Get tomorrow's brief in your inbox
Today: Anthropic pushed Claude Cowork into full enterprise release with finance and HR plugins, making AI agents actually deployable at work. Companies are paying 20% more for workers with AI skills, and 98% of execs now want their entire workforce trained on AI tools.
Anthropic's Claude Cowork is moving from research preview to real enterprise deployment, and this time it's built for the messy reality of corporate IT departments. The company launched domain-specific plugins on Tuesday for finance, engineering, HR, and legal teams, plus enterprise connectors for Gmail, DocuSign, and Clay. This is the first serious attempt to make AI agents fit into how companies actually work, complete with private software marketplaces, controlled data flows, and the admin controls IT departments demand.
Why this matters: 2025 was supposed to be the year of enterprise AI agents, but most attempts failed because they didn't account for security, data governance, or how different departments actually operate. Anthropic's approach acknowledges that a finance team needs different skills than an HR team, and that companies will customize these plugins to match their specific processes. The finance plugin handles market research and financial modeling. The HR plugin generates job descriptions and offer letters. Each one encodes the basic workflows common across companies, then lets you modify them.
If you run a business with knowledge workers doing repetitive research, analysis, or document generation, this is the first enterprise AI agent system designed to be deployed without your IT team vetoing it on security grounds. Claude Cowork is positioning itself as infrastructure, not a chatbot experiment.
Read more: Anthropic updates Claude Cowork tool | Anthropic launches new push for enterprise agents
Cursor ships parallel agent execution on virtual machines
Cursor updated its AI coding agents to run multiple tasks in parallel on their own virtual machines, meaning developers can now have 10-20 autonomous coding tasks running simultaneously without competing for resources on their local laptop. The agents record their work via videos, logs, and screenshots, and can be triggered from web, desktop, mobile, Slack, or GitHub. Cursor says 35% of its internal pull requests are now generated by agents running on VMs. The company crossed $1 billion in annualized revenue in November and is fighting off competition from Anthropic's Claude Code ($2.5 billion run-rate) and OpenAI's Codex (1.5 million weekly users).
If you manage a dev team, the ability to offload grunt work to parallel agents means your engineers spend less time editing boilerplate and more time on architecture decisions. This is what "10x developer" actually looks like in 2026.
Read more: Cursor announces major update
Google adds workflow automation to Opal vibe-coding app
Google added automated workflow creation to Opal, its no-code app builder, using the Gemini 3 Flash model. The new agent feature lets users create mini-apps that plan and execute multi-step tasks via text prompts. For example, it can use Google Sheets to maintain a shopping list across sessions for an e-commerce app, planning next steps autonomously. The agents ask clarifying questions if they need more info. Opal launched in the U.S. in July 2025 and is now available in 160+ countries.
If you've been putting off building internal tools because you don't have dev resources, Opal's workflow agents are worth testing. Competitors include Lovable, Replit, and Rocket.new.
Read more: Google adds automated workflows to Opal
ProducerAI joins Google Labs for AI music creation
Google folded ProducerAI (backed by The Chainsmokers) into Google Labs, powering it with the Lyria 3 music-generation model. The platform lets you generate music, remix songs, and create custom instruments via natural language prompts. Unlike slot-machine AI music generators, ProducerAI is designed for iterative collaboration, like working with a producer. It uses Gemini for chat, Nano Banana for album art, and Veo for music videos, all coordinated by your AI producer. Free tier available; paid plans start at $8/month for ~600 songs.
If you need custom background music for videos, presentations, or podcasts and don't want to deal with royalty nightmares, this is a legitimate option. Grammy-winning rapper Wyclef Jean used Lyria 3 on his recent track "Back From Abu Dhabi."
