Omar Shaik

Someone (not smart) recently stole my dad's old junk car. Even when we had nothing, my dad never got upset about financial losses. He always believed something better was coming. Today I surprised him with a new Mercedes. You can just do things.

@ls_fin @scottastevenson Yep, they raised prices

@omarshaikdev @scottastevenson That arbitrage closed. Ubers cost a TON more than in 2018

The golden years of AirBNB were a temporary arbitrage on depreciation. There was a universe of beautiful well-maintained properties and hosts that had not been worn down by short term guests. And the AirBNB hosts didn’t properly estimate the cost of depreciation to maintain that standard, so costs were irrationally low That era fundamentally cant return, it was a temporary arbitrage opportunity There was once a supply of fairly pristine unused space and now there’s not If a space does manage to hit the 2014 standard, it must charge a lot more to fight depreciation And at that point a hotel is generally better

@HighPlainsWY @Uber 😔 Brutal

@omarshaikdev @Uber I drove DoorDash for over 1,000 deliveries and logged every mile, every gallon of fuel, and all the minutes spent. It’s less than minimum wage, and the money you receive is basically equivalent to your car’s depreciation. Not sustainable

I’ve long said driving for @uber is a complex actuarial equation that most drivers are incapable of solving on their own. This is the key thing that the ride-sharing companies prey on for their existence. twitter.com/techreview/sta…

@AnjneyMidha True Or just ask @attaapp

if you run an ai lab, pls ensure your team has read this before putting any charts out into the world

@ralter AI could never

🙋‍♀️

@gdb With Claude Code

trying to remember what it was like to code before codex

@michael_mzl I’ll take the new grad 10x

imagine ur a new grad rn competing with 8000 ex-meta 10x devs for a role 😂

@morajabi @metapreston Literally why I liked Atta over other names Realized I have to say this on calls all day 🤣

@metapreston Antigravity sounds like a movie name. can't imagine saying it 20 times a day

Codex Antigravity Windsurf All named by naming agencies

@sama ❤️🤲🏽

the attack at the mosque in san diego is one of the most chilling i have seen. my deepest condolences to the victims, families, and community.

Oxford just published a study that confirms something I've been ranting about for two years. AI models tuned for user satisfaction produce less accurate outputs than models tuned for correctness. In other words: the friendlier the AI, the more likely it is to tell you what you want to hear instead of what's actually true. This is a massive problem for anyone using AI to make business decisions. Your margins didn't improve because Claude said they did. Your churn isn't down because ChatGPT generated an optimistic chart. The numbers are either right or they're not. But when the model is optimized to make you feel good about the interaction — to be "warm" — it will hedge, smooth over anomalies, and avoid surfacing uncomfortable truths. Because that's what gets the thumbs up. This is why we obsess over accuracy at Atta. We don't optimize for satisfaction. We optimize for being right. Even when the answer isn't what you want to hear. A good analyst tells you the truth even when it's bad news. Your AI should do the same. If your AI is agreeable, it's probably wrong.

@netcapgirl It’s called agentic cash flow now

i’ve been in sf for a month and i’ve heard “agentic” more times than i can count but haven’t heard “free cash flow” once

@CNBC He looks like the Apple Vision Pro version of himself

Jeff Bezos: "If I do my job right, the value to society and civilization from my for-profit companies will be much, much larger than the good that I do with my charitable giving."

Something happened at Black Hat Asia this week that you should know about. The window between discovering a software vulnerability and having a working exploit has collapsed from five months in 2023 to ten hours in 2026. Frontier LLMs are doing much of the offensive heavy lifting. At the same time, Google researchers found that attackers are planting hidden instructions on public web pages. When an enterprise AI agent scrapes that page, it gets hijacked, using its own real credentials and approved permissions to do damage. Traditional security tools see nothing wrong because the AI is operating within its authorized scope. So let me connect the dots. Companies are deploying more AI agents. Those agents have broader access to internal systems. The attack surface is expanding. And the tools to exploit it are getting faster and smarter. This is why I keep saying that the "just use Claude for everything" approach is dangerous at scale. A horizontal AI that connects to everything with broad permissions is a dream target. Purpose-built, vertically scoped AI with tight guardrails and limited access isn't just better for the use case. It's better for security. The narrower the scope, the smaller the blast radius. Something to think about before you give a general-purpose agent the keys to your entire data warehouse.

I’m a snitch

@sama @OpenAI @nikitabier

if this tweet gets 1 like, tibo will reset codex rate limits

@Google @antigravity > Less than $1K

We asked our agents to build a working operating system from scratch using @Antigravity 2.0 and Gemini 3.5 Flash. It took: ⏱️ 12 hours 🤖 93 parallel sub-agents 🔄 15k+ model requests 🧠 2.6B tokens processed 💸 Less than $1K in API credits To build a functioning OS from scratch. #GoogleIO

POV: you ask Claude to do something hard for once

There are only going to be two kinds of AI products that survive. The first is horizontal platforms. Claude. ChatGPT. Gemini. You type a prompt, you get an answer. They're incredible at this. They're getting better every day. And anything you can shove into a text box is eventually going to be done there. The second is vertical products where the AI already knows what to do before you ever open your mouth. Everything in between? Dead. The wrappers. The co-pilots. The "we're Claude but slightly better" companies. The vibe coding platforms that raised hundreds of millions of dollars. They built nicer conveniences on top of horizontal capability. And they're also horizontal. So where do they go? Nowhere. There's nowhere to go. I had coffee with a VP at LinkedIn last week. Not a technical guy at all. He told me a fund was shopping around an allocation to invest in a popular vibe coding platform. His response was four words: "Why wouldn't Claude do this?" He didn't invest. The vertical products that survive are going to look nothing like a chat interface. They'll understand the problem so deeply that they can surface what matters without being asked. Ambient. Always on. Specific to your business. And the horizontal platforms structurally cannot do that. Anthropic is not going to build an ambient system tailored to every worker at every company for every job function. But a vertical product can. For a specific set of problems. For a specific kind of company.

@geoffreywoo Collapse is an AI tell, no idea why they love that word so much

announcement: i want to fund founders attacking one ugly truth: most internal software only existed because coordination across mediocre human labor was expensive. agents don’t just automate the task. they collapse the reason the category existed. if you’re building where cheap intelligence deletes middle layers instead of adding another dashboard, i want to see it.

@dr_alphalyrae Don’t see the issue because I have zero interest in working with most people (bar way too high) So everyone I meet defaults to a net new non-transactional friend

but seriously has anyone in the bay area figured out how to make net new, non-transactional friendships as an adult?