Claude Cowork reports for work - Sync #554
Plus: Apple chooses Gemini; ads in ChatGPT; Claude for Healthcare; China’s second AI tiger goes public; CRISPR pioneer launches a new company; Walmart and Wing expand drone delivery services; and more
Hello and welcome to Sync #554!
It’s a rather big issue this week. We will take a closer look at Claude Cowork—Claude Code for the rest of your work, as Anthropic describes it. Other than that, it feels like everyone was waiting until the second week of the year to make announcements and releases.
Apple has confirmed that it will use Gemini to power its AI features, including an upgraded Siri. OpenAI has officially confirmed that ads are coming to ChatGPT. Meanwhile, Anthropic—apart from launching Claude Cowork—has introduced Claude for Healthcare and Labs, and cracked down on unauthorised usage of Claude by third parties. We also have Google launching tools for agentic commerce and Personal Intelligence in the Gemini app, as well as an update to the recent Grok controversy.
Over in robotics, Walmart and Wing plan to expand their drone delivery service by adding it to 150 more Walmart stores over the next year. 1X has launched a new world model for robotics, and Skild AI has raised about $1.4 billion to build a brain for general-purpose robots.
This week’s issue of Sync also features a new company founded by a CRISPR pioneer, Nvidia investing in Eli Lilly, OpenAI investing in Sam Altman’s neurotech startup, why 2026 may be the year of mega IPOs, and more!
Enjoy!
Claude Cowork reports for work
Claude Code reimagined how we write code. For many, this tool transformed the way they create software. Some went further and started applying Claude Code to tasks beyond coding. Anthropic noticed that and created Claude Cowork—Claude Code for the rest of your work, as the company calls it.
Claude Cowork works in the same way as Claude Code. You give it access to your computer, define the task, and then Claude prepares a plan on how to solve it, presents it and, once accepted, executes it. Unlike Claude Code, Cowork is designed with non-technical people in mind. Instead of working with a command line interface, which can be intimidating for many people, Anthropic's new product comes with a graphical interface that is easy to use and understand.
What Claude Cowork can do
According to Anthropic, Claude Cowork excels at tasks that involve file management and knowledge work. You can point Claude at folders on your machine, and it can read, create, edit, and organise files directly. This includes organising downloads, batch renaming files with consistent patterns, and converting file formats.
Beyond simple file operations, Cowork can also handle more complex tasks. It can create Excel spreadsheets with working formulas and conditional formatting, generate PowerPoint presentations from rough notes or meeting transcripts, and produce polished reports from scattered information. The tool can also synthesise research across multiple sources, extract themes and action items from meeting notes, and analyse your personal knowledge base to surface patterns and connections you might have missed.
For particularly complex work, Cowork can break tasks into subtasks and coordinate parallel workstreams using sub-agents. Each sub-agent starts with fresh context, avoiding the accumulation problems of long chat threads. This architecture allows Cowork to handle long-running tasks without hitting context limits or conversation timeouts. You can describe a task, step away, and come back to finished work.
Cowork becomes more powerful when paired with other Claude features. If you’ve set up Model Context Protocol (MCP) connectors, Cowork can access external information from your connected tools—whether that’s searching through Slack messages, pulling data from Google Drive, or accessing other services you’ve linked to Claude.
When combined with Claude in Chrome, Cowork gains the ability to complete browser-based tasks. It can read web pages, fill out forms, extract data from sites without APIs, and navigate across tabs.
The promise of automation
Tools like Claude Cowork can be very useful. Many people, including myself, have tasks that could be automated or wish they didn’t have to do them manually. This is where Cowork comes in. Instead of writing scripts, putting different tools together with orchestration software, or doing the work manually, Cowork can do it for you. At least that’s the promise.
I tested Cowork, and I can see myself using this tool. I started with the example task of cleaning up the desktop. Cowork did the job, although not exactly as I expected. For example, it put my notes on GPT-5.2 into a Gaming & Entertainment folder, when those notes have nothing to do with gaming or entertainment. Cowork did a better job when I tested it on my Obsidian notes. There was specific work I wanted to do around reformatting some of my notes. It performed very well here, saving me from writing a script to do that.
Early reviews have been generally positive, though with caveats. WIRED described it as "the first agent that has really clicked" after testing numerous disappointing AI agents. However, other testers note there's a learning curve—the asynchronous workflow requires unlearning chatbot habits and getting comfortable with handing off tasks rather than supervising every step.
