OpenAI Just Launched an AI That Never Stops Working

Morning, crowd.

Four stories breaking today:

• 🤖 OpenAI just launched "o" — an AI that never stops working — Persistent agents handling multi-step tasks, autonomous workflows, continuous execution. Sam Altman keynote at 10 AM PT. Live now.

• 🎬 Runway is now built into Adobe Premiere Pro and After Effects — Generate video, restyle footage, drop results straight into timeline. Professional video editing just went AI-native.

• 💻 Google launched Aluminium OS for laptops — Android-powered, Gemini-native, shipping fall 2026. Chrome OS is officially dead.

• 🏴 Every major AI lab just shipped persistent agents this month — ByteDance Seedance 2.5, Kimi K2.7, Project Genie, Microsoft Diagnostic AI, Apple Intelligence AFM. The agent race got real.

Here's what founders need to understand:

Read time: 5 mins

NVIDIA's Founder Says Farmers Should Absolutely Use AI

“If I were a farmer, I would absolutely use AI.” 

That’s Jensen Huang, founder and CEO of NVIDIA. 

And he's pointing to one of AI’s biggest untapped opportunities: Farming. It’s an industry facing mounting pressure to produce more with less and it’s still massively under-automated. 

, nutrition automation, and real-time data to livestock production, helping ranchers boost productivity and get more from every animal. 

And it’s already proven in one of the world's toughest livestock environments: 

• 500+ units deployed 

• 370,000 head of livestock on the platform 

• Up to 55% higher daily weight gain 

Now expanding into the U.S. and Brazil, is targeting a 300M+ head cattle market. And the biggest barrier to adoption? Gone. Ranchers get the technology for free when they sign up for a three-year nutrition plan. 

can scale adoption faster, which means more data and more recurring revenue. 

Invest before this early-stage opportunity gets harder to access.

^{𝘐𝘯 𝘮𝘢𝘬𝘪𝘯𝘨 𝘢𝘯 𝘪𝘯𝘷𝘦𝘴𝘵𝘮𝘦𝘯𝘵 𝘥𝘦𝘤𝘪𝘴𝘪𝘰𝘯, 𝘪𝘯𝘷𝘦𝘴𝘵𝘰𝘳𝘴 𝘮𝘶𝘴𝘵 𝘳𝘦𝘭𝘺 𝘰𝘯 𝘵𝘩𝘦𝘪𝘳 𝘰𝘸𝘯 𝘦𝘹𝘢𝘮𝘪𝘯𝘢𝘵𝘪𝘰𝘯 𝘰𝘧 𝘵𝘩𝘦 𝘪𝘴𝘴𝘶𝘦𝘳 𝘢𝘯𝘥 𝘵𝘩𝘦 𝘵𝘦𝘳𝘮𝘴 𝘰𝘧 𝘵𝘩𝘦 𝘰𝘧𝘧𝘦𝘳𝘪𝘯𝘨, 𝘪𝘯𝘤𝘭𝘶𝘥𝘪𝘯𝘨 𝘵𝘩𝘦 𝘮𝘦𝘳𝘪𝘵𝘴 𝘢𝘯𝘥 𝘳𝘪𝘴𝘬𝘴 𝘪𝘯𝘷𝘰𝘭𝘷𝘦𝘥. 𝘋𝘐𝘛 𝘈𝘨𝘛𝘦𝘤𝘩 𝘩𝘢𝘴 𝘧𝘪𝘭𝘦𝘥 𝘢 𝘍𝘰𝘳𝘮 𝘊 𝘸𝘪𝘵𝘩 𝘵𝘩𝘦 𝘚𝘦𝘤𝘶𝘳𝘪𝘵𝘪𝘦𝘴 𝘢𝘯𝘥 𝘌𝘹𝘤𝘩𝘢𝘯𝘨𝘦 𝘊𝘰𝘮𝘮𝘪𝘴𝘴𝘪𝘰𝘯 𝘪𝘯 𝘤𝘰𝘯𝘯𝘦𝘤𝘵𝘪𝘰𝘯 𝘸𝘪𝘵𝘩 𝘪𝘵𝘴 𝘰𝘧𝘧𝘦𝘳𝘪𝘯𝘨, 𝘢 𝘤𝘰𝘱𝘺 𝘰𝘧 𝘸𝘩𝘪𝘤𝘩 𝘮𝘢𝘺 𝘣𝘦 𝘰𝘣𝘵𝘢𝘪𝘯𝘦𝘥 𝘩𝘦𝘳𝘦: https://bit.ly/4bzuWCi​  ​}

FRESH NEWS

Getty Images

Sam Altman just keynoted at Fort Mason. The announcement: "o" an AI assistant that doesn't close after your chat ends.

Here's what that means: Instead of asking ChatGPT to "write a 10-part email sequence" and then waiting for the response, you tell "o" to handle it. The agent plans the emails, writes them, schedules them, monitors replies, and keeps working without you checking back. It's autonomous.

The architecture under the hood: "o" has memory (remembers context), scheduling (plans multi-step workflows), and permissions (knows what it's allowed to do in your systems). It's not a chatbot. It's a task executor.

This changes everything because most work is actually multi-step. Writing a pitch deck involves research + drafting + design feedback loops. Building a feature involves planning + coding + testing + deployment. These are agent workflows, not chat workflows.

The competitive signal: Every AI platform is shipping this today or next week. Kimi added agents in Sept 2025. Runway launched Agent in 2026. Project Genie went interactive in Jan 2026. Microsoft diagnostic AI is already live. Now OpenAI is validating it at scale.

