Claude Cowork goes mobile

In partnership with

You've seen the AI demos. Viktor does it without you watching.

The AI tool you tried last quarter waited for a prompt, hallucinated a number, then asked if you'd like a summary.

Viktor opened a PR at 2am, rebased it against main, ran your test suite, and posted a note in #eng: "Two flaky tests in payments service, both pre-existing. Recommended merging after fixing them." Then drafted the customer reply for the support ticket the bug created.

That's 619K autonomous actions per day across 20,000+ teams. Not chat replies. Real work shipped to GitHub, Stripe, Linear, Notion, and 3,000+ other tools, from inside Slack and Microsoft Teams.

You don't supervise him any more than you supervise a senior engineer.

SOC 2 certified. Your data never trains models.

"It's what you probably originally thought AI was going to be when you first heard of it in sci-fi movies." Tyler, CEO.

Get Started for Free

In today's issue

01 · Models

Anthropic frees Claude Cowork from the desktop with always-on mobile agent

Anthropic untethered Claude Cowork from your laptop: the AI agent now runs scheduled tasks overnight without you keeping your desktop open. The upgrade rolls out first to Max plan subscribers ($100/month), then to Pro ($20/month), letting users request work from their phone and return to finished results hours later.

The details:

Cowork originally shipped in January 2026 but required an active desktop session; mobile access existed only through Dispatch, which forwarded requests from phone back to computer.
Business process work and content creation are the two largest usage categories, including data reports, checklists, slide decks, and partnership proposals.
OpenAI launched Codex Remote last month and added task creation and management to its iOS app in July; Google shipped its always-on agent Spark in the first half of 2026.

Why it matters: Always-on agents are now table stakes across the major AI platforms—Claude, OpenAI, and Google all shipped versions in the last few months. For knowledge workers, this means the difference between AI as a helper you supervise and AI as a night-shift coworker that handles overnight requests without human babysitting.

Read the full story →

02 · Models

DeepSeek plans its own inference chips to cut Nvidia and Huawei reliance

DeepSeek is designing its own data center chips for running AI models after spending about a year on the project. The move lets the Chinese AI company dodge US export controls on Nvidia and reduce reliance on Huawei, which dominates China's chip market.

The details:

DeepSeek is focusing on inference chips—the hardware that powers AI models for users—rather than training chips, making them cheaper to build and directly tied to cutting costs.
Huawei currently controls roughly half of China's data center chip market, giving it outsized power over costs that DeepSeek wants to escape.
Alibaba and Baidu are pursuing similar custom silicon projects, and OpenAI just unveiled its own inference chip called Jalapeño, signaling a broader shift toward vertical integration in AI.

Why it matters: Building a competitive chip from scratch typically takes three to five years, and DeepSeek hasn't named a manufacturing partner or timeline yet—so this is a long-term bet, not an imminent product. The real win, if it ships, is controlling the profit margin on every AI response the company sells rather than paying Nvidia or Huawei as middlemen.

Read the full story →

03 · Business

SK Hynix plans $28B US listing as AI memory demand fuels 260% stock run

SK Hynix, South Korea's memory giant, is selling shares on Wall Street this week in a $28 billion offering that gives US investors direct access to the second-largest memory chipmaker globally. The timing capitalizes on an AI boom: SK Hynix's revenue jumped 200% year-over-year, and its stock is up 260% this year as data centers desperately need more memory chips to run AI models.

The details:

The company will sell 17.8 million ADRs (each representing one-tenth of a share), priced Thursday and trading Friday, letting US investors skip Micron as the only pure memory play.
SK Hynix and Samsung have committed over $550 billion combined to build new memory factories, one of the largest coordinated equipment spending sprees in chip industry history.
Micron, the closest US competitor, has climbed 700% in a year to over $1 trillion in market value, driven by the same AI memory shortage pushing SK Hynix's valuation higher.

Why it matters: SK Hynix's US debut plugs a gap for investors betting on AI infrastructure, but it carries real risk: memory is cyclical, and if new factories come online just as AI workloads shift toward less memory-hungry designs or specialized chips, both SK Hynix and Samsung could face a costly overcapacity crash like past memory downturns.

Read the full story →

AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

● ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat

● Generate images & video with Sora, Veo, Ideogram

● Compare any two models side by side

● From $8.99/mo · 80+ models, all included

Try AI Box →

aibox.ai

Trusted by 3,000+ teams

04 · Tools

Solos launches AirGo A6 smart glasses at 19 grams, skipping cameras for voice AI

Solos just released the AirGo A6, ultra-light smart glasses that weigh 19 grams and skip cameras entirely in favor of a voice-only AI assistant. That's roughly one-third the weight of Meta's latest glasses and sidesteps the privacy backlash that's dogged camera-equipped frames.

The details:

The A6 weighs half as much as last year's AirGo A5 (36–40 grams) by using slimmer temple arms; slimmer weight means less nose and ear pressure, the biggest pain point keeping people from wearing smart glasses all day.
Solos released three privacy add-ons for its older AirGo V2 camera glasses: $39 transparent temples, $49 camera shield with sunglasses, and $79 bundle—letting owners disable the camera when they want.
The A6 has no display and no camera; it handles voice queries, real-time translation, calendar reminders, and calls while letting wearers still hear their surroundings through speakers behind the ears.

Why it matters: Solos is betting that voice-only smart glasses can win where cameras lose: you can wear them anywhere without making people uncomfortable about being recorded. Without a camera, the A6 trades visual lookup and scene description for all-day comfort and privacy peace of mind—a trade that matters more as smart glasses move from niche gadget to everyday wear.

Read the full story →

05 · Business

Vercel's Guillermo Rauch bets on splitting AI models from agents

Vercel, the platform that hosts millions of web deployments, is betting that the future of AI belongs to companies that keep models and agents separate — not locked together by a single AI lab. The startup processes 6 million deployments daily and over 1 trillion AI tokens, giving it a front-row seat to how companies actually build with AI.

The details:

Half of Vercel's deployments are now triggered by coding agents, and the company shipped Eve (a natural-language agent framework) and Vercel Sandbox to let teams control what data internal agents can access.
Customers are abandoning single-lab bets: companies standardized on OpenAI or Anthropic last year but now mix models from Gemini, DeepSeek, GLM-5.2, and others as interchangeable parts.
CEO Guillermo Rauch framed the shift as a direct counter to OpenAI's push to let users build websites inside ChatGPT — Vercel wants to be the infrastructure layer models recommend, not the other way around.

Why it matters: The real battle is over who controls the stack. If companies can swap AI models like components, they won't be locked into one lab's ecosystem — which means infrastructure platforms like Vercel gain leverage. That's why OpenAI is moving into hosting: both sides know the layer that connects models to production wins.

Read the full story →