😸 You can now build agents and apps inside ChatGPT

Welcome, humans.

Figure really does seem like the company to beat in humanoid robotics. Tesla Optimus certainly has scale, but what’s been shown so far from Figure’s robots is much more impressive. Apparently, the bots have been working 10 hour days for 5 months straight at BMW.

Interestingly, China apparently already has a “robot army” of 300,000 workers being deployed in factories to offset the aging workforce.

AI is amazing, huh? In one corner, you have actual working human robots that can be a drop in for human labor, and in the other corner, you have Stephan Hawking at the X Games.

Here’s what happened in AI today:

1. We break down the big launches at OpenAI’s DevDay.
2. AMD x OpenAI inked a megadeal (and infinite money glitch brrr).
3. DeepMind previewed CodeMender, a code security agent.
4. Senate report warned AI could automate ~100M US jobs.

Advertise in The Neuron here

OpenAI just launched an AI that builds its own AI tools, and now you can build agents and apps inside ChatGPT.

DEEP DIVE: Everything OpenAI launched at DevDay 2025, explained

Yesterday, OpenAI dropped a suite of developer tools that fundamentally changes how AI applications are built. The headline-grabber: their new AI coding agent, Codex, wrote 80% of their new Agent Builder tool in under 6 weeks. If you feel like AI is moving fast, this is why. The tools are now building themselves.

Here's the breakdown of the biggest announcements:

• Codex is now an AI co-worker. It's an AI agent that codes alongside you, writes its own tests, and even reviews pull requests. A new Slack integration lets you assign it tasks by just tagging @Codex.
• AgentKit is a no-code builder for AI agents. It features a visual canvas called Agent builder (video) where you can drag-and-drop components to create complex, multi-agent systems. Think n8n, but for building agentic workflows inside ChatGPT’s playground.
• Apps can now live inside ChatGPT. A new Apps SDK lets developers embed interactive UIs—like maps, order forms, or video players—directly into a chat (video), creating a seamless experience without ever leaving the conversation.

There’s also new models available over the API, including Sora 2 and GPT-5 Pro; we break it all down in the deep dive above!

The demos were absolutely wild. One engineer used voice commands to get Codex to build a complex app on the fly, controlling a professional camera and stage lighting (first with an xbox controller, then his voice) via functions coded in real time. Another built and deployed a production-ready customer service agent in just 8 minutes using the new visual Agent Builder. You can watch the full keynote here.

What this means for you:

The era of needing deep technical expertise to build powerful AI is ending. We are now officially in the “AI building AI tools” phase, and the implications are massive.

• For founders and creators, prototyping complex, agentic software just went from a multi-month engineering project to a weekend hackathon. You can now use Agent Builder to visually map out and test an idea for a multi-step AI workflow without writing a line of backend code. This dramatically lowers the barrier to entry for building the next generation of AI-native companies.
• For professionals, this signals a shift from using AI as a simple tool to becoming an architect of AI systems. Start thinking about your repetitive workflows: researching topics, creating reports, and managing projects as systems you can design. Your job isn't just to do the work, but to design the agent that does the work for you. Start by playing with Agent Builder to automate a small, personal task.

There’s tons of resources published with this release; we also synthesized it all together to create a handy tutorial walking you through the process. Check it out!

The bottom line is this: the speed of innovation is about to compound. The feedback loop of AI building better tools to build better AI is now in full swing. What was considered science fiction a year ago is now available via an API.

FROM OUR PARTNERS

Don't Miss PyTorch Conference 2025, October 22-23 in San Francisco!

PyTorch Conference is the premier forum for AI innovation. Join 2,500+ AI engineers and researchers for 6 technical tracks, 3 co-located summits, training & certification, a Startup Showcase, poster sessions + MORE. Whether you’re scaling a startup or advancing the state of AI, PyTorch Conference is your gateway to the people, tools, and insights driving the next era of innovation.

Save 50% off your attendee pass with code NEURON. Register today!

