Balu

fashion, econ/policy, culture, supply chain nerd. multicultural-lingual immigrant. building something new. ex @tommyhilfiger @Bain @Wharton @goldmanSachs #AAUSF

[Usman Gul @usmangul]

💥 New Product Launch 💥

After two years of development, @metalraise is launching Autopilot -- our automation platform that draws on proprietary intelligence to run the round process for founders.

Our thesis is simple -- in the future, founders that are best positioned to raise will not run the round process themselves. Instead, they will use the right tools to automate in ways that create unique advantages.

From research, round collateral and pipeline formation to building calendar density, Metal Autopilot brings precision via proprietary intelligence and speed via AI-driven automation.

Link to join the waitlist below.

[hari raghavan @haridigresses]

@nikunj Nudged me to write this :)

https://t.co/Q94QbyhL43

[Balu @whynotbalu]

wise

[Alfred Lin @Alfred_Lin]

https://t.co/Der6H48R8U

[Will Manidis @WillManidis]

https://t.co/55JiWR8AEu

[Balu @whynotbalu]

@jaltma 2026 version of ‘which companies are safe from X?’

1970s: computers
1980s: software
1990s: internet
2020s: AI
2030s: ?

Long enough timeline...doesn't AI eat ANYthing that only exists b/c of computers, software, or the internet? ‘Safe from AI’ logic feels like self-soothing ;)

[Balu @whynotbalu]

@katekiewel Miami Beach vibes 😜

[Aaron Levie @levie]

https://t.co/UVlRDGJu5I

[Muratcan Koylan @muratcan]

After reading Karpathy's YC talk and year-end review, it's clear that there's a real, defensible layer between foundation models and end users.

The pattern he calls out is "Cursor for X". It's what Cursor revealed about how LLM apps should be architected.

Karpathy identifies four things these apps do:
1. Context engineering. The app decides what goes into the context window. You don't manually copy-paste code files and error logs. The app does the retrieval, embedding, and curation. This is a ton of hidden work.

2. Multi-call orchestration. Under the hood, there are embedding models for your files, chat models for reasoning, models that apply diffs. The user sees one experience. The app runs a whole orchestra.

3. Application-specific GUI. This is undersold. Text is hard to audit. Seeing red/green diffs uses your visual system, which is way faster than reading. Command+Y to accept, Command+N to reject. You're not typing "yes I accept this change" into a chat box.

4. Autonomy slider. Cmd+K changes a small chunk. Cmd+L changes a file. Cmd+I does more autonomous work. The user controls how much the AI does at once.

The human verification step is the bottleneck. AI generates instantly. But you're still responsible for the output. If you get a 1,000 line diff, you have to verify it actually works, introduces no bugs, and has no security issues. That takes time.

So there are two levers:
- Speed up verification (GUIs, visual diffs, good UX)
- Keep the AI on a leash (smaller chunks, clearer prompts)

If your prompt is vague, AI does something unexpected, verification fails, you re-prompt. You're now spinning in a loop. Better to spend more time on a precise prompt that increases the probability of successful verification on the first pass.

"This is the decade of agents"

Right now, with where models are, you want suits. Partial autonomy products where the human stays in the loop, the generation-verification cycle is fast, and there's an autonomy slider you can push right over time.

Suits augment. Robots operate autonomously. The suit still has a human making decisions. The robot flies around on its own.

Build Iron Man suits, not Iron Man robots.

[Josh Kopelman @joshk]

Instead of just throwing AI at its sales team and expecting results, @NotionHQ did something different: they turned an engineer into a full-blown BDR for a month.

This immersion led to a much clearer understanding of the right problem to solve with AI.

“From the outside, it seems obvious we should make the sales team move faster by doing account research for them. But when Theo came in and did research, where he ended up was actually account prioritization,” says @PraveshMistry, Notion’s Head Global of Sales.

The result was an internal tool that gives reps product signals they use to better prioritize which accounts to reach out to, while also providing them with customized messaging to edit and use in that outreach.

If you’re developing internal AI tools, this process is a clear example of how to find the real problem before writing a single line of code.

Read the story here: https://t.co/lhqU8VNLwu

[Usman Gul @usmangul]

Most founders end up talking to the wrong investors.

As a result, they end up getting the wrong feedback from the market.

We built @MetalRaise to help founders identify investors that are actually likely to invest, using real historical data.

