If you trust Instinct with your credit card number and passwords using their Vault, don't. The "security" is a mirage: the agent can read what it swears it can't.
Today, I was wondering if Instinct could actually help me book a hotel room on https://t.co/qHFW578m88, so I asked how it might do it.
It promised me that it could fill out the form to book without EVER seeing my credit card details as long as I use their Vault.
As an engineer, I immediately thought: bullsh*t.
Instinct kept insisting that the security was rock solid and that the Vault & browser are built in a way where it can NEVER see the card details, by "plumbing."
And yet, because I know how browser agents work, I knew that can't be true.
Turns out, it wasn't. My card details showed up unmasked in Instinct's own view of the page. The protection it told me was guaranteed by design wasn't there.
I've been building startups my career, so I believe in moving fast & breaking things. I know things are messy at the beginning. But the bar for personal agents getting access to my personal info is extremely high, and telling users your architecture guarantees something it doesn't should be a red line that we don't cross.
By the way. Same goes for passwords at Instinct. Same glaring hole. But when I asked Meta Muse the same q, at least it was straight with me: it said that it can probably read all my secrets, I’d just be trusting it not to. The candor was refreshing.
Heard at YC: "We just finished sophomore year when we had dropped out. So yeah, the beginning of the marginal diminishing utility of college."
Can't believe I invested 4 years into that sh*t, a whole 2 years too much 🫠
@higgsfield raised $400M at $5.4B led by DST Global, grew $20M to $700M ARR in 12 months, valued at roughly ~7x ARR. Another "wrapper" that continues to show incredible momentum leveraging its distribution, product maturity, and user data advantages!
@WisprFlow raised $280M on $2B to keep growing voice dictation. Story: incredible user love, proprietary model, more workflow (like note taking).
Like a lot of other "wrapper" companies, Wispr continues do well despite heavy competition due to advantages in distribution and maturity of product.
It hopes to maintain its lead by aggressively partnering with other companies in the ecosystem and introducing new workflow tools (like note taking!)
With the launch of Canto, it's leveraging its proprietary real-world data to try to create differentiated results (and control over COGS) with its own speech model.
With this raise, it joins the consumer/prosumer AI unicorn club -- a group of companies that have done especially well this year by leveraging viral GTM + charging $10 - $30 / user / mo!
@stripe has agreed to acquire @OpenRouter for more than $8B in cash & stock at a roughly 55x to 60x revenue multiple.
And according to reporting by @theinformation, the company was recently generating about $12m revenue per month ($140m annualized), with gross margins of 71.5%.
The company was founded in early 2023.
Amazing outcome for everyone involved, and the second large realization (after Cursor) for AI-native darlings. @a16z a big winner in both!
(analysis of returns by share class by Claude using data from @PitchBook)
@AndreyOstrovsky Is competition really top of mind for systems? Every report -- and every health system's actions -- indicates that patients cannot access care as is, and that providers are overworked. Can you shed some more light?
@garrytan Can you define what this means?
Capital hires labor. Capital fires labor. Capital determines labor's pay and benefits, when and how labor is promoted, and how it's structured.
What makes you unique?