Hamza Farooq

Founder @traversaalai
Adjunct Professor @ucla
Bootstrapped to $1MM

[Hamza Farooq @athamzafarooq]

Hi, I'm new to X and looking to #connect with the #buildinpublic and tech communities.

I’d love to meet folks in:
🚀 SaaS & Startups
📷 Web or Mobile Developer
📷 Builders

What are you working on?

[Hamza Farooq @athamzafarooq]

Running a series on bootstrapped founders on my 15k substack.

If you want to be featured , drop your startup here.

Substack: https://t.co/hFd91a9wsa

[Hamza Farooq @athamzafarooq]

AP x Swatch Royal Pop broke the internet this weekend. So I built it a website. In one afternoon.

Github: https://t.co/k9H9weUDZ9

[Hamza Farooq @athamzafarooq]

Karpathy used the phrase "AI psychosis" on a recent pod. I've seen it in every cohort I've run. It's the moment someone demos an agent that worked once and calls it reliable. The gap between "it did the thing" and "it will do the thing" is where most AI projects fail.

[Hamza Farooq @athamzafarooq]

Someone asked why I add "Co-Authored-By: Claude" to commits. I didn't have a clean answer. It's not legal. It's not performance. In six months I'll open that repo and want to know: did I understand this code, or just accept it? The attribution is a note to future me.

[Hamza Farooq @athamzafarooq]

I've been teaching ML long enough to remember when "building AI" unambiguously meant training. Now it might mean picking the right open model, doing the right RL, and shipping a product people actually use. None of those is lesser. They're just a different kind of hard.

[Hamza Farooq @athamzafarooq]

Cursor's Composer 2 is built on Kimi K2.5. Kimi confirmed it. Cursor's CEO said they're "not a model provider and not an app maker." I think that's the most honest description of where most serious AI companies actually land in 2026.

[Hamza Farooq @athamzafarooq]

The part worth arguing about isn't "did they use an open model." It's: their moat is the domain-specific RL on coding tasks + the product layer. Not the base weights. That's a real bet. It just requires believing the product + fine-tuning gap stays wide enough to matter.

[Hamza Farooq @athamzafarooq]

I add "Co-Authored-By: Claude" to commits by default now. Didn't think much about it until someone asked me why. It's not for legal reasons. It's not that I think the model cares. I think I want a record of how the code actually got written. The git log is a lie otherwise.

[Hamza Farooq @athamzafarooq]

@clairevo @lennysan Submitting a knowledge graphs implementation: https://t.co/43TBSFVDVZ

[Aakash Gupta @aakashgupta]

.@athamzafarooq on why context engineering is now more critical than prompt engineering
Most AI PMs focus on prompt engineering—writing better instructions to get better outputs.

But that only works for generic responses. The moment you need personalization, prompt engineering falls apart.

Consider a finance agent. One user wants conservative S&P 500 investments. Another wants high-risk crypto trades. How does the same LLM give both users relevant advice?
Context engineering solves this by layering multiple information sources: system prompts, user prompts, long-term memory from past interactions, and relevant data pulled from RAG.

"Prompt engineering is what you tell an LLM. Context engineering is how you design the instructions for your LLM. That's the beauty of having the knowledge of context engineering because it makes your entire ecosystem dance."

Fine-tuning works differently—it's task adaptation. You train the LLM on thousands of examples to specialize it for specific outputs, like generating Python code or understanding pharma industry vocabulary.

Lesson: Context engineering is now more important than prompt engineering because modern AI products require personalization at scale. You need to orchestrate memory, retrieval, and prompts simultaneously to deliver relevant responses to each user.

[Rapid Response 47 @RapidResponse47]

Corporate lawyers and others with agendas are creating a lot of FAKE NEWS around President Trump's H-1B Proclamation, but these are FACTS:

1. The Proclamation does not apply to anyone who has a current visa.

2. The Proclamation only applies to future applicants in the February lottery who are currently outside the U.S. It does not apply to anyone who participated in the 2025 lottery.

3. The Proclamation does not impact the ability of any current visa holder to travel to/from the U.S.

[Hamza Farooq @athamzafarooq]

@hthieblot Data OS for Supply Chain
https://t.co/09ZlKgTEa9

[Hamza Farooq @athamzafarooq]

@yanabana @a16z Intelligence that runs your data
https://t.co/09ZlKgT6kB

[Hamza Farooq @athamzafarooq]

@cliftonsellers Intelligence that runs your data
https://t.co/09ZlKgT6kB

[Hamza Farooq @athamzafarooq]

@PLBompard Intelligence that runs your data
https://t.co/09ZlKgT6kB

[Hamza Farooq @athamzafarooq]

@atShruti I think they stopped being boring a while ago, @elonmusk used to be on H1b too

[Hamza Farooq @athamzafarooq]

@aryanvs_ @a16z we are building Data OS, already working with three design partners :)

[Hamza Farooq @athamzafarooq]

Video of people offloading Emirates flight after the executive order was passed.

Here's the video: https://t.co/quZ5TJz5CA

Not sure about the authencity of the video

[Shreyas Doshi @shreyas]

Something to consider:
If these poll results accurately represented the distribution of employees in your company, what would be the state of your company?

a) Way better than it is today
b) About the same as today
c) Way worse than it is today