In this video, we explain with examples how it works:

[hari raghavan @haridigresses]

I started saying “gippity” instead of ChatGPT and it’s unexpectedly catching on with friends, family, colleagues, etc. 😂

[Balu @whynotbalu]

Boom! The power of raise intelligence.

[Usman Gul @usmangul]

💥 New Announcement 💥

A few months ago, we started working with the @speedrun team to explore how Metal could benefit founders in their programs. Today, a vast majority of founders in the latest @speedrun batch have become paying customers.

In the below video, our editorial team narrates how things played out!

[tobi lutke @tobi]

What a legend

[Balu @whynotbalu]

@thatguybg @microHQ super excited to give this a shot @thatguybg ! amazing launch video

[USCIS @USCIS]

The International Entrepreneur Rule lets startup founders & their spouses work in the US. We are actively working new applications with no backlog. Check out our new FAQs:
https://t.co/c2QnGspKXZ

[Vinod Khosla @vkhosla]

Hope you are wrong. @realDonaldTrump will set back this country's institutions, protections, climate for many years past his presidency!

[Aman Dhesi @amansplaining]

Introducing Superpipe Studio: a free and open-source tool to curate datasets, run evals, conduct experiments and optimize your LLM pipelines for accuracy, speed and cost.

@benscharfstein and I have spent the last year building with LLMs and working closely with a number of companies and encountered the same problems repeatedly:

- evaluating LLM software is hard, because...

- collecting high-quality, representative, correcty-labeled data is hard...

- which makes it hard to objectively compare different techniques, models and parameters...

- which prevents continuous iteration and improvement of LLM software

If you want to build a winning AI product - it all comes down to continuous evaluation and optimization. Unlike traditional software, you can't build it once, write some unit tests and sleep well at night.

You need a virtuous cycle where product usage generates data that goes into your eval system, which is used to evaluate new models/techniques or fine-tune your own model.

@HamelHusain captured it best in this diagram --

We tried existing tools in the market and found none of them really worked well to implement this virtuous cycle, so we built a lightweight tool for internal use: Superpipe Studio.

It helps you do 3 things:

1. curate and manage datasets

2. run experiments and compare them on accuracy/speed/cost

3. monitor your production pipelines (and expand your datasets with production data)

We think everyone should have these abilities, so we're making Studio completely free and open-source with no restrictions.

Studio is a work in progress and rough around the edges but still robust. We encourage you to try it and modify it to suit your needs.

Building a winning AI product is more about process than about any one feature or idea — and it all starts with high-quality labeled data and a robust evaluation and optimization system.

[Balu @whynotbalu]

Opportunity of a lifetime

[Josh Kopelman @joshk]

Are you a creative and unconventional thinker, yet also highly analytical and logical?

Come be my next Chief of Staff!

This person will spend the next 2+ years working alongside me in all areas of venture capital. From inbox ✉️ to investment review 💰 through exit 💸!

Learn more -- and apply here:

https://t.co/b2hz2Zgu1f

[Jack Altman @jaltma]

Founder equity is a sensitive subject but I wanted to share a couple thoughts I have on the subject that I think align everyone’s interests better and are just generally more fair than the current status quo:

1. I think founder vesting ought to be longer than 4 years. It’s just not really enough time from day zero, there is too much work to do ahead at day 1,460. This might sound “founder unfriendly” but when you think about it a bit longer it’s really not. A startup is almost never done in less than 4 years, and if it is everyone will get vested anyway. So really this is friendly to the founders who stay longer and do the work between days 1,460 and 5,000+. There are a lot of scenarios where a founder leaves after 3 or 4 or 5 years, with 20% of the company, and the remaining founder now has a huge amount of work left to do with a big hole on the cap table that no one sizing up the situation would really consider fair.

2. After the initial vesting period, I think founder refreshers tend to be way smaller than they ought to be. VCs will go through humongous contortions to tell founders why they don’t deserve the same refresh grant that it would cost to hire a replacement for their role. And at the end of the day all the arguments for paying the founders less eventually seem to boil down to “because we can get away with it”, which is not really a great way to do business. Same goes for any other exec here imo.

So that’s where I think things ought to be — longer initial vesting periods, larger refresh grants — but unfortunately there are a lot of incentives making it tough to make changes on either.

[Aman Dhesi @amansplaining]

@sama Creating material abundance is the long-term arc of all technological progress.

In developing countries people embrace this because they see their lives improve in their lifetime. In developed countries many people take their quality of life for granted