Read more: ProducerAI joins Google Labs | This Chainsmokers-approved AI music producer is joining Google
Oura launches women's health AI model
Oura released a proprietary AI model in its Advisor chatbot focused on women's health, covering menstrual cycles through menopause. The model uses established medical standards reviewed by board-certified clinicians, plus your biometric data (sleep, activity, cycle, stress) to deliver personalized guidance. It's designed to be non-dismissive and emotionally supportive, though it's not a replacement for a doctor. The model runs entirely on Oura-controlled infrastructure; conversations are never shared or sold. Available now in Oura Labs (opt-in experimental features).
If you use an Oura ring and have struggled to get useful health guidance from generic AI chatbots, this is the first model trained specifically on women's reproductive health data.
Read more: Oura launches women's health AI
Zapier's team recorded their messy AI workflows to teach AI fluency
Zapier launched an internal series called "Behind the Build" where employees record themselves solving real problems with AI, dead ends and all. Instead of teaching specific tools (which become obsolete in months), they're teaching decision-making patterns: when to use AI, when to iterate vs. accept first output, when to stop and bring human judgment. One example: an employee building a talent sourcing tool kept pushing back when the AI claimed it was searching real candidates but the profiles didn't exist. That persistence, not the tool knowledge, is the transferable skill.
If you're trying to build AI fluency at your company, stop training on "how to use ChatGPT." Instead, ask 3-4 people to record themselves solving a real problem with AI (5-15 minutes), then share what worked and what didn't. The messy middle is where learning happens.
Read more: I asked my team to record their messy AI workflows
Uber engineers built an AI version of their CEO for presentation prep
Uber CEO Dara Khosrowshahi revealed that some engineering teams built a "Dara AI" chatbot to practice presentations before meeting with him. About 90% of Uber's engineers now use AI in their work, with 30% classified as "power users" completely rethinking system architecture. Khosrowshahi says AI is changing productivity "in a way I've never, ever seen before."
If your team is preparing high-stakes pitches or presentations, consider building a custom GPT trained on past feedback, decision patterns, and common objections from leadership. It's cheaper than a consultant and available 24/7.
Read more: Uber engineers built an AI version of their boss
India's AI boom hits monetization wall
India became the world's largest market for AI app downloads in 2025 (207% year-over-year growth), but accounts for only 1% of AI app revenue despite driving 20% of global downloads. OpenAI ended its free ChatGPT Go access in India, and Perplexity ended its bundled Pro offer with Indian telco Airtel. ChatGPT's revenue in India dropped 33% in November and 32% in December after launching sub-$5 ChatGPT Go. The promotional free-tier era is ending; conversion to paid subscribers is the next test.
If you're building an AI product for price-sensitive markets, India is the canary in the coal mine. Downloads don't equal revenue. Consider what pricing models will work post-promotion.
Read more: India's AI boom pushes firms to trade near-term revenue for users
98% of execs plan to hire or train for AI skills, paying 20%+ premiums
Zapier surveyed 550 executives and found 98% want workers who can use AI, with 44% planning to hire new talent and 65% planning to train current employees. 56% of execs already have employees in AI-focused roles (machine learning engineers, AI automation engineers). 60% predict AI jobs will see pay bumps, with some offering over 20% more for AI skills. The job market is brutal right now, but AI fluency is the skill employers are willing to pay for.
If you're job hunting, get documented AI skills. If you manage a team, start training now or you'll lose talent to companies that do.
Read more: Despite a tight job market, more than 40% of execs plan to hire for AI skills
Meta strikes up to $100B chip deal with AMD, not Nvidia
Meta signed a multiyear deal to purchase potentially up to $100 billion in AMD chips (MI540 GPUs + latest CPUs) to drive ~6 gigawatts of data center power. AMD issued Meta a performance-based warrant for up to 160 million shares (10% of AMD) at $0.01 each, vesting as milestones are hit. The full stock award requires AMD's share price to hit $600 (closed Monday at $196.60). Meta CEO Mark Zuckerberg says the deal supports Meta's push toward "personal superintelligence." Meta plans to spend $135 billion on AI infrastructure in 2026 alone.
