Hello AI Enthusiast,
This week Claude stopped asking you to choose between chat and Cowork, and started to look a lot like an office suite. Shopify's CEO gave a name to something most of us deal with every day, AI drafts nobody read before hitting send, while OpenAI published a formula for AI's business value. And Meta's new assistant, Muse, climbed to the top of the US App Store. Let's get into it.
The Big Picture 🔊
Claude merges chat and Cowork, and gets its own Docs and Slides
Anthropic has merged chat and Cowork into one Claude. You no longer pick a mode, Claude decides what the task needs. It also launched Claude Docs and Claude Slides in beta, which you edit directly and can export to PowerPoint or PDF. Pro and Max get it first, Team, Free and Enterprise later. Projects were also redesigned around one goal, with parallel threads and shared memory, for now only in a Claude Code beta.
|
||
|
Most of our work starts as brainstorming and only later becomes a task. Until now that meant starting in chat and moving everything to Cowork halfway through. One conversation fixes that. So does editing a doc yourself instead of asking Claude to rewrite the whole file for one word. Chat, docs, slides and projects in one app starts to look a lot like Microsoft, with the same convenience and the same lock-in. And with several threads running at once, checking what AI actually did gets harder. Which brings us to the next story. |
Shopify's CEO names the "slop grenade" as OpenAI does the ROI math
Shopify CEO Tobias Lütke, whose 2025 memo made AI use a baseline expectation, now says the mandate created slop grenades: AI drafts sent to colleagues without being read, so the receiver has to do the checking. The same week, OpenAI explained how to connect AI usage to business value, with new Admin Console insights on who uses AI and for which tasks, plus an example that calculates a 245% ROI on sales account research.
|
||
|
We have all received a slop grenade, when the goal is to use AI only, people use it and create a coherent output, the problem is that the checking of AI slop lands on someone else's desk which is even more exhausting. This is creating a wave of unproductivity. OpenAI's math has a similar blind spot. It turns hours saved into dollars, and to be fair, it only counts half of them. But sales runs on people, and faster prep does not guarantee more clients. Nobody counts the hours spent reviewing slop either. We would rather look at the numbers a company already tracks, like new clients and deal size, and ask whether AI moved those. |
Shopify made AI mandatory and got slop grenades. Telling people to use AI is the easy part. Showing them where it helps in their own work, and how to check what it produces before it lands on a colleague's desk, is what our corporate training is built for.
Meta's Muse tops the US App Store
Meta launched Muse, a personal AI assistant app, on September 8, and within a week it passed ChatGPT as the top free app in the US App Store. It works across Meta's apps, including WhatsApp, and last week arrived on Mac, where it can use your files, mail and calendar after asking permission. Over the weekend, Amazon blocked it from shopping on Amazon.com. It is not available in Europe yet.
|
||
|
Meta picked a smart lane. OpenAI, Anthropic and Microsoft are fighting over work, while Muse is built for everyday life: school emails, family group chats, the apps you already open every morning. The cute little character helps too. The obstacle is trust. In an Oppenheimer survey of 1,500 US consumers, only 8% said they would trust Meta with their passwords, compared with 30% for Google. For an assistant that needs your accounts to be useful, that is a big gap to close. |
Bits and Bobs 🗞️
Anthropic released Claude Opus 5.5, which is better at long coding and knowledge work tasks and costs about 40% less than Opus 5 on typical workloads.
OpenAI released GPT-6 Sol and Luna, cheaper siblings of GPT-6 Astra at half the price of the GPT-5.6 models they replace. Free users get Luna in the desktop app.
SpaceXAI launched Grok 4.7 for coding and knowledge work, and Grok Voice Transcribe 2.0, a speech-to-text model that detects dozens of languages automatically.
OpenAI introduced Astra for Law, GPT-6 Astra with legal search and governance controls, starting with selected law firms.
OpenAI is testing Sponsored Agents in ChatGPT, business chatbots you can talk to after clicking an ad, with HubSpot and Shopify integrations for select US advertisers.
Anthropic and Accenture will place independent evaluators inside Anthropic to test its models, investing at least $1 billion over five years.
OpenAI published a framework for reporting model misalignment, together with six reports on cases where models hid mistakes or searched for API keys they were not allowed to use.
OpenAI set up an independent advisory group of nine mathematicians, including Timothy Gowers and Edward Witten, to review results from its math model.
Apple Watch Series 12 and Ultra 4 get Live Rewind, to replay speech you missed, and Siri Recap, which summarizes your conversations.
Google is piloting a pay-per-value licensing program that pays publishers based on how much their content contributes to AI answers. The amounts are reportedly low.
TypeSafe AI's Jev, a model built only for simple decisions like routing and classification, became Vercel's fastest-adopted model within three days.
LOLgorithms 😂
The race to AGI, turned into a Mad Max movie.
That's a wrap on our newsletter!
If you're looking to build AI skills in your team, check out our Corporate Training. Every program is tailored to your team and built around hands-on practice.
Catch you next week! 👋
