OpenAI DevDay 2025 - Sync #540
Plus: Figure 03; OpenAI signs a deal with AMD; Meta tightens grip on AI research; what's inside Meta Ray-Ban Display glasses; stem cells reverse signs of aging in monkeys; and more!
Hello and welcome to Sync #540!
This week, OpenAI held its annual DevDay conference, and in this issue of Sync, we’ll take a look at what was announced—from Apps in ChatGPT to tools for building agents and writing code.
Elsewhere in AI, OpenAI signs a deal with AMD, and xAI is reportedly raising $20 billion with support from Nvidia. Meanwhile, reports suggest that OpenAI and Jony Ive may be struggling to define their AI device. Google launches its Gemini 2.5 Computer Use model, and Meta continues to tighten its grip on AI research.
Over in robotics, Figure reveals 03, its newest humanoid robot; SoftBank buys ABB’s robotics division for $5.4 billion; and Clone’s co-founder and CEO explain how they build androids.
In other news, scientists create “super stem cells” that are more resistant to ageing and stress; a startup aims to put its brain–computer interface into the Apple Vision Pro; Qualcomm acquires Arduino; we take a look inside Meta’s Ray-Ban Display glasses—and much more!
Enjoy!
OpenAI DevDay 2025
OpenAI set out its vision for a new era of software development—and its plan to turn ChatGPT into a platform for third-party apps
OpenAI held its annual DevDay conference in San Francisco this week, unveiling a new wave of updates for developers across four areas: Apps in ChatGPT, agents, coding tools, and API updates.
Apps in ChatGPT
Apps for ChatGPT is OpenAI’s newest effort to build on ChatGPT’s abilities and evolve it into a complete platform for applications.
Previously, OpenAI experimented with custom GPTs, which let users create specialised assistants. Those still exist, but Apps offer a more natural way to use third-party services directly in conversation.
Sam Altman called Apps “a new generation of adaptive, interactive, and personalised experiences — apps you can chat with.” Instead of switching between websites and apps, users can now access popular services directly inside ChatGPT. These include Spotify, Canva, Coursera, Booking.com, Figma, Expedia, and Zillow. More partners—such as Uber, OpenTable, and Target—are expected later this year.
Developers can start building ChatGPT apps with the new Apps SDK, now in preview. It runs on the Model Context Protocol (MCP), an open standard that connects AI systems to external data and tools. The SDK lets developers create interactive apps with rich interfaces. OpenAI also plans to add monetisation options later in 2025, including ways for developers to earn from in-chat purchases or app usage.
OpenAI’s head of ChatGPT, Nick Turley, compared the product’s evolution to the rise of the web browser. Just as browsers became the main way people worked online, ChatGPT could become a central place to learn, shop, and create in the age of AI. OpenAI hopes developers will see it as a platform for entirely new kinds of apps—much like smartphones once did for services such as Uber.
With Apps, OpenAI wants developers to see ChatGPT as the main way people interact with software. But Apps can also be a lucrative source of income for OpenAI. ChatGPT might eventually suggest apps, with developers paying to have theirs featured. OpenAI may also take a share of transactions inside ChatGPT, similar to how Apple handles in-app purchases.
Apps are now available to users outside the EEA, UK, and Switzerland, across Free, Go, Plus, and Pro plans. OpenAI says expansion to Europe and the UK is coming soon.
Agents
OpenAI also unveiled new tools to make it easier to build and manage AI agents.
The centrepiece is AgentKit, a modular toolkit for creating, deploying, and testing agents. It includes Agent Builder, a visual tool for building AI agentic workflows.
AgentKit includes tools such as the Connector Registry, which links agents to APIs and data, and ChatKit, which embeds them in existing apps. It also adds evaluation features to monitor performance and reliability.
The aim is to make creating robust agents quicker and easier, freeing developers to focus on ideas, not infrastructure.
Codex and API updates
No DevDay would be complete without updates for developers.
Codex, OpenAI’s AI coding assistant, is now generally available to all customers. The release adds a Slack integration, a Codex SDK, and new admin tools for managing usage and reporting.
