every week a new ai tool drops and everyone loses their mind. i test them like someone who actually has to use them at work and in real life. 👀 ux teardowns 🔍 the workflow step the demo always skips ✅ tools worth using after week one 💬 honest takes when it breaks.

ChatGPT is getting a serious upgrade. GPT-6 + a smarter, more interactive interface are now rolling out to everyone.

We're releasing pplx-embed-v2-late, two late-interaction embedding models that retrieve text, images, and pages with a shared embedding space for cross-model querying. Both models achieve frontier performance and are publicly available on Hugging Face. perplexity.ai/hub/blog/multi…

a few of iPhone 18 Pro Max owners on AT&T are learning the difference between prevention and repair. Apple says iOS 27.0.1 plus a carrier settings update can help stop the cellular failure from happening. however, if the phone is already stuck in SOS mode, the update will not restore it. the device needs to be replaced. that distinction should be the first sentence in every support message: still connected? update now. already disconnected? contact Apple or AT&T for a replacement. support UX matters most when the product cannot support its basic function.

Claude can now edit Google Docs, Sheets and Slides without making you copy everything into another tab. that is the obvious improvement. the better product decision is smaller: Ask before edits is the default. each change appears as an approval card before Claude touches the file. there is also an Accept all edits mode for people who want speed. same AI. two levels of trust. good AI UX does not force everyone into automation. it lets the user decide when assistance becomes permission. which mode would you actually use? claude.com/resources/article…

mathematicians about to open their inbox and find 400 new side quests i’ll believe the number when OpenAI publishes the list and independent experts have checked the proofs.

An AI that lives in your house, remembers everything, works 24/7 and pays no rent. very on brand 😭 local first is the interesting part though. i would want a very obvious warning whenever a feature sends anything outside the device.

Gap is adding conversational shopping across its brands. instead of selecting: women → dresses → formal → black → size 10 someone could ask: “I need a black dress for an outdoor wedding that works in warm weather.” that is a better input because it contains the occasion, environment and intent. but a good result still needs visible reasons: why this fabric? why this fit? why this price? replacing filters with a chat box only works if the recommendation is easier to understand than the menu it replaced.

OpenAI is about to release 400 papers on every math problem it’s solved today. Maybe in a few hours.

TikTok’s new Shopping Assistant can answer questions about product details, sizing, availability and shipping then help move the user towards checkout. the missing UX question is, does it explain why it recommended that product? “this matches your size, budget and delivery deadline” is useful. “we thought you’d like this” hides whether the result is relevant, sponsored or simply profitable for the platform. conversational shopping removes filters. it should not remove transparency.

honestly the left handed part would annoy me the most. spending that much on a phone and still having to reach awkwardly across the screen every day would drive me nuts

to be honest free tier that is already good enough.

apple’s response to ai agents asking for broad mac access is a product-design correction. “full disk access” is not a normal permission. it can include files, emails, messages and browsing history. when an app asks for access that powerful, the interface should not make consent feel like another box to click through. good permission design makes the consequence visible before the user says yes. explicit action is not friction when the request is extraordinary. reuters.com/business/retail-…

smart glasses succeed when they stop feeling like a gadget someone is wearing. they need to work as: a frame people actually like a camera people around them can recognise an interface usable without a screen a product that signals recording clearly the difficult part it is making technology feel natural without making everyone nearby feel watched. good wearable design has to serve the wearer and the room.

AI companies are spending billions on chips, power and data centres. but users do not experience a financing package. they experience: whether the product works what it can access what happens to their data whether a mistake can be corrected the next competitive advantage is not only more compute. it is making powerful systems understandable at the moment someone needs to trust them. infrastructure creates capability. product design turns it into confidence.

america.gov’s new chatbot is reportedly giving answers that conflict with statements from the administration behind it. that exposes a basic product problem. when a government chatbot answers a question, the user needs to see: which source supplied the claim when that source was updated whether the answer is official policy where conflicting information exists a polished response can still leave the user unsure which version to trust. for public service ai, citations are part of the interface.

that some fun easter eggs is all you had to say. i am going in for the API guides and will somehow spend 40 minutes looking for a hidden button.

Claude.dev is our new home for developers building with Claude. You'll find engineering deep dives, Claude Code and API guides, tips from the teams building Claude, and some fun easter eggs.

my to do list just found out it can do things without me. bit awkward for both of us. but for anyone in corporate, this could actually be useful because it can set an automation to check for updates in Slack and Gmail every morning, then pull together what needs your attention before the first meeting.

chatgpt agent can browse websites, work with files and complete multi step tasks. the useful question is what happens before it takes an action that matters. say you ask it to prepare for a trip. finding flights and comparing hotels can happen in the background. booking a room, sending a message or paying for something should bring you back into the process with a clear approval screen. a useful agent should show: what it plans to do which account it will use what information it will share whether the action can be reversed the agent can handle the steps but the user still needs control over the commitment.

calling Fable vital to Xbox’s DNA while the team faces layoffs is a rough contrast. if it’s that important to the platform, the people making it should feel that too.