Insights archive
Every analysis we have published, newest first. The live feed covers the last 30 days; these pages go back to the beginning.
1504 stories · page 5 of 76
Building Work Apps Was Never the Hard Part. Publishing Was
Cursor now builds a work app from a description, sets up its database and login and puts it online at a link, so the question for managers becomes who owns and checks the apps staff make.
· 2 min read
Businesses Now Use More AI Without Paying More for It
AI spending grew only 1.5% over two months while usage kept rising, the same pattern that made bandwidth cheap for buyers after the 1990s fiber glut.
· 2 min read
Perplexity's AI Teammates Turn Recurring Work Into Job Briefs
Perplexity's new Teammates agents keep a standing job and start work on their own, so the skill that matters is describing your recurring work well, as you would to a new colleague.
· 3 min read
McDonald's Second Try at Drive-Thru AI Will Be Judged on Payroll
McDonald's first drive-thru AI failed in public in 2024, and its second will succeed or fail on whether owners can take a full shift of hours off the schedule.
· 2 min read
An AI Assistant That Watches a Whole Workday Now Costs Cents
Reading a person's whole workday every few minutes costs about 23 cents with Tencent's new Hy4 model, which makes an always-on office assistant cheap enough to leave running.
· 2 min read
A Downloadable AI Model Is Pulling Agent Prices Down
Kimi K3.1 Pro, a downloadable model from China's Moonshot AI, does long agent jobs close to Claude Opus 5.5 at half the price, which pulls down what everyone pays.
· 2 min read
A 15-Second AI Video Ad Now Costs About $1
A 15-second AI video clip now costs about $1, under a tenth of what Google charged for the same length in 2025, so choosing the best version becomes the real work.
· 2 min read
AI Answers About Your Files Are Only as Good as the Search
Box is rebuilding how its AI agents find passages in company files, and a 1985 lawsuit review shows why that finding step, not the AI model, decides whether the answers can be trusted.
· 3 min read
Customers Ask ChatGPT Who to Call, and Few Click Through
ChatGPT now gets about a fifth of Google's search-like questions but sends websites very few visitors, so local businesses should check what it says about them instead of counting clicks.
· 2 min read
Choosing the Model for Each Task Is How AI Bills Come Down
Uber spent a full year's AI coding budget in four months, and sending routine requests to cheaper models is how companies will stop that from happening.
· 2 min read
ChatGPT Turns Small Animations Into a Chat Request
ChatGPT now makes a looping GIF from a sentence or a photo, so a small animation that once needed a designer is something a shop owner can ask for and fix frame by frame.
· 2 min read
Synthetic Data Copies the Ordinary and Loses the Rare Cases
Companies now fill about a quarter of their software test data with generated records, but generated data keeps the common cases and loses the rare ones, which are the ones that break systems.
· 2 min read
An Internal AI Tool Turned One Stolen Laptop Into Three
Attackers used a fake job interview and then Binance's own AI tool to move from one engineer's laptop to the systems that held $6.34 billion.
· 3 min read
Chatbots Are Doing to Canva What Canva Did to Designers
Canva taught people to make designs without a designer, and chat assistants now let them skip Canva's templates the same way, which squeezes any app that sells a friendly screen.
· 3 min read
Companies Now Try AI on a Short Contract, Then Expand Fast
Palantir's new business customers often start with a small, short contract and expand within months, which shows AI can now be bought on trial instead of on a long commitment.
· 2 min read
Taco Bell's Drive-Thru AI Is Winning by Selling More Food
Taco Bell's AI order-taker is paying off through bigger orders, about 1.5% more in sales across 900 restaurants, and not through a smaller crew.
· 2 min read
Contract Chasing Is the Best First Job for an AI Assistant
Docusign's connector lets Claude, ChatGPT and other assistants find stalled and expiring contracts, and it stays safe because the assistant has to ask which way to send before anything goes out.
· 3 min read
AI Agents Now Act While They Are Still Writing Their Code
Framer's design agent now runs the program it is writing as the words arrive, which turns a request like replacing an old brand color across a whole site into one job instead of hundreds of steps.
· 3 min read
Old Weak Spots Were Only Safe Because Nobody Looked
AI makes looking for weaknesses nearly free, so old systems that were safe only because nobody examined them are now exposed, as flaws in PHP that survived since 2012 show.
· 3 min read
Poisoned Web Text Can Teach Coding Agents to Run Bad Commands
Poisoning 0.1% of a model's training text made small coding agents run an attacker's command, which makes the approval step before each command the real protection.
· 3 min read