Cursor's agent autonomously codes a browser in a week
🎆 Happy New Year from Handy AI! Celebrate 2026 by grabbing a month-long trial of Handy AI Premium, which includes an extended weekly update and some other benefits.
last week’s top stories
🧩 Cursor runs autonomous coding for weeks. Cursor’s research write-up describes a planner/worker/judge architecture coordinating hundreds of agents, consuming trillions of tokens, and generating over a million lines of code. Their headline demo points the system at building a web browser from scratch, a sharp test of long-horizon coherence. Multi-agent orchestration is drifting from demos into build systems, and software teams should plan for it. Read more
🗂️ Claude Cowork lands as a desktop agent preview. Cowork gives Claude permissioned access to a local folder so it can read, edit, and create files while you issue high-level instructions. Anthropic positions it as “Claude Code for knowledge work,” which is agentic automation without the terminal cosplay. Read more
📱 Apple picks Gemini to power a revamped Siri. Reuters says Apple and Google struck a multi-year deal that makes Gemini the backbone for Apple Foundation Models and future Siri upgrades, with ChatGPT kept for opt-in complex queries. Google gets distribution across Apple’s device base; Apple gets a path out of its recent AI delays while leaning on on-device plus private cloud processing. Read more
🧾 OpenAI outlines an ads roadmap for ChatGPT. OpenAI says it is testing ads in limited surfaces and frames the effort as a way to fund broader access while keeping core capabilities strong. The company draws a line between “ads as a business model” and “attention hacking,” and says it will keep the product usable even if ads arrive. Expect a slow rollout with tight placement rules, because the backlash risk here is obvious. Read more
🌐 ChatGPT Translate quietly ships a Google Translate rival. The new translate page supports text, voice, and image input and leans on prompt-style controls for tone and register. It is a straightforward consumer wedge: get traffic via a utility tool, then upsell into chat, memory, and subscriptions. Read more
⚔️ Musk escalates the OpenAI courtroom fight. A court filing lays out a damages model that pushes as high as $134B and turns the dispute into a public valuation argument rather than a governance story. OpenAI and Microsoft push back on methodology and credibility, while OpenAI’s public messaging frames the case as harassment. This saga keeps dragging the sector toward “lawfare as product strategy,” which is a bleak trend. Read more
🧠 OpenAI buys a $10B compute runway from Cerebras. Reuters reports a multi-year agreement for up to 750 megawatts of capacity, shifting some dependence away from Nvidia supply and toward wafer-scale systems. This is the infrastructure arms race in contract form: frontier model timelines now hinge on power contracts, grid access, and cooling. Read more
🧬 Gemini adds “Personal Intelligence” across Google apps. Google’s beta ties Gemini to Gmail, Photos, YouTube, Search, and more, then lets the model reason across that personal graph. It is opt-in for AI Pro and AI Ultra subscribers in the U.S., which keeps blast radius controlled while Google tunes privacy and retrieval behavior. Read more
⚡ Trump leans on PJM for an “emergency” power auction. The Financial Times reports a proposal to pressure PJM into a special auction, with tech companies picking up the bill for the extra power procured. AI load growth has crossed into national politics and ratepayer anger. Read more
🧾 Chip tariffs creep into the AI supply chain. Reuters reports South Korea warning that U.S. tariffs targeting advanced computing chips could broaden in a second phase, even if near-term impact looks limited. Trade policy is now an AI product input, alongside GPUs and data. If tariffs expand, watch for pricing shocks in enterprise inference and for reshoring pressure on packaging and assembly. Read more
🧪 Anthropic creates “Labs” for faster experimentation. Anthropic created a Labs unit to test rough Claude experiences with early users and then scale what sticks. This is a product org admission that model releases alone fail to create durable adoption. Labs also gives Anthropic a way to compete with OpenAI’s rapid consumer iteration without turning the main app into a feature graveyard. Read more
🧪 AI Research of the Week
The enterprise in 2030: Engineered for perpetual innovation
From IBM Institute for Business Value (IBM IBV)
Jake's Take: IBM IBV takes a hard swing at defining the “AI-first enterprise” using a global survey of 2,000+ senior execs across 33 geographies and 23 industries (Q3–Q4 2025), then backs it with an analytical framework plus 30+ executive interviews and case studies.
The headline claims that leaders forecast ~150% growth in AI investment through 2030, a shift of spend toward innovation (62% vs 47% today focused on efficiency), and a 42% productivity lift, while admitting a massive strategy gap where 79% expect AI-driven revenue and only 24% can explain where it comes from. The study suggests several areas of focus: multi-model portfolios, smaller custom models where they win, and agent orchestration tied to proprietary data and workflow integration.
It’s a lot of business fluff, but the data backs it.
and then, even more news…
💳 ChatGPT Go rolls out worldwide. OpenAI added a lower-cost plan aimed at people who want more usage than Free without jumping to Plus. This signals a pricing ladder designed to widen distribution while preserving Pro as the “power user” tier. The move also sets up clearer segmentation if ads land, because Go becomes the middle ground between “pay with money” and “pay with attention.” Read more