Jose's take: The winners in this next phase won't be "best at chatting." They'll be "best at autonomous execution." If your AI can't remember context between sessions, schedule tasks, and execute without supervision, it's already obsolete. Every platform is converging on agents. The margin is in execution quality, speed, and integration depth — not model capability.

Adobe

This week, Runway Plugins went live in Adobe Premiere Pro and After Effects. Not as an export tool. Not as a separate window. As a native panel inside your edit interface.

What this actually means: A Hollywood editor can now highlight a clip, type "make this slower motion + color grade for sunset," and Runway generates the variation directly in the timeline. Takes 30 seconds. Used to take an hour of render time or a separate trip to After Effects.

Runway Gen-4.5 is powering it. Also available: Gen-3, Seedance 2.5 (ByteDance), Kling, MiniMax. Editors can now switch models mid-project without leaving Adobe.

The bigger story: Professional software just stopped being "features + plugins." It's becoming "AI + your interface." Adobe didn't build video generation. They became a distribution layer for Runway, ByteDance, and others. This is the pattern: whoever owns the interface wins, not whoever owns the model.

Real-world impact: Feature films shipping in 2027 will have AI-generated shots nobody notices. B-roll, color corrections, crowd effects, background reworking — all invisible, all seamless, all generated in-timeline.

Jose's take: If you're building professional tools, the game just shifted from "how do we add AI features?" to "how do we become invisible plumbing for AI models?" The winners aren't building new software. They're embedding existing AI into workflows builders already use.

Google

Google announced Aluminium OS this week, a new operating system for laptops, built on Android.

The feature that matters: "Magic Pointer" — when you shake your cursor over any text or image, Gemini gives you contextual suggestions. Screenshot + cursor shake = instant AI analysis of what's on screen.

This is not Chrome OS 2.0. This is Google saying "the operating system itself is now AI-first." Every interaction feeds context to Gemini. Every action Gemini can anticipate.

Shipping: Fall 2026, on custom Qualcomm silicon (Snapdragon AI Halo). This competes directly with Microsoft Project Zenith (local AI on Windows PCs).

The strategic move: Google is betting that builders will adopt a new OS if it bakes AI into the kernel, not just as a feature. They're right.

Jose's take: OS vendors just entered the AI infrastructure race. If Google ships Aluminium OS with better Gemini integration than Windows has with Copilot, developer mindshare flips. The "always-on AI assistant" is now a selling point for operating systems, not just models. Whoever wins OS distribution in 2027 owns the AI interface layer.

Topbots

One story got buried: this is not OpenAI innovation. This is industry convergence.

The timeline:

• Sept 2025: Kimi adds "OK Computer" (agentic AI)

• Jan 2026: Google launches Project Genie (interactive worlds)

• June 2026: Runway launches Agent (AI describes video, generates it, assembles it)

• June 2026: ByteDance Seedance 2.5 (generative video at scale)

• June 2026: Microsoft releases Diagnostic AI (beats doctors)

• Sept 2026: Kimi K2.7 Code ships

• Sept 2026: OpenAI launches "o"

The pattern: Persistent agents, autonomous execution, and real-time adaptation are not differentiators anymore. They're table stakes. The margin is in where agents run (on-device vs. cloud), how fast they execute, and what systems they can control.

Why it matters: Startups building "an AI agent for X" are already too late if X is generic. The wins will go to teams building agent infrastructure (scheduling, memory, permissions), agent integrations (connected to real systems), and agent vertical optimization (agents for specific workflows, not generic agents).

⚡ Quick Hits: The Ecosystem Moving

🎥 Runway Gen-4.5 now powers Agent mode — Describe a full video production in text, Agent plans shots, generates each one, assembles the final cut. Agentic video creation is production-ready.

🇨🇳 Seedance 2.5 available on Runway platform — Western builders can now use ByteDance's SOTA video model. Chinese labs shipping faster than US labs. Speed > brand loyalty.

🏥 Microsoft Diagnostic AI outperforms doctors 4x on complex cases — Diagnostic bottlenecks just became solvable by AI. Healthcare builders have a runway.

💻 Kimi K2.7 Code (June 2026) still shipping faster than OpenAI — One major release every 2-3 months. Moonshot AI proving speed matters more than brand.

🎮 Project Genie still breaking game studios — Interactive worlds from text prompts, real-time rendering. Game dev tools just got disrupted. Unity and Unreal updating playbooks.

🍎 Apple Intelligence AFM v3 released (June 2026) — Third-party devs get access to proprietary Apple foundation models via API. Apple is building its own model stack.

Hunt Accordingly

Three patterns define this week.

One: Persistent agents are industry standard now, not innovation. Every lab ships them. The winners won't be "best at agents." They'll be "fastest at new capability cycles" + "best integrations" + "vertical specialization." Speed beats smarts.

Two: Professional software is becoming an AI interface layer. Runway didn't disrupt Adobe; Runway became embedded in Adobe. The pattern repeats: whoever owns the interface wins, not whoever built the model.

Three: Chinese labs are shipping faster than US labs. Seedance 2.5 (May), Kimi releases (Feb, April, June), Project Genie agents (Jan). Speed compounds. By next quarter, builders stop picking by brand and pick by capability + latency. That rewards whoever ships fastest.

See you soon,

Jose, Alberto, and the Founderscrowd editorial team