Current limitations
As a research preview, Cowork has some limitations. It’s currently available only on the macOS desktop app and doesn’t sync across devices or to web or mobile apps. Sessions stay entirely on your machine, which means faster iteration and direct file access, but also means you can’t share chats or artefacts with others.
Several Claude features aren’t available in Cowork yet. You can’t use it within Projects, it doesn’t have Memory capabilities to retain information across sessions, and it’s not compatible with the GSuite connector. The desktop app must remain open while Claude is working—closing the app ends your session.
Usage is another consideration. Working with Cowork consumes significantly more of the usage allocation than standard chatting with Claude. Complex, multi-step tasks are compute-intensive and require more tokens to execute. If you’re a Pro subscriber hitting usage limits, you might find yourself running out of capacity more quickly when using Cowork extensively.
For enterprise users, there are additional constraints. Cowork activity isn’t captured in Audit Logs, Compliance API, or Data Exports, which means it shouldn’t be used for regulated workloads. There are also no role-based access controls available yet.
Security considerations
The biggest concerns around Cowork centre on security. Anthropic has built defences against potential security threats, but agent safety remains an active area of development.
Cowork runs in a virtual machine on your computer, which provides some isolation from your main operating system. You must explicitly specify which folders Claude can access—if you don't grant permission to a folder, Claude cannot see it. This controlled environment offers some security benefits, but Claude does have real access to the files you permit it to use.
Cowork can read, write, and permanently delete files in folders you grant it access to. While the tool asks for permission before taking significant actions, there's always some chance Claude might misinterpret your instructions. Similar AI agents have accidentally deleted important data—in one case, removing an entire disk, in another, deleting a production database.
Given these risks, Anthropic strongly advises being cautious about granting access to sensitive information like financial documents, credentials, or personal records. The company recommends saving backups of critical files and creating dedicated folders with non-sensitive information for Claude to work with.
When using Claude in Chrome for browser-based tasks, there are additional risks. Like most AI agents, Cowork is susceptible to prompt injection attacks—hidden messages in online content that attempt to trick the AI and deviate it from its intended tasks. Anthropic warns that hidden code in websites could potentially steal data, inject malware into your system, or take over your system if you grant browser access. Users need to be particularly careful about which websites they allow Cowork to interact with.
These security risks aren’t new with Cowork, but this might be the first time many users are working with an advanced tool that moves beyond simple conversation and can take actual actions on their computer. The combination of file system access and potential for destructive actions means users need to understand the risks before diving in.
Availability and future plans
At the moment, Claude Cowork is in research preview and available for Claude Pro and Max subscribers. Additionally, the new product works only on macOS. Anthropic is releasing it early because the company wants to learn what people use it for and how it could be improved. If this beta program for enthusiasts proves successful, we could see Cowork become more widely available and expand to Windows and other platforms. Anthropic has indicated plans to add cross-device sync and continue making safety improvements as they gather feedback.
Final thoughts
Overall, Claude Cowork is an interesting demonstration of what using AI tools could look like in practice. Claude Code has reimagined how people work with code, and Cowork has the potential to do the same for all kinds of knowledge work tasks. I wouldn’t be surprised if we see a similar tool coming from OpenAI or Google soon. While it has limitations and security concerns that users need to understand, it represents a step toward AI agents that can actually handle real work, rather than just generating text in a chat window.
If you enjoy this post, please click the ❤️ button and share it.
🦾 More than a human
OpenAI Invests in Sam Altman’s New Brain-Tech Startup Merge Labs
OpenAI has invested $252 million in Merge Labs, a new neurotech company cofounded by OpenAI CEO Sam Altman. Unlike implant-based systems such as Neuralink, Merge plans to use non-invasive methods like ultrasound and AI to read and influence brain activity, with a long-term goal of bridging biological and artificial intelligence to maximise human ability, agency, and experience.
🧠 Artificial Intelligence
Apple picks Google’s Gemini to run AI-powered Siri coming this year
Apple has entered a multiyear partnership with Google to use Gemini to power future AI features, including a long-awaited Siri upgrade. In a short joint statement, the companies said Google’s Gemini technology was chosen because it offers a strong base for Apple’s AI systems, while still running on Apple devices and its private cloud.
OpenAI to begin testing ads on ChatGPT in the U.S.
Ads are officially coming to ChatGPT. OpenAI announced that it will soon start testing ads in ChatGPT for adult free users in the US, while paid users will not see them. The ads will appear at the bottom of replies and will be clearly marked. OpenAI says ads will not affect answers, and it will not sell user data. OpenAI said ads will not influence answers, user data will not be sold, and feedback will be used to refine the experience.
