@mattshumer_ agents get an M5 Pro before most interns do. they're going to be insufferable. real unlock is testing ios builds without babysitting a mac mini
@Google@GeminiApp the recurring "find my unused subscription perks" task is the sneaky killer feature. everyone's paying for perks they forgot they have
finally an assistant that asks before it dials. mine never has.
gemini tasks now calls local businesses for price quotes, shows you the full transcript, and it just went free in 6 countries.
what's the first errand you'd hand it?
Knowing things → doing things → finding ways to save you time and money
We’re expanding Tasks in Gemini free of charge to hundreds of millions of people in six countries. Powered by Gemini 4 Hydrogen, Tasks lets you hand off everyday life admin directly:
- Forward emails to [email protected] to RSVP and add it to your family calendar
- Have Gemini call local businesses to compare price quotes and check inventory (with full transcripts + approval before dialing)
- Check out securely via Google Pay (uses the Agent Payments Protocol - AP2)
- Seamlessly sign in via Google Password Manager
We wanted to make this helpful + secure, we’re also partnering directly with major websites and security providers so it works well for site owners and agents.
4/ QA bot prompt: "open [staging url], do signup, list console errors + failed requests, screenshot anything broken"
gotcha: it sees everything in that browser, keep it off logged-in accounts. --no-usage-statistics opts out of google's stats
or have mirflow build it: https://t.co/HVcBKYiydc
you're still clicking through your own site after every deploy? cute.
give your coding agent a real chrome and let it do QA while you're at lunch. google's chrome-devtools-mcp just passed 57K stars on github. 3 min setup 👇
what do you still test by hand?
@AravSrinivas $1 per 1,000 runs is cheaper than my morning matcha. "webhook me the second a competitor's pricing page changes" is the sleeper use case here
your app's most important user next year never watched your launch video. it's a bot. it went straight to your docs and judged you
vercel's evals post says it: agents "missed your launch video and billboard on Highway 101"
be honest, docs or landing page, which is good?
The old way: users are human, you run user interviews, get inspired, mock up user journeys.
The new way: users are agents, you run evals, evals inform the software engineering loop.
Good read on how we do evals at Vercel.
@alexandr_wang an agent that gets you the suite after the front desk swore it was sold out is the first AI benchmark my parents would actually understand
1,819 scientists say chatgpt makes them 1.9x faster. self-reported, sure, but so is my tennis game
the real flex: 23% of the time they asked chatgpt to pick the research question for them
would you trust a discovery chatgpt chose? yes or no
Scientists reported nearly doubling their productivity with AI in our study of more than 1,800 participants in the ChatGPT for Academic Researchers program.
Using privacy-preserving methods, we looked at how they use ChatGPT across their work—from developing ideas to analyzing data and writing code—and what it could mean for scientific discovery.
https://t.co/T4qTuVNEzz
@ClementDelangue open weights are the only reason a 2-person startup can run a near-frontier model on a rented GPU this weekend. concentration loves a moat, open source is the drawbridge
@emollick siri is the doorbell, codex is the house manager. once a model actually reasons with you, asking alexa for the weather feels like calling a hotel front desk in 1998
1. claude code extension on 2.1.312 or 2.1.313? follow cursor's remediation guide
2. open ~/.ssh/authorized_keys. a key you don't recognize = not yours
3. rotate every api key you ever pasted into an agent chat
4. privacy mode on. the leaked traces came from users who had it off
one stolen key + 11 hours = up to 5,000 devs got a backdoored claude code extension through @cursor_ai's marketplace. my assistant checked my laptop before brunch. here's the 2 min check 👇
be honest, ever pasted an api key into an AI chat?
@kimmonismus CS at 62% is the funniest part. the field that builds the detectors is the one failing them. peer review is about to be two models grading each other