In partnership with
TL;DR
Cursor agents get their own dev environments, a mobile app, and a GitHub rival in Origin
Anthropic brings Claude into Slack as a taggable teammate
Figma turns code into a design material with code layers
Superhuman buys GPTZero to build an authenticity layer
Stripe launches a search engine built for AI agents
Zeely’s competitor spy and UGC builder turn rival ads into your next batch of creatives
Latitude turns your agent's production failures into fixable signals
Every World Cup match is a market.
48 games. 32 countries. One tournament. From the group stage through the final, every outcome is tradeable in real time on Kalshi, a federally regulated exchange and official regional partner of the Argentine National Team.
You're not picking a spread. You buy "Yes" or "No" shares on what you think happens: who wins, who advances, who scores first. Earn returns if you're right. Peer-to-peer. No house. Cash out before the final whistle.
Trade $10, get $10 free to start.
Trade responsibly.
CURSOR GOES ALL IN ON AGENTS THAT NEVER CLOCK OUT
At Cursor's first Compile keynote, the company stopped talking about coding assistants and started talking about teammates.
The headline move is cloud agents that get their own dev environment. Real compute, cloned repos, installed dependencies, the works. They can run tests, click through UIs, take screenshots, and ship a merge-ready PR while you're asleep. Cursor says these agents are now 3x faster with three nines of reliability, which is the part that actually matters. An agent that runs for days is useless if it falls over every few hours.
Then there's Cursor Mobile, an iOS app in beta. You can check on every agent you have running, see what's blocked, annotate a screenshot, and nudge it back to work from your phone. The pitch is simple: an idea hits you on a run, you open your phone, the agent starts building.
The third drop is the big one long term. Origin, a Git platform built from scratch for agent scale. Cursor's bet is that GitHub was built for humans pushing a few commits a day, not thousands of agents hammering the same repo at once. Origin auto-resolves merge conflicts, fixes CI failures, and only pings a human when it actually needs one.
And buried at the end: Cursor confirmed it's training a new frontier model from scratch, running on 10 to 20x more compute than anything they've shipped before, expected in the next few weeks.
OUR TAKE
The agents-as-colleagues framing isn't just marketing. Giving agents their own machines and their own Git infrastructure is Cursor admitting that bolting AI onto human-built tooling has a ceiling.
Origin is the one to watch. If thousands of agents really can push to the same repo without humans drowning in conflicts, that's not a feature, that's a new floor for how software gets built.
The model news got buried at the bottom of the keynote, but training from scratch at 10-20x the compute is the part that should worry every other coding tool not named Claude or Codex.
QUICK HITS
Anthropic brings Claude into Slack as an actual teammate: Claude Tag lets your whole team tag @Claude into a channel and watch it actually work the thread, pick up bugs, hand off tasks, even grind on a project solo for days. Everyone in the channel shares the same Claude, so anyone can jump in and steer it. Live now in beta for Team and Enterprise plans
Figma says design & code shouldn't be separate tools: Figma's annual Config conference began this week and they aren't holding back. The headline is code layers, letting you turn any design layer into interactive code with one click, edit it on the canvas with your team, then push changes back and forth. They also teased Figma Motion, shader fills, generative plugins, and a Figma agent.
Superhuman just bought the company built to catch AI fakes: Superhuman acquired GPTZero, the AI detection startup with 19 million users and $30M ARR. The plan is to fold its hallucination detection and authorship verification into Superhuman Go, building an authenticity layer into its AI assistant.
Stripe just built Google search, but for AI agents shopping for APIs: Stripe Directory is now in public preview, letting developers and AI agents search across Stripe Apps, Stripe Projects, and pay-per-call APIs to find and integrate services. Pull it up from the terminal with the Stripe CLI, and agents can search in JSON, parse the results, and wire up the integration themselves using the returned endpoint.
The smartest founders never pick their own 3PL.
Researching 3PLs yourself feels responsible. It's how you end up with a great sales pitch and a bad fit. There are over 2,800 of them, and the right one rarely markets the loudest. Here's why the founders who hand off the search get better answers, faster.
TURN COMPETITOR ADS INTO YOUR NEXT WINNING CREATIVE
Steal what's already working, then make 20 versions of it in less than an hour.
⏱️ 30-45 min | 🔧 Zeely AI
Why build this? Most people start ad creative from a blank page, which is the slowest and dumbest way to do it. Your competitors are already spending money to figure out what works, and if an ad has been running for weeks, that's not an accident, that's a signal. Zeely now has a built-in competitor spy tool that shows you exactly what's running, plus a massively upgraded UGC video builder. Together they cut out the screenshotting, the folder of saved ads, and the slow one-at-a-time generation grind.
Steps:
Open Competitor Spy inside Zeely and search your category or follow specific competitors directly.
Filter by format (image, video, carousel) and sort by active ads. Active means they're still spending, which means it's still converting.
Download a reference ad that has the look you want.
Drop that reference into the image builder along with your product. Zeely rebuilds the format around your product, keeping the structure that works while swapping in your actual value props.
Switch to the video studio and pick a UGC preset (digital demo, service showcase, product showcase, or fashion try-on).
Browse the avatar library, now way bigger and more diverse, and pick faces that actually match your audience. Want a UK accent for a UK audience? Done.
Batch generate 10-20 static variants at once across different avatars and references before spending any credits on animation.
Pick your winners, let Zeely auto-write the script (it pulls from your product and even competitor positioning), tweak the voiceover tone and accent, and animate.
Expected outcome: A real batch of scroll-stopping UGC ads, built around proven formats instead of guesses, without touching a camera or hiring a single creator.
Full tutorial:
TOOL OF THE DAY
Your agent works great in testing. Then it ships and you have no idea what it's actually doing until someone complains.
Latitude fixes that blind spot. It clusters every production conversation into patterns, so you can actually see where users hesitate, escalate, or just give up. When the same failure keeps happening, it bundles those into one issue with a severity score instead of making you dig through thousands of logs to spot it yourself.
It plugs straight into your coding agent via MCP. So when something breaks, you pull the real failure data straight into your editor, turn it into a test case, and confirm the fix actually holds before you ship again.