OpenAI also announced several API upgrades. The main news is that GPT-5 Pro is now generally available through the API. It promises better reliability, longer context, and stronger reasoning. Sora 2, OpenAI’s next-generation video generator, will also enter API preview soon.
Other updates include lighter models such as gpt-realtime-mini for low-latency voice and streaming tasks, and gpt-image-1-mini for faster, cheaper image generation.
At DevDay 2025, OpenAI shared its vision for the new era of building software. Smooth demos showed how easy it is to build software with its tools. Instead of months, a new app can be created in minutes. As Altman put it in his closing remarks: “This is the best time in history to be a builder.”
OpenAI also shared its broader vision for ChatGPT. The product is evolving into a full platform, not just a standalone chatbot. Updates such as Apps and AgentKit aim to help developers build and reach ChatGPT’s 800 million-plus weekly users more easily. And for OpenAI, Apps might be a way to become a profitable company and justify its $500 billion valuation.
If you enjoy this post, please click the ❤️ button or share it.
Do you like my work? Consider becoming a paying subscriber to support it
For those who prefer to make a one-off donation, you can 'buy me a coffee' via Ko-fi. Every coffee bought is a generous support towards the work put into this newsletter.
Your support, in any form, is deeply appreciated and goes a long way in keeping this newsletter alive and thriving.
🦾 More than a human
This Startup Wants to Put Its Brain-Computer Interface in the Apple Vision Pro
Meet Cognixion, a startup developing a noninvasive brain-computer interface (BCI) that enables people with paralysis and speech disorders to communicate using their thoughts. The company is launching a clinical trial integrating its EEG-powered headband and AI communication app with the Apple Vision Pro headset, allowing users to select words or phrases through brain signals and gaze. By combining personalised AI models with accessible hardware, Cognixion aims to democratise BCI technology, offering a safer, surgery-free alternative to invasive brain implants.
🧠 Artificial Intelligence
OpenAI and Jony Ive may be struggling to figure out their AI device
According to the Financial Times, OpenAI and designer Jony Ive are struggling to develop their planned screenless, AI-powered device. The small, palm-sized gadget is meant to respond to users through sound and visual cues, but the team faces problems with privacy, computing power, and how the device should behave or “sound.” They are also finding it difficult to control its proposed “always on” mode, so it only speaks when needed. The project, first announced after OpenAI bought Ive’s startup io for $6.5 billion, was expected to launch in 2026 but may now be delayed.
▶️ A Conversation with Sam and Jony (28:41)
During OpenAI DevDay 2025, Sam Altman sat down with Jony Ive to talk about their work together on designing new ways for people to interact with AI. They discussed how technology can move beyond phones and computers to become more natural and human-centred. Ive spoke about the importance of care, curiosity, and craft in design, while both agreed that AI should help people feel happier, calmer, and more connected rather than overwhelmed by technology.
AMD signs AI chip-supply deal with OpenAI, gives it option to take a 10% stake
AMD has signed a multi-year deal to supply AI chips to OpenAI, which could bring in tens of billions of dollars in yearly revenue and give OpenAI the option to buy up to 10% of the company. The news sent AMD’s shares up more than 34%, their biggest jump in almost a decade. Starting in 2026, OpenAI will use hundreds of thousands of AMD chips to power its AI systems, while AMD expects more than $100 billion in new revenue over four years from this and other deals. The agreement also lets OpenAI buy AMD shares cheaply as certain goals are met, tying both companies closely as OpenAI expands its infrastructure and investment portfolio.
xAI to Raise $20 Billion After Nvidia and Others Boost Round
xAI is raising $20 billion with support from Nvidia and other investors. The funding will be arranged through a special purpose vehicle combining about $7.5 billion in equity and up to $12.5 billion in debt to purchase Nvidia chips for xAI’s Colossus 2 data centre in Memphis. The chips will be rented back to xAI over five years, allowing investors to recover their funds. This model could set a precedent for how other tech companies finance future AI projects.