Prompt Tip of the Day

As part of today’s release, OpenAI published a Sora 2 prompting guide (and Sora 2 is now available via API) Here’s the TL;DR:

• Set the container in code: model, size, seconds—text won’t change these.
• Write one shot, one move, one clear action, timed to clip length (beats like t=1s, t=last).
• Use concrete visuals + 3–5 palette colors; reuse phrasing for continuity.
• Image reference pins character/wardrobe/set; Remix to change one thing at a time.
• Keep dialogue short in a separate section; label speakers.

And here’s a mini prompt template you can use:

Treats to Try

1. Sora is not the only AI video social network btw: Pika Predictive Video (featured above) turns a thought into a full clip on iOS; type a “vibe” (lol, we know) like “im rapping in a studio, fish-eye”, and it auto-writes the script, picks shots, lighting, music, and camera moves so you can crank out share-ready videos in seconds.
2. Speech AI provider Hume released Octave 2, which converts your text into natural-sounding speech in 11+ languages with under 200ms latency—try it here, use it in your apps via API here.
3. Monologue turns your voice into perfectly formatted text that matches how you'd actually write, i.e casual for friends, professional for your boss—free trial (1K words), then $100/year (early bird pricing, regularly $144/year).
4. Zo Computer gives you a personal cloud computer you control through chat to write, code, host websites, generate images, and automate tasks—all stored on your own server (credits 25% off right now btw).
5. Supermemory helps AI retain context between sessions with its universal memory API (founded by a 19 y/o and already raised $2.6M from Google, DeepMind, and OpenAI executives).
6. Nothing Playground transforms widget ideas into reality on Nothing phones using natural language prompts (raised $200M Series C).
7. Write It Down helps track expenses and incomes with a focus on written commitment for better financial accountability.

P.S: OpenAI also hosted a live podcast from DevDay featuring some partners talking about how they build with these tools, including:

• SchoolAI, which personalizes lessons to each student's learning pace and style, while creating lesson plans, rubrics, and assessments for you in minutes.
• JamDev, which records bugs with one click, capturing all the console logs and repro steps in a link you can instantly share (and they just announced a tool called “plsfix” which converts your point-and-click edits into the actual code changes for your dev team)
• Abridge, which records patient visits for doctors and writes clinical notes for you, freeing you from the keyboard during appointments.
• And Cursor, who you probably know by now (its the AI coding IDE) which autocompletes your code as you type and turns your descriptions into working software.

Around the Horn

A breathtaking chart.

• AMD partnered with OpenAI to provide 6 gigawatts of GPU computing power starting in 2026 (in exchange for a potential 10% of the company), sending AMD shares up 35%; Alex Kantrowitz & M.G. Sieglar break this down here.
• Rohan Paul says OpenAI and Jony Ive are working on a “palm-sized, screenless” assistant to be released in 2026, but uh, it’s got aways to go…
• Google DeepMind introduced CodeMender, an agent that automatically finds and fixes software security vulnerabilities, and has already submitted 72 security patches to open source projects
• A Senate Democrat report predicted AI could eliminate nearly 100M U.S. jobs in the next decade, with 89% of fast food and 64% of accounting roles at risk.
• Nintendo denied lobbying Japan's government over generative AI after a politician falsely claimed the company was doing so to protect its IP.
• Bank of England Governor Andrew Bailey called for a pragmatic UK approach to AI that solves risks rather than just broadcasting them.
• University of Liverpool researchers created a low-cost, AI-powered handheld blood test that detects Alzheimer's biomarkers early with hospital-grade accuracy.

FROM OUR PARTNERS

Elon Musk: “Robots Will…Do Everything Better”

And it’s already happening. Just look at fast food. Miso’s kitchen robots already fried 4m+ baskets of food for brands like White Castle. With 144% labor turnover and $20/hour minimum wages, they’re not alone. Partnered with NVIDIA and Uber’s AI team, Miso’s new robot sold out initial units in one week. Invest before Miso’s bonus shares change on 10/9.

Take Advantage While You Can

_{This is a paid advertisement for Miso Robotics’ Regulation A offering. Please read the offering circular at invest.misorobotics.com.}

A Cat’s Commentary