Google Bets on AI-Based Shopping With New AI Agents for Retailers
Google has introduced new AI tools to help retailers use shopping and ordering agents as AI-driven commerce grows. Companies such as Kroger, Lowe’s and Papa Johns are testing the technology to personalise shopping, assist customers and make buying easier, while keeping control over their own platforms. Retailers can use the recently released Universal Commerce Protocol (UCP), which aims to standardise how AI agents interact with retailers’ systems, making it easier for automated assistants to browse products, check availability and complete purchases securely.
MiniMax, China’s second ‘AI tiger’ to go public, doubles in value in Hong Kong debut
MiniMax Group, the second of China’s so-called “AI tigers” to go public, had a strong first day on the Hong Kong stock market, with its shares doubling in price and valuing the AI startup at about $13.7 billion after raising $616 million to fund research.
Anthropic announces Claude for Healthcare following OpenAI’s ChatGPT Health reveal
Anthropic has introduced Claude for Healthcare, a new set of AI tools for healthcare providers, insurers, and patients. Like its rival ChatGPT Health, revealed last week, the product can sync health data from phones and wearables without using that data for training, but Anthropic says Claude offers more advanced features. While concerns remain about AI errors in medical advice, and both companies urge users to consult professionals, Anthropic sees strong demand as millions already use large language models to discuss their health. Meanwhile, following an investigation by The Guardian, Google disabled AI Overviews for certain health-related queries.
Gemini introduces Personal Intelligence
Google introduces Personal Intelligence, a new feature for Gemini that lets users connect Google apps like Gmail and Photos to get more helpful, personalised answers. Available as a beta in the US, it allows Gemini to find details from emails, photos and searches, suggest ideas, and help with everyday tasks, while giving users full control over what data is connected and keeping privacy and security at the centre.
Musk’s X to block Grok AI tool from creating sexualised images of real people
The latest Grok controversy continues as pressure grows on Elon Musk’s AI tool after it was used to create sexualised and fake images without consent. Since Monday, California’s attorney general has ordered xAI to stop allowing such content, while campaigners have called on Apple and Google to remove Grok from their app stores. In the US, Ashley St. Clair, the mother of one of Musk’s children, has filed a lawsuit claiming Grok generated explicit fake images of her and that the company failed to act. In the UK, the government has welcomed X’s move to block Grok from creating sexualised content of real people, although Ofcom’s investigation remains ongoing under new laws criminalising such material.
Microsoft on track to spend $500 million per year on Anthropic AI
According to a new report from The Information, Microsoft is deepening its ties with Anthropic, with spending on the company’s AI expected to reach about $500 million a year. Alongside this increased spending, Microsoft is also investing in Anthropic and integrating its AI into key products such as GitHub Copilot and Microsoft 365 Copilot.
Mira Murati’s startup, Thinking Machines Lab, is losing two of its co-founders to OpenAI
Mira Murati’s startup, Thinking Machines Lab, has seen two of its co-founders leave the company and return to OpenAI less than a year after it was founded. The departures come despite the startup raising $2 billion and being valued at $12 billion.
Taiwan will invest $250 billion in U.S. chipmaking under new trade deal
The United States and Taiwan have agreed on a trade deal that will see Taiwanese chip companies invest at least $250 billion in building semiconductor factories in the US, supported by loan guarantees from Taiwan. In return, the US will cut tariffs on Taiwan and offer exemptions for certain products, encouraging firms such as TSMC to expand US production while still operating in Taiwan. The deal aims to strengthen US chip manufacturing, reduce supply risks, and improve certainty for the semiconductor industry.
China blocks Nvidia H200 AI chips that US government cleared for export
China has blocked shipments of Nvidia’s H200 AI chips despite US approval for their export, according to a Financial Times report citing informed sources, though Reuters said it could not independently verify the claims. Chinese customs officials reportedly halted the chips at the border, prompting suppliers to pause production, while authorities also warned domestic firms against buying them. The situation adds to confusion in US–China tech relations, as the US government has imposed a 25% tariff on the chips by requiring them to pass through the US before delivery, a move that has further complicated sales to China.
Meta Unveils Sweeping Nuclear-Power Plan to Fuel Its AI Ambitions
Meta has signed major deals to secure nuclear power in the US to meet the soaring electricity demands of its AI data centres. The company will help fund new nuclear reactors and buy electricity from existing plants in Ohio and Pennsylvania. Although financial terms were not disclosed, the agreements are among the most ambitious yet between technology firms and nuclear providers, and aim to bring new power online by the early 2030s.