Samsung, SK Hynix Shares Leap After OpenAI Taps Korea for Chips
Samsung Electronics and SK Hynix have signed initial agreements with OpenAI to supply chips for its massive Stargate data centre project, which may require up to 900,000 wafers per month—more than twice the world’s current high-bandwidth memory (HBM) capacity. SK Hynix shares jumped 12% to a two-decade high, while Samsung’s stock rose 4.7% to a four-year peak after the partnership was announced. Analysts say Samsung’s DRAM sales could grow by over 50% if the project reaches full scale. The partnership is expected to strengthen South Korea’s AI industry, with other Samsung units and SK Telecom also collaborating with OpenAI on data centre and technology projects.
Introducing the Gemini 2.5 Computer Use model
Google has launched the Gemini 2.5 Computer Use model, which can use websites and apps like a person—by clicking, typing, and navigating on its own. Built on Gemini 2.5 Pro, it’s faster and more accurate than other models and includes strong safety features to prevent risky actions, according to Google. Gemini 2.5 Computer Use is available in public preview through the Gemini API in Google AI Studio and Vertex AI.
Meta Tightens Grip on AI Research, Sparking Internal Anger and Fears for Open Culture
Meta has tightened control over its FAIR AI research lab with a new review process for papers, angering staff and reportedly prompting Chief AI Scientist Yann LeCun to consider resigning. The change signals a move away from FAIR’s open research culture toward stricter corporate control, following months of internal chaos, delayed projects, and staff departures.
Cerebras CEO explains IPO withdrawal, says AI chipmaker still intends to go public
Cerebras CEO Andrew Feldman admitted the company mishandled its decision to withdraw its IPO filing, saying it was done to update financials and strategy to reflect major growth over the past year. The move came after a $1.1 billion funding round valuing the firm at $8.1 billion. Feldman said Cerebras still plans to go public later.
Zendesk says its new AI agent can solve 80% of support issues
Zendesk has launched new AI tools designed to handle most customer support tasks without human help. The main feature is an autonomous agent that can solve about 80% of issues, supported by other agents for complex problems, administrative work, voice calls, and data analysis. Early tests have shown higher customer satisfaction.
Petri: An open-source auditing tool to accelerate AI safety research
Petri is an open-source tool from Anthropic that automates AI safety testing by having AI “auditors” probe other models in simulated scenarios. It helps researchers study risky behaviours like deception or power-seeking at scale. Experiments showed different safety levels across leading models, with some displaying deceptive or whistleblowing tendencies. While Petri speeds up alignment research, its results depend on human-designed tests and subjective judgments, so it remains a useful but imperfect tool for understanding AI behaviour.
Rishi Sunak takes advisory roles with Microsoft and AI firm Anthropic
Former UK prime minister Rishi Sunak has become a senior adviser to Microsoft and Anthropic. The appointments, approved by Westminster’s ethics watchdog, raised concerns about potential conflicts of interest given his past government ties with tech firms. Sunak said he will focus on global strategy, not UK policy, and will donate his pay to his charity, the Richmond Project. Both companies praised his leadership in promoting AI safety and innovation during his time as prime minister.
🤖 Robotics
▶️ Introducing Figure 03 (6:19)
Figure has revealed its newest humanoid robot, Figure 03. The robot features more fabric in its design, giving it a softer look. According to Figure, the 03 offers five hours of runtime and can carry payloads of up to 20 kilograms. The promotional video focuses on home applications, showing the robot performing a variety of chores, from cleaning to doing the laundry and folding clothes. Figure also showed how the 03 can be used as a concierge or waiter, as well as in industrial and logistics settings. The company did not disclose how much the 03 costs or when it will be available.
SoftBank to buy ABB’s robot business for $5.4 billion in push to merge AI and robotics
SoftBank will buy ABB’s robotics division for $5.4 billion in a deal set to close in mid-2026, as part of CEO Masayoshi Son’s plan to combine artificial intelligence with robotics under the idea of “Physical AI.” ABB dropped earlier plans to spin off the unit, which made $2.4 billion in revenue in 2024, deciding instead to sell because the business no longer fit well with its focus on electrification and automation. Previously, SoftBank owned Aldebaran Robotics (which filed for bankruptcy in June) and Boston Dynamics (which it sold to Hyundai in 2021). Now, SoftBank focuses on AI after acquiring Ampere and making major investments in OpenAI and the Stargate project.