For businesses: the AI chip wars are diversifying beyond Nvidia. CPUs are becoming core to AI inference because they're efficient and easier to scale. This matters if you're evaluating cloud providers or building AI infrastructure.
Read more: Meta strikes up to $100B AMD chip deal
MatX raises $500M to challenge Nvidia with 10x better AI training chips
MatX, founded by two former Google TPU engineers, raised a $500 million Series B led by Jane Street and Leopold Aschenbrenner's Situational Awareness fund. The company's goal: make processors 10x better at training LLMs than Nvidia's GPUs. Investors include Marvell Technology and Stripe co-founders. MatX will manufacture with TSMC and plans to start shipping in 2027. Competitor Etched raised $500M at a $5 billion valuation last month.
If you're planning AI infrastructure investments for 2027-2028, keep an eye on these Nvidia challengers. Diversification is coming.
Read more: Nvidia challenger AI chip startup MatX raised $500M
OpenAI COO: "We have not yet really seen AI penetrate enterprise business processes"
OpenAI COO Brad Lightcap admitted that despite launching OpenAI Frontier for enterprises, AI hasn't actually penetrated enterprise business processes at scale yet. OpenAI ended 2025 with over $20 billion in annualized revenue, but Lightcap says they're still "having to manage too much demand." OpenAI will measure Frontier's success based on "business outcomes, not seat licenses." The company partnered with Boston Consulting Group, McKinsey, Accenture, and Capgemini to deploy enterprise agents.
Translation: the SaaS replacement narrative is hype. AI tools are being used by individuals, but enterprise-wide process transformation hasn't happened yet. Don't rip out your existing software stack based on agent promises.
Read more: OpenAI COO says 'we have not yet really seen AI penetrate enterprise business processes'
Anthropic vs. Pentagon: The "any lawful use" standoff
Anthropic is in weekslong negotiations with the Department of Defense over three words: "any lawful use." The Pentagon wants terms that would allow unrestricted use of Claude for mass surveillance and lethal autonomous weapons (AI with full power to track and kill with no human decision-making). OpenAI and xAI reportedly already agreed to these terms. Pentagon CTO Emil Michael (former Uber exec) threatened to designate Anthropic a "supply chain risk," which would force defense contractors like AWS, Palantir, and Anduril to drop Claude. This would end Anthropic's $200 million Pentagon contract and devastate its broader enterprise business. CEO Dario Amodei met with Secretary Pete Hegseth on Tuesday.
For businesses: this is about acceptable use policies vs. "any lawful use." If you care about AI ethics or have regulatory compliance concerns, watch how this plays out. It sets precedent for what AI companies can enforce in their terms.
Read more: Inside Anthropic's existential negotiations with the Pentagon
Use Claude Code or Cursor to create code walkthroughs for projects you don't understand
Simon Willison shared a pattern for using AI coding agents to generate structured walkthroughs of codebases you didn't write (or vibe-coded and don't remember). Prompt the agent to read your source code, plan a linear walkthrough, then use a tool like Showboat to create a markdown document that explains how everything works, with code snippets pulled directly from the repo (no copy-paste hallucinations). This works for learning new frameworks, documenting inherited code, or understanding your own vibe-coded projects after the fact.
Try this: point Claude Code at a repo and ask it to create a step-by-step walkthrough using actual code snippets. Even toy projects become learning opportunities.
Read more: Linear walkthroughs
Enterprise AI agents are finally graduating from demos to deployment, but only if they play by IT department rules. Anthropic's plugin approach acknowledges that AI tools need admin controls, data governance, and department-specific customization to work in real companies. Meanwhile, the job market is brutal, but companies are paying 20%+ premiums for AI skills and training entire workforces. If you're waiting for the AI hype to die down before investing in skills or tools, you're making a costly bet. The companies hiring and training today are the ones who'll dominate operational efficiency tomorrow.