agent systems — and how they’re already making an impact, from research workflows to human gatherings.In This Episode:Thomas shares Anthropic’s findings from building the multi-agent AI "Research" feature.

He also explores the AI Village experiment, where AI agents collaboratively raised funds for charity and organized a real-life event in San Francisco.

And finally, he shares a helpful tutorial that will teach you to building your own AI agent with tools from Coinbase's Agent Kit, The Graph and Vercel.Also covered:- Why multi-agent systems outperform single-agent setups- How Claude agents compress insights and share tasks - The flexibility of collaborative AI in unpredictable workflows- Ethical and social implications of agents organizing IRL eventsChapters01:20 Exploring Anthropic's multi-agent design for Research05:40 AI agents organizing an in-person event from @AiDigest_09:55 Ethical questions around real-world coordination10:04 Tutorial: build an on-chain agent using tools from @coinbasedev, @graphprotocol and @vercel11:08 AI in the news: 3 headlines you might have missedAgents Unleashed: Cannes — June 30, 2025The event is free, but space is limited.🎟️ RSVP here: https://lu.ma/tp6rhykzResources & Links:How we built our multi-agent research system: https://www.anthropic.com/engineering/built-multi-agent-research-systemAI Village by AI Digest: https://theaidigest.org/villageTutorial: https://x.com/cryptomastery_/status/1931117128951992489Sponsored by Olas — the best way to own and monetize your AI agents.🧠 Learn more at https://www.olas.network🎵 Theme music: "Forward" by Grand Project on Pixabayhttps://pixabay.com/users/grand_project-19033897/💬 Like, subscribe, and drop a comment!📱 Follow Thomas on X: @ThomasMaybrier🚀 Run your own AI agent today → https://olas.network

Welcome to Episode 10 of the Agents Unleashed Podcast — the show that helps you find signal in the noisy world of agentic AI.

Hosted by Thomas Maybrier (Digital Media Producer, Valory), this episode features a wide-ranging conversation with Trent McConaghy, co-founder of Ocean Protocol and the Artificial Superintelligence Alliance (ASI).

In This Episode: Trent shares his journey, from early AI-for-chips design to launching Ocean Protocol, an early AI and crypto project. He explains his view on what a truly autonomous AI agent is, why they need sovereignty, and why blockchains are essential.

We also explore:

- What makes an agent “sovereign”, and why many agents today aren’t

- How Ocean Protocol evolved

- “Bitcoin as a life form” and what that could mean for AI

- Reasons blockchains matter for AI agents

- What’s next for AGI, ASI, and the future of brain-computer interfaces

Resources & Links:

Trent McConaghy: https://trent.st

Trent on X: @TrentMc0

Sponsored by Olas — the best way to own and monetize your AI agents.

Learn more at https://www.olas.network

🎵 Theme music: "Forward" by Grand Project on Pixabay: https://pixabay.com/users/grand_project-19033897/

💬 Like, subscribe, and leave a comment!

Follow Thomas on X: @ThomasMaybrier

Run your own AI agent today → https://olas.network

Welcome to Episode 10 of the Agents Unleashed Podcast — the show that helps you find signal in the noisy world of agentic AI.

Hosted by Thomas Maybrier (Digital Media Producer, Valory), this episode dives into where Web3 AI is actually working with insights from guest David Attermann, Head of Web3 Investments at M31 Capital.

In This Episode:

David Attermann breaks down why verifiable privacy, composability, and on-chain autonomy are more than buzzwords: they’re essential components of real AI agents.

He shares the investor's lens on what’s working, where real-world use cases are gaining traction, areas where Web2 adoption of crypto infrastructure is already underway.

We also explore:

- Why most AI agents aren't truly autonomous (yet)

- The role of trusted execution environments (TEEs)

- The evolution from speculative hype to practical, composable AI systems

- Web3’s unique edge in secure, economically-incentivized AI

Agents Unleashed: Cannes — June 30, 2025

The event is free, but space is limited.

RSVP here: https://lu.ma/tp6rhykz

Resources & Links:

David Attermann at M31 Capital: https://m31.ventures/

David on X: @atterX_

Sponsored by Olas — the best way to own and monetize your AI agents.

