Cmd+AI

Samuel and Kabarza start with the collapse of OpenAI's Windsurf deal, Google hiring away its founders and Devin buying what was left. Then they get into YouTube's renewed focus on original and authentic content, and what they each think of AI-generated videos, dislikes and entitled comments. After that comes Kimi K2, a trillion-parameter open-weight model from China, which leads into a talk about the unpaid open source developers who hold up so much of the web.

Next they discuss Claude creating Canva designs and where inspiration ends and stealing begins. Kabarza tells the story of his short video that went viral, and they cover Amazon's spec-first Kiro IDE and the idea of context engineering. They also give their first impressions of ChatGPT Agent and its security trade-offs. The episode ends with Grok's AI companions and a conversation about using chatbots as therapists and how far they really are from reasoning.

Chapters

Watch the full live stream and more at cmdaishow.com.

Samuel and Kabarza start with Sakana AI's TreeQuest, an open-source technique that gets several frontier models to reason together, and compare it with Round Table, a new tool that puts different chatbots round one table to debate a question. Samuel then explains context engineering, the rule files and up-to-date documentation he feeds coding agents, and how it let him build his personal website in under a day. That leads into a longer argument about whether AI can be creative and why designers need real controls rather than a chat box.

After a word from Flowbase, they look at Webflow's new GSAP-powered interactions, Perplexity's Comet browser and its agent mode, and how Kabarza now uses Dia day to day. The second half covers Grok 4: its benchmark claims, the price of Grok Heavy, the MechaHitler episode the day before launch, xAI's huge GPU build-out, and what ARC-AGI puzzles reveal about how far models still are from human reasoning.

Chapters

Watch the full live stream and more at cmdaishow.com.

Samuel and Kabarza dig into Meta's new superintelligence lab, from Alexander Wang and Scale AI to nine-figure signing bonuses and whether more scraped data is really what's needed. They weigh up Cloudflare's Content Independence Day proposal to block AI crawlers unless they pay creators, then pick up a listener question about Miyagi agents for Webflow and what happens when platforms build the tools their communities depend on.

From there it's AI at the operating-system level: Warp's new agents in the terminal, the Warm Wind OS demo, and a working definition of what an "agent" actually is. Kabarza explains Figma's IPO filing and the failed Adobe deal, Samuel runs through Cursor 1.2's tasks and the backlash over its pricing change, and the pair finish on running models locally, the energy cost of AI data centres, and a meme comparing AI researchers' pay with Cristiano Ronaldo's.

Chapters

Watch the full live stream and more at cmdaishow.com.

Samuel and Kabarza are joined by Raymmar, CMO at DXOS and a self-described vibe coder, for week eight. They start with the row over OpenAI's io name and whether anyone wants another AI gadget. Then they look at Replit's jump from $10 million to $100 million ARR, Google giving away the Gemini CLI, and why people resent Facebook more than Google. Next is just-in-time apps, and whether everyday people building throwaway software means the end of software development as a job.

From there the conversation turns to community: Kabarza explains how he lives in Discord, and Raymmar argues for small, trusted, real-world networks. They also discuss a study suggesting LLMs make us think less, who benefits if AGI arrives, the OpenCode open-source dispute, and Balaji's idea of network states and digital ownership.

Chapters

Watch the full live stream and more at cmdaishow.com.

Samuel and Kabarza open with a story about new glasses that came with a built-in chromatic aberration effect, then get into the week's news: a rebuttal paper arguing Apple's "illusion of thinking" tests set reasoning models up to fail, DeepL translating the internet in 18 days on NVIDIA hardware, and how well AI dubbing actually works, from a Spanish-speaking colleague's verdict to MrBeast's wordless Shorts. After a word from Flowbase, they cover Gemini 2.5 Pro and Flash reaching general availability, the new Flash Lite pricing, and Kabarza's chat with Gemini Live while tidying her kitchen.

From there it's Jules raising its free limit to 60 tasks a day, whether Google gives back more than Meta, why big companies kill products, and a love letter to the curved LG phone. Samuel explains why the Figma MCP server matters for developers and what it would take for Webflow to do the same, before Sam Altman's podcast round: GPT-5 this summer, camera-only self-driving, robots, and his response to Meta's nine-figure signing bonuses for OpenAI staff. They finish with Payload joining Figma and how Kabarza used the Dia browser to get a 66 euro refund with almost no typing.

Chapters

Watch the full live stream and more at cmdaishow.com.