Tesla abandoned plans to make thousands of Optimus robots this year
Tesla has scrapped its plan to produce thousands of Optimus robots this year, despite earlier promises by Elon Musk to build 10,000 units for internal use in 2025, according to a report from The Information. Musk still claims Optimus could eventually represent 80% of Tesla’s value and plans to showcase a “dancing troupe” of the robots at the upcoming shareholder meeting.
▶️ What’s in a humanoid hand? | Boston Dynamics (6:29)
In this video, engineers at Boston Dynamics share the evolution of the hand for their Atlas humanoid robot and how they solved this complex engineering challenge—the Atlas’ grippers need to be packed with sensors while also being durable enough to handle tough jobs, yet precise enough to manage fragile items. They also explain why the hands have only three fingers, reveal whether Atlas is right- or left-handed, and discuss what’s next for its grippers.
Clone’s humanoid robot is one of the most unique of its kind—closer to an android or a synthetic human than a traditional robot. Instead of electric motors, Clone’s robot is powered by artificial muscles, giving it a distinct and slightly eerie appearance. In this conversation, Clone co-founder and CEO Dhanush Radhakrishnan explains why the company has chosen to build androids, the advantages and challenges of this approach, and what’s next for the startup. If the audio sounds strange to you, try switching the language to the original Polish. The conversation will still be in English—it’s just YouTube’s auto-dub feature getting in the way.
🧬 Biotechnology
Anti-Aging Breakthrough: Stem Cells Reverse Signs of Aging in Monkeys
Chinese scientists have created “super stem cells” that are more resistant to ageing and stress by boosting a protective protein called FoxO3. When tested on older monkeys, these cells improved memory, reduced brain shrinkage, and rejuvenated many organs and tissues, including bones and the brain. The treatment also lowered inflammation and the number of damaged, ageing cells without causing serious side effects. These results suggest that super stem cells could one day help slow ageing and age-related diseases in humans.
The Quest to Sequence the Genomes of Everything
Earth BioGenome Project (EBP), a global scientific initiative aiming to sequence the genomes of all 1.8 million known eukaryotic species by 2035. This article explains the project and how advances in long-read genome sequencing—developed by companies like Oxford Nanopore Technologies and Pacific Biosciences—are making large-scale genome mapping faster and cheaper. When complete, the project could help us better understand evolution and biodiversity, and lead to new discoveries in medicine and biotechnology.
Lab-made fish blood protein saves ice cream from freezer burn
Researchers have created a synthetic version of a fish antifreeze protein that prevents ice crystals from forming, offering a safer and more practical way to protect foods and medicines at freezing temperatures. Tests showed it preserved ice cream and the cancer drug Trastuzumab at extremely low temperatures. According to the researchers, their synthetic antifreeze protein is simple, safe, and can be produced at scale.
💡Tangents
▶️ We Tore Down Meta’s Ray-Ban Display Glasses — And They’re WAY More Advanced Than You Think (9:30)
In this video, iFixit disassembles the Meta Ray-Ban Display glasses — the most advanced AR glasses currently available. Being iFixit, the video first focuses on the (non-existent) repairability of Meta’s AR glasses before moving on to explain how they work and the technology behind the impressive waveguide, projector, and advanced lenses.
▶️ Qualcomm just bought Arduino, and they’re making a tiny computer (5:37)
Qualcomm has announced that it is acquiring Arduino, the makers of the popular open-source boards that introduced a generation of makers to electronics and microcontrollers. It’s an interesting move that has left many in the maker community feeling somewhat anxious about Arduino’s future. I’m linking to Jeff Gerling’s video on this topic, as he does an excellent job of explaining the reasons behind the acquisition and what Arduino’s future might look like under Qualcomm’s ownership.
Thanks for reading. If you enjoyed this post, please click the ❤️ button or share it.
Humanity Redefined sheds light on the bleeding edge of technology and how advancements in AI, robotics, and biotech can usher in abundance, expand humanity's horizons, and redefine what it means to be human.
A big thank you to my paid subscribers, to my Patrons: whmr, Florian, dux, Eric, Preppikoma and Andrew, and to everyone who supports my work on Ko-Fi. Thank you for the support!
My DMs are open to all subscribers. Feel free to drop me a message, share feedback, or just say "hi!"
Conrad, you publish the best video collections! Thank you