Learn more at https://www.olas.network

🎵 Theme music: "Forward" by Grand Project on Pixabay: https://pixabay.com/users/grand_project-19033897/

💬 Like, subscribe, and leave a comment!

Follow Thomas on X: @ThomasMaybrier

Run your own AI agent today → https://olas.network

Welcome to Episode 8 of the Agents Unleashed Podcast — your signal in the noise of agentic AI. Hosted by Thomas Maybrier (Digital Media Producer, Valory), this episode explores major developments from the leaders and projects shaping AI's future.

In This Episode:

Sam Altman (OpenAI) on the future of AI as a strategic business problem-solver

Dario Amodei (Anthropic) predicts up to 50% of entry-level white-collar jobs could vanish in 10 years

ByteDance’s Bagel: a powerful new open-source multimodal model

Zochi: the world’s first AI-generated scientific paper to be accepted by a top academic conference

Plus quick hits on Meta, the FDA’s generative AI tool, Builder.ai’s collapse, Yoshua Bengio’s LawZero, and more

Resources & Links:

• Sam Altman at Snowflake Summit 2025: https://www.businessinsider.com/sam-altman-betting-ai-agent-could-soon-become-your-coworker-2025-6
• Dario Amodei on Future of Work: https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic
• ByteDance’s BAGEL Multimodal Model: https://seed.bytedance.com/bagel
• Zochi's AI-Generated Research Paper: https://www.intology.ai/blog/zochi-acl
• Builder.ai Collapse: https://www.ft.com/content/67f0277a-10fd-463c-8946-af0be2b4028f
• FDA’s Generative AI Tool ‘Elsa’: https://www.fda.gov/news-events/press-announcements/fda-launches-agency-wide-ai-tool-optimize-performance-american-people
• Meta’s AI-Driven Ad Strategy: https://www.reuters.com/business/media-telecom/meta-aims-fully-automate-advertising-with-ai-by-2026-wsj-reports-2025-06-02/
• Yoshua Bengio’s LawZero: https://lawzero.org/en
• SecureMinions from Ollama: https://ollama.com/blog/secureminions

Run an AI agent that you own: https://olas.network/operate

🎟️ Don't Miss The Next Live Event! Agents Unleashed is also a global event series on agentic AI — the next event is June 30th in Cannes, France. RSVP now at https://lu.ma/tp6rhykz

Sponsored by Olas — the best way to own and monetize your AI agents.

Learn more at https://www.olas.network

🎵 Theme music: "Forward" by Grand Project on Pixabay: https://pixabay.com/users/grand_project-19033897/

💬 Like, subscribe, and leave a comment! Follow Thomas on X: @ThomasMaybrier

This is episode 7 of the Agents Unleashed podcast — your guide to filtering the signal from the noise in the fast-evolving world of agentic AI.

In this episode, host Thomas Maybrier explores two major developments in the world of agentic AI: Google DeepMind's groundbreaking Veo3 model and the open-source II-Agent framework from Intelligent Internet.

Thomas breaks down what makes Veo3 a leap forward in text-to-video generation—including its synchronized audio capabilities, pricing model, and implications for creators and the broader AI economy.

He then shifts to the decentralized side of the spectrum with II-Agent, a flexible, Claude-powered agent framework that recently outperformed competitors on the GAIA benchmark.

The episode covers use cases, architectural features, and how frameworks like II-Agent can plug into the Olas ecosystem via the SDK, the Mech Marketplace, or the Olas Accelerator.

Resources mentioned in this episode:

II-Agent Github: https://github.com/Intelligent-Internet/ii-agent

Olas Accelerator program: https://olas.network/accelerator

Google DeepMind Veo3: https://deepmind.google/models/veo/

🔔 Don’t forget: Agents Unleashed is also a global event series on agentic AI — the next event is June 30th in Cannes, France.

RSVP now at https://lu.ma/tp6rhykz

Sponsored by Olas — the best way to own and monetize your AI agents.

Learn more at https://www.olas.network

🎵 Theme music: "Forward" by Grand Project on Pixabay: https://pixabay.com/users/grand_project-19033897/