In partnership with
Reading time: 6 minutes
🗞️ In this edition
Google Is Deleting Your Gems
Sponsored: TigerData
arXiv Told Scientists To Stop Submitting
Barclays Is Handing Half Its Developers To Claude
⚡ YOUR MOVE For people building the life they want, now.
In other AI news:
Ant Group Put 560 Billion Parameters On Free Trial
OpenAI Is Teaching A Model To Drive Chip Software
Google Gave MIT $2.1 Million To Fix Your Commute
4 must-try AI tools
Google is deleting a feature you may have saved things in. arXiv put a hard cap on its own scientists. And Barclays named the share of its engineers it wants handed over by December.
Keep reading.
androidheadlines.com
WHAT HAPPENED
Google is retiring Gems, the saved prompt feature in the Gemini app, and replacing it with Skills. A Skill is a set of instructions you save once and then call up by typing a forward slash and its name inside any chat. You can stack several at a time, so a brand guide and a tone of voice can run together, and you can attach reference files like PDFs and images. Gemini can also notice a prompt you keep retyping and offer to save it as a Skill. Every Gem you already made converts across on its own. Personal accounts move in November 2026, work and nonprofit Workspace accounts in March 2027, and education accounts in June 2027. Skills work on free accounts, and Gems began life as a paid feature.
WHY IT MATTERS
The slash command is the quiet part. Gems sat in a side panel you had to go and find, which is why most people made one and never opened it again. A Skill sits in the prompt box, so it gets reached for in the middle of a job rather than at the start of one, and that changes how often a saved instruction actually gets used. Making it free matters more. Google is handing its entire user base a way to pin down house style, reading level and formatting rules, which is most of what people currently pay separate writing tools to do. Anyone selling a thin layer of saved prompts on top of somebody else's model just watched the model vendor ship it as a feature.
OUR TAKE
The obvious read is a tidy rename. We think it is a land grab on the prompt layer, and the giveaway is the free tier. You do not move a paid feature down to free unless the point is to make everyone's saved instructions live in your app instead of somebody else's.
Watch whether Skills get to talk to each other. If Google lets one Skill call another, or wires them into Gemini's agent features, before the end of 2027, this was the first brick of an app platform and Gems was in the way. If they stay a private shortcut list with nothing connected to them, this was housekeeping with a better keyboard shortcut.
From Our Sponsor:
Analytics on Live Data Without Leaving Postgres
When analytics on Postgres slows down, most teams add a second database. Then come the pipelines, the sync jobs, and a copy of your data that's always a little behind.
TimescaleDB takes a different approach: extend Postgres instead of splitting away from it. Hypertables partition your data automatically as volume grows. Hypercore compression cuts storage up to 95%. Continuous aggregates keep dashboards live without re-querying everything.
CERN runs Postgres this way for sensor data from the Large Hadron Collider.
No split architecture, no pipeline lag, no new query language to learn. Same SQL, same drivers, same tools.
Start on Tiger Cloud and get $1000 in credits.
Source: arXiv.com
From October 1, anyone posting to arXiv can put up two papers a calendar month and hold three in the queue at once, across every field. September brought 40,363 submissions, against 20,569 in the same month of 2024 and 9,869 ten years before that, and roughly 9,000 support tickets landed in that one month. arXiv blames thin papers of narrow scope, and what it calls salami papers, meaning one piece of work sliced into several, with AI tools making both easy to produce.
How did today's stories land?
Source: Banking40.ro
Barclays wants half its developers using Claude Code by the end of 2026, and most of its software engineers on it through 2027. The groundwork is already running. More than 16,000 staff use an internal assistant built on Claude, it has handled over a million searches, and a separate system sorts 120,000 emails a day inside the markets division. All of it sits under a push to find £2 billion of savings. Co-chief operating officers Anne Marie Darling and Craig Bright are the two names on it.
⚡️ YOUR MOVE
FOR PEOPLE BUILDING THE LIFE THEY WANT, NOW.
🧪 Your coding assistant keeps inventing package names that do not exist, and attackers have started registering them and waiting. The Cloud Security Alliance checked 2.23 million code samples across 16 models and found made up packages in 19.7% of them, more than 205,000 distinct names. → Check before you install
📄 LlamaIndex Extract v2.5 pulls fields out of messy documents and now points at the spot on the page it took each one from. Accuracy on the cheapest tier went from 87.1 to 93.9, and the per page price did not move. → Run it on your worst PDF
💬 Photon raised $4.5 million so one agent can reach people on iMessage, WhatsApp, SMS and RCS through a single interface, written once. Revenue grew tenfold between May and August. → Pick your channel
🧵 Restate took $20 million for a runtime that lets a long agent job survive a crash or a dropped connection and pick up where it stopped. Replit is already a customer. → Fix your retry logic
Ant Group Put 560 Billion Parameters On Free Trial – Ling-3.1-flash uses about 25 billion of those parameters per answer, is held to 256,000 tokens of context during a two week free trial, and the company says it will open source the model once the trial ends.
OpenAI Is Teaching A Model To Drive Chip Software – GPT-Synopsys is a multi year revenue sharing deal that lets AI agents operate the design tools used to lay out semiconductors, with early semiconductor customers already testing it.
Google Gave MIT $2.1 Million To Fix Your Commute – Google.org funded a hub that merges transit monitoring, control room decisions and passenger messaging into one system, with agencies in Washington DC, Chicago, London, Boston, Tokyo and Hong Kong involved.
Notion AI: Finds the answer buried in the workspace you stopped organising.
FlowiseAI: Wires your agents together on a canvas instead of in a codebase.
Screenshot To Code: Turns a picture of a design into the front end that runs it.
Briefly AI: Cuts an hour of recording down to the part you actually needed.
Three organizations drew a line this week, and all three drew it around volume. Google decided your saved prompts belong in its prompt box rather than a side panel nobody opens. arXiv decided the problem is no longer getting science published but getting too much of it through the door. Barclays decided the question is not whether the tools work but how many of its own engineers it can move onto them by December. Nobody is arguing about what these systems can do anymore. They are arguing about how much of it to let through.
Decide what you are letting through.
Interested in featuring your services with us? Email us at [email protected]