Anthropic cracks down on unauthorized Claude usage by third-party harnesses and rivals
Anthropic introduced new safeguards to stop third-party tools from pretending to be its official Claude Code client, which cut off tools like OpenCode that let users run heavy automated coding on flat-rate subscriptions. It also separately blocked xAI from using Claude models through Cursor, saying this broke its commercial rules. Many developers reacted negatively, saying the changes broke their workflows and felt unfriendly to customers, while others supported Anthropic, arguing it was fair to stop misuse of subscriptions and protect its systems.
Anthropic: Introducing Labs
Anthropic is launching Labs, a new team focused on testing and building early, experimental products that explore what Claude can do. The goal is to try ideas with users, learn what works, and then turn the best ones into reliable products.
Introducing the Raspberry Pi AI HAT+ 2: Generative AI on Raspberry Pi 5
Raspberry Pi has launched the AI HAT+ 2, a $130 add-on for Raspberry Pi 5 that enables local generative AI. It uses the Hailo-10H accelerator, includes 8GB of on-board memory, and delivers up to 40 TOPS of performance. According to Raspberry Pi, it can run LLMs, VLMs, and other GenAI workloads fully offline, keeping latency low and data private. However, the real-world performance is a bit mixed, as Jeff Geerling has found out in his review.
Anthropic invests $1.5 million in the Python Software Foundation and open source security
Anthropic has entered into a two-year partnership with the Python Software Foundation and contributed $1.5 million to support Python and its community. The funding will improve security for Python and PyPI, help protect users from malicious packages, and support the PSF’s ongoing work maintaining the language, infrastructure, and community programmes.
Games Workshop bans use of AI in its designs
Games Workshop has banned staff from using AI to create content or designs, saying it wants to protect the ideas and artwork behind its Warhammer games. In doing so, the company echoes the Imperium of Man from the Warhammer 40,000 universe, which also forbids the use of artificial—or “abominable”, as it calls it—intelligence.
OpenAI Partners with Cerebras to Bring High-Speed Inference to the Mainstream
OpenAI and Cerebras have announced a multi-year partnership to deploy 750 megawatts of Cerebras’ wafer-scale systems for AI inference starting in 2026. The capacity will come online in multiple tranches through 2028. While the companies did not disclose the deal’s value, the Wall Street Journal reports it is worth around $10 billion. Sam Altman is also a personal investor in Cerebras, and the two companies previously explored a potential partnership as early as 2017.
Global AI computing capacity is doubling every 7 months
Epoch.AI's new research has found that the total available computing capacity from AI chips across all major designers has grown by approximately 3.3x per year since 2022. Nvidia AI chips currently account for over 60% of total compute, with Google and Amazon making up much of the remainder.
▶️ Top 10 largest AI Datacenters in 2026 (11:49)
In this video, SemiAnalysis lists the top 10 AI data centres of early 2026, based on their power consumption and installed hardware. The video gives a clear sense of the scale of AI data centre build-up and just how massive these facilities are.
The Data Center Resistance Has Arrived
This article highlights growing opposition to AI data centre development in the US, with local communities raising concerns about rising electricity bills, heavy use of water and land, and limited benefits for residents. New research shows that this pushback has delayed or stopped tens of billions of dollars’ worth of projects and is increasingly influencing local politics and elections.
World ‘may not have time’ to prepare for AI safety risks, says leading researcher
David Dalrymple, an AI safety expert at the UK’s Aria agency, has warned that AI is advancing so quickly that there may not be enough time to make it safe. He says machines could soon perform most valuable work better and more cheaply than humans, while governments and the public underestimate the risks. Dalrymple argues that advanced AI systems cannot be trusted to be reliable and could cause serious economic and security problems unless urgent action is taken to control and reduce their potential harms.
Dutch couple’s marriage annulled due to ChatGPT speech
Another reminder to double-check responses from AI chatbots. A Dutch court annulled a couple’s marriage after the officiant used a light-hearted, ChatGPT-generated speech that did not meet legal rules. Even though the couple said “I do” and intended to be married, the ceremony did not include the required promise to take on marital duties. Because of this, the judge ruled that the marriage was not legally valid and ordered it to be removed from the city registry.
🤖 Robotics
Wing is bringing drone delivery to 150 more Walmart stores
Walmart and Wing plan to expand their drone delivery service by adding it to 150 more Walmart stores over the next year, with more than 270 drone delivery locations planned by 2027. Wing, which is part of Alphabet, has seen rapid growth and strong repeat use, while Walmart says drone delivery helps customers get last-minute items quickly. The service will roll out in several major US cities and will operate under US aviation rules, alongside Walmart’s partnerships with other drone companies.
