Sora 2, OpenAI AgentKit, and the Future of Agent Management — Claude 4.5, N8N, and the Battle for AI Orchestration
🎙️ Remy (Barcelona) and Emrick (NYC) are back, diving deep into the week’s most chaotic and fascinating AI news. From OpenAI’s Sora 2 launch to AgentKit, Claude 4.5, and the rise of agent orchestration, this episode covers the real state of AI workflows in 2025—where everything works a little too well, and nobody knows what to do while waiting for their agents to finish running.🎥 Sora 2 and the new video frontierRemy admits he was wrong: Sora 2 isn’t a flop—it’s a monster. The video quality blew expectations away, and the Cameo-style feature for generating your own likeness is shockingly good. But within a week, OpenAI added censorship layers, quantization, and lower-tier models. Emrick breaks down how OpenAI probably manages GPU budgets, scaling down from high-fidelity “Pro” models to cheaper quantized versions. They debate whether the hype cycle hides a quiet cost-optimization strategy, and how this shift mirrors the VO3 vs. Sora pricing war.🎧 AgentKit vs N8N — No-code wars get realOpenAI’s new AgentKit lets you visually build multi-step agents, but it’s a walled garden—GPT-only. Emrick argues that N8N, the open-source automation powerhouse with 150 K+ stars on GitHub, still wins for flexibility. They compare AgentKit’s flow builder to Zapier, Make, and Cursor’s background agents, asking whether OpenAI’s closed ecosystem can attract devs who already live inside GitHub and MCP-based frameworks. The key question: Can GPT’s smooth UX beat open-source extensibility?💼 Claude 4.5 and Anthropic’s B2B pivotThe hosts dissect Anthropic’s slick new Apple-style marketing campaign—“AI for the problem solvers.” Remy calls it “for the weirdos who actually build things.” They discuss why Anthropic’s B2B-first model might outlast OpenAI’s B2C focus, especially as OpenAI adds ads and app integrations into ChatGPT. Emrick predicts Anthropic will dominate enterprise contracts while OpenAI chases consumer traffic and revenue through embedded apps like Spotify, Booking, and Expedia.📱 The Apps SDK — GPT becomes a platformOpenAI’s Apps SDK introduces in-chat integrations. Think ChatGPT as a mini web, where apps surface inside conversations. Emrick questions whether this will become an ad layer—where apps pay OpenAI for top placement—creating a new “search monopoly” inside the chat UI.🧠 The Agent Management ProblemRemy and Emrick hit a nerve for every AI developer: what do you do when you press Enter? Real builders juggle 3–10 concurrent agents—Cursor, Claude, OpenAI—all running in different terminals. There’s no orchestration layer, no shared context, no “manager agent.” Emrick envisions a dashboard that acts like a CEO of agents, pulling reports, syncing objectives, and enforcing permissions via agent cards, A2A (agent-to-agent) protocols, and MCP routing. 🧩 Micro-Hard, Crew.ai, and the future of orchestrationThey revisit Elon Musk’s rumored “Micro-Hard” concept—companies that run without humans—and the forgotten hype of Crew.ai, once hailed as the next agentic revolution. The missing piece, they argue, is a unified protocol: something between MCP and Agents.md that defines how sub-agents communicate, report, and think. “Every 1 % model improvement scales the whole virtual company,” says Emrick. “We just don’t have the org chart yet.”🍔 Between calls: the developer’s dilemmaRemy jokes about running “tiny societies” inside his MacBook—dozens of branches, multiple AI threads, and cloud quotas always at risk. Together they imagine future tools that spin up isolated VMs for every feature branch, allowing multiple agents to build in parallel without conflicts.🧩 Mentions:Sora 2 · Sora 2 Pro · VO3 · Claude 4.5 · AgentKit · N8N · Cursor · Crew.ai · MCP · Agents.md · Apps SDK · Anthropic · OpenAI🎧 Hosted by Remy and Emrick — Watch Out for AGI, the AI podcast for solopreneurs#AI #LLMs #GPT4 #Claude #Cursor #Startups #Automation #SoftwareDevelopment