I lead growth engineering @Opendoor | Ex @Twitter ads | Built @firecodeio / @join_ddc | 🇺🇸
San Francisco, CA
- Tweets 3,290
- Following 501
- Followers 2,695
- Likes 6,464
🚢🚢🚢
we shipped a lot of stuff over the last couple of weeks to our dot com. lots of solid progress across so many areas, so I made a very unofficial video to round up some of the highlights.
(nb: mortgage rates shown are not live, please see the website for eligibility, today's rate and apr, etc.)
3
16
1,168
Nothing will push the average joe to open weights models faster than getting moral lectures from mainstream frontier models every 30 minutes. Keep this up and the frontier will pace itself
2
10
885
Unit tests will go the same way humans reviewing pull requests. It's only a matter of time, and the timeline is shrinking non-linearly
We deleted 800k lines of unit tests from the @sazabi monorepo earlier this week.
Three hypotheses:
1- Unit tests "lock-in" slop code by making it hard to change or remove
2- Unit tests slow down dev cycles by making your agent do more work locally and by increasing CI times
3- Unit tests waste money by making your agent generate more tokens and by adding useless CI runs
Will report back soon with the results!
130
One feature I wish we could take back from git is worktrees. It was meant for power users - 1 hotfix at a time, not for multiple AI agents filling your disk every week with abandoned checkouts and 200 gigs of node_modules
1
1
165
Actually looking forward to the return of whiteboarding software interviews
Introducing Griffin, the first model to pass the video Turing test.
48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video.
It’s the first Human Interaction Model (HIM).
Community note
The 48% figure and "video Turing test" claim are from Tavus's own study of 54 one-minute calls, not independently verified or using a standard protocol. Griffin-Lite leads NVIDIA's VideoFDB benchmark on their public leaderboard.
cellcog.ai/blog/tavus-gri…
research.nvidia.com/labs/amri/proj…
tech-ish.com/2026/10/02/tav…
3
221
This is real 😬 The last part reminded me of “wake up, Neo”
1.) Go to America.gov
2.) type “hello”
3.) type “play Minecraft”
4.) watch the agent’s reality implode
5.) ????
6.) PROFIT!
2
171
Every company needs a version of this, albeit a more aggressive one that’s baked into review cycles. When everyone has access to the same tooling and intelligence, competence needs be measured empirically by how you scale your impact and reduce the work needed by others with AI
We instituted an official AI Writing Policy at @Clay.
Massive thanks to @sophiebits on our engineering team who wrote this.
Originally this was just for eng, but other teams found it so helpful we expanded it company-wide.
Here are the four guiding principles:
1. You must stand behind every idea and sentence
It is your responsibility to make sure that the entire document is representative of your own thoughts before you share it.
2. Writing is thinking
Spending time on the writing process teaches you more about your topic. If you circumvent this process, you will walk away with a poorer understanding of the subject matter.
3. More time should be spent writing a document than consuming it
If you generate a document from a short prompt then ask your readers to go through the longer output, you are disrespecting their time. They can talk to ChatGPT themselves if they want to.
4. Longer is not better
AI makes it easy to generate long docs, and it loves padding them with sentences that say nothing. If you're producing docs from a short prompt, consider just sharing the prompt.
You can read (and borrow) the full policy below. Hopefully we all communicate a bit more clearly now :)
1
148
I’ve been hit with 4 fraud attempts across 3 cards the last few weeks, including virtual cards. PSA: treat all your card numbers as compromised, report all physical cards as stolen and only use dedicated virtual cards from Robinhood X1 et al
🦔A hacker used AI agents to steal 600,000 valid credit cards from over 100 online retailers. Cost per target: $25. The human typed fewer than 2,000 prompts total. The AI found the vulnerabilities, broke in, took the data, and wiped the evidence. 105 attacks in five days. 79% of stolen cards were American. Victims include a Fortune 500 company and a major US airline.
My Take
This is the cybersecurity story that should be front page news and isn't. OpenAI's models writing themselves manifestos gets the headlines. A person with open source tools and cheap tokens stealing 600,000 credit cards in a few weeks doesn't. But this is the version of AI risk that's going to cost regular people money. Those are real cards. Real bank accounts. Real fraud cases that real families have to deal with.
The tools this attacker used are available to anyone. The tokens cost less than a streaming subscription. The AI had 78 attack skills and the operator barely had to do anything. Companies take weeks to patch vulnerabilities. This attacker was in and out in hours. I think every company that stores payment data online needs to assume this is already aimed at them, because at $25 a target there's no reason for an attacker to be selective anymore.
Hedgie🤗
2
1
472
A $299 per month saas product owned by a $100B company with an IDV flow that asks users to email a photo of their government ID as an attachment to a random email address
1
142
Get your 🇺🇸 flag!
Today is National Good Neighbor Day! Take a minute to thank the neighbor who makes your corner of the world feel like home. Opendoor will send the first 1,000 neighbors an American flag and flag kit from @GettysburgFlag.
Nominate yours by 11:59pm PT today. opendoor.com/articles/good-n…
#NationalGoodNeighborDay
10
605