Neo humanoid maker 1X releases world model to help bots learn what they see
1X has introduced a new physics-based AI system, the 1X World Model, designed to help its Neo humanoid robot learn new real-world tasks by combining video with prompts, even without prior training examples. While CEO Bernt Børnich claims this marks the beginning of self-teaching robots, the company acknowledges current capabilities are limited to simple household actions, though these are seen as an early step towards more complex behaviours as Neo prepares for a home release.
Humanoid and Schaeffler Announce Strategic Technology Partnership
Humanoid, a UK-based robotics and AI company, and Schaeffler have announced a five-year partnership to bring hundreds of humanoid robots into Schaeffler’s factories to support industrial automation. The first test deployments are planned for 2026–2027 to check performance, safety, and how well the robots work with existing systems, before wider rollout. The agreement also covers the supply and joint development of robot actuators, as well as data collection and skill training to improve robot capabilities for real industrial environments.
Robotics Startup Skild AI Valued Above $14 Billion in New Funding Round
Skild AI has raised about $1.4 billion in new funding, valuing the company at more than $14 billion—over three times its value seven months ago. Founded in 2023, Skild AI is building a general-purpose AI “brain” that helps robots learn and adapt to many tasks and environments by watching and practising, rather than being programmed for one job. Skild plans to use the funding to expand and train its software and deploy more robots across areas such as healthcare, warehouses and construction.
🧬 Biotechnology
Crispr Pioneer Launches Startup to Make Tailored Gene-Editing Treatments
Aurora Therapeutics is a new company cofounded by gene-editing pioneer Jennifer Doudna that wants to make personalised gene therapies available to more people with rare diseases. Using a new FDA approval pathway, the company plans to treat many different genetic mutations with the same underlying CRISPR technology, starting with the disorder phenylketonuria. This approach could make it faster and easier to develop treatments and help more patients who currently have few options.
Nvidia to Invest $1 Billion in AI Drug Laboratory With Eli Lilly
Nvidia will invest $1 billion over five years in a new joint laboratory with Eli Lilly in Silicon Valley to help speed up drug research using AI. The partnership aims to reduce the time and effort needed for lab work by using AI to automate some tasks, bringing together Nvidia’s technology and Lilly’s scientific expertise to push forward AI-based drug discovery.
Aging Weakens Immunity. An mRNA Shot Turned Back the Clock in Mice.
Scientists have developed an experimental mRNA treatment that restores immune function in older mice. The therapy uses mRNA injections to increase the production of healthy T cells, which are vital for fighting infections and cancer. Treated mice responded better to vaccines and cancer therapies, suggesting that age-related immune decline may be partly reversible, although the research is still at an early stage.
When Google Locked the Door, Three MIT Students Picked the Lock
Three MIT researchers were not happy when DeepMind closed off AlphaFold 3, so they created Boltz—a free, open-source alternative—and started a company around it. The article explains how Boltz plans to make money by helping pharmaceutical firms use and customise the models rather than selling the software itself.
The golden age of vaccine development
This article follows the history of vaccines, from their beginnings as a lucky accident to today’s carefully designed technologies. It shows how growing knowledge of germs, immunity and genetics turned vaccines from fragile, hard-to-make products into safer, more precise and faster-to-develop tools. It argues that we are in the golden age of vaccine development and that the future holds even greater breakthroughs, but only if we continue to invest in them.
💡Tangents
2026 May Be the Year of the Mega I.P.O.
2026 is shaping up to be a big year for IPOs, with OpenAI, Anthropic and SpaceX all preparing to go public. If these listings go ahead, they would rank among the largest ever and deliver a huge windfall for Wall Street and Silicon Valley after years of weak offerings. The deals could give public investors long-awaited access to the AI boom, while also testing whether these fast-growing companies can justify their sky-high valuations.
MIT’s pills that communicate from stomach integrates biodegradable radio frequency antennas
Researchers at MIT have created a pill that can signal when it has been swallowed. The pill contains a small, biodegradable antenna that sends a safe radio signal from the stomach, then breaks down in the body, while a tiny chip passes naturally through the digestive system. This technology could help doctors check whether patients are taking important medicines, especially for long-term conditions, and reduce health problems caused by missed doses.
Thanks for reading. If you enjoyed this post, please click the ❤️ button and 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!"