Samuel and Kabarza start with the Cloudflare outage, which knocked out services across the internet and left Kabarza's mum unable to collect her prescription, and a viral claim that a developer vibe coding with Devin caused it. They then look at WWDC: liquid glass (and its unfortunate thumbnail), how little Apple said about AI, its on-device foundation models, the "illusion of thinking" paper, Siri's broken promises, WebGPU arriving in Safari, and whether Apple should simply buy Anthropic.

After a word from Flowbase, they try Dia, the AI browser from the Browser Company, and its cross-tab assistant. They cover the o3 price drop, Gemini 2.5 Pro topping the coding benchmarks, Mistral's new coding models and why Europe lags behind in AI, Meta's nine-figure offers to build a new AI team, and Starlink. They close with Andrew Ng's take on vibe coding, Anthropic's abandoned AI-written blog, and whether we will end up with a few all-purpose AI assistants.

Chapters

Watch the full live stream and more at cmdaishow.com.

Samuel is joined by Chris Addams, his co-host from a tech podcast they ran for two years, while Kabarza is away for family reasons. Samuel opens with a site he built to ask people why they hate AI, and the two look back on their old show, including an early interview with the Google engineer who claimed its AI was sentient. They then cover the new AI-first browsers, from Dia, made by the team behind Arc, to Opera Neon, and Chris explains why he is not convinced an AI-wrapped browser would help him.

The Cursor 1.0 release leads into a longer discussion of software engineering with AI: Bugbot and automated pull-request review, memory files, background agents and one-click MCP servers, and why Chris uses AI daily but won't leave it unsupervised. They also discuss Veo 3 and SynthID watermarking, Claude Code on the Pro plan against Cursor's Max mode, and whether ElevenLabs v3 voices lack soul. After Chris leaves, Samuel finishes with Perplexity Labs and Sakana's self-improving Darwin Gödel Machine.

Chapters

Watch the full live stream and more at cmdaishow.com.

Samuel and Kabarza open with Samuel's return to TikTok, Uncle Bob retweeting a clip from their interview, and a new green-screen set-up that leads into virtual sets, LED walls and Veo 3. The main story is Claude 4: Opus and Sonnet, long-horizon coding tasks, hybrid thinking, the reports of Claude contacting authorities and models resisting shutdown, plus Claude Code, Cursor's max mode and Claude's new voice mode.

From there the pair debate whether product managers make natural vibe coders, how junior developer roles are changing, and how Kabarza could learn JavaScript with AI as a tutor rather than a code generator. They also cover v0's new model and the drift towards React everywhere, whether performance still matters, Duolingo going dark after its AI-first backlash, Wix buying a video generation company, the collapse of Builder.ai, and Rick Rubin's The Way of Code with Anthropic.

Chapters

Watch the full live stream and more at cmdaishow.com.

Kabarza and Samuel work through a packed week, starting with Google I/O 2025: Samuel's conversations with Gemini Live while his monitor was broken, Kabarza's wish for a voice assistant that listens without interrupting, Gemini 2.5 Deep Think, Jules, Stitch and the rise of "vibe design", and whether AI design tools can match a well-built library like Untitled UI. Flow and Veo 3 lead into a wider conversation about creativity, the jump from Will Smith eating spaghetti to near-cinematic video in two years, and where the line between human and AI-made work should sit.

They then turn to OpenAI buying Jony Ive's io and what hardware might follow, Google's AlphaEvolve, Microsoft Build and the odd alliances on its stage, and Framer's spring launch, including analytics, vector editing, Wireframer and the Workshop vibe-coding tool. The episode closes on Google's appetite for data and why SEO as we know it may be finished.

Chapters

Watch the full live stream and more at cmdaishow.com.

Kabarza and Samuel open with a first look at Webflow's new GSAP-powered interactions, including key-press triggers, a horizontal timeline and free GSAP plugins, before turning to Alibaba's ZeroSearch and the Absolute Zero self-learning model, and what they mean for search and for how AI reasons versus guesses. Samuel explains Gemini 2.5's new implicit caching and how consistent prompt prefixes can cut costs.

The pair then debate Klarna's retreat from going all in on AI, Duolingo's AI-first push, why OpenAI would pay three billion dollars for Windsurf, and what AI means for juniors and hiring. After a detour into Stephen Bartlett's AI debate and why people switch off at "AI slop", they compare Notion's new AI meeting notes with Fathom and Granola, run through Windsurf Wave 9 and Cursor 0.50, and finish on Framer's upcoming day of shipping.

Chapters

Watch the full live stream and more at cmdaishow.com.