ashu garg

Enterprise VC @FoundationCap | Early investor in @databricks @tubi & 6 other unicorns- @cohesity @eightfoldai @turingcom @amperity @alation @anyscalecompute

[ashu garg @ashugarg]

https://t.co/DBjBSJpFTu

[ashu garg @ashugarg]

Many graduates I talk to are asking some version of the same question: is the game over?

It’s a fair fear. AI writes 75% of new code at Google and north of 90% at Anthropic. The on-ramps that used to look safe - first-year tracks in banking, consulting, law, and programming - no longer do. Computer science enrollment just posted its steepest drop on record.

But the frontier always looks claimed to the people standing closest to it.

Consider the class of 2002. The dot-com bust had just happened, and Yahoo, AOL, eBay, and Amazon looked like they had absorbed all the available oxygen. The smart move was to go to business school or get a “real job” in finance.

Then came Facebook in 2004, YouTube in 2005, the iPhone in 2007, and Airbnb, Uber, Instagram, Stripe, and Databricks soon after. Most of the companies that defined the internet for two decades were built by people who, in 2002, were told they were too late.

Twenty years from now, people will look back on this stretch of AI the way we look back on dial-up. You are earlier than you think.

More on what that means for the class of 2026: https://t.co/lRMviApPOj

[ashu garg @ashugarg]

A few more pieces of advice for new grads here: https://t.co/ryKHJhrTLs

[ashu garg @ashugarg]

@nair_shreyas That is so kind of you. Made my day🙏

[ashu garg @ashugarg]

Playing it safe is itself a bet, and usually a worse one than it looks.

This matters most if you just graduated, because the math tilts hardest toward taking risks in the years right after school.

Loss aversion is a powerful force, but over the course of a career, risk and return are tightly correlated. Consistently choosing the higher-upside path is one of the biggest determinants of how far you will go.

What you want to look for is convexity: risks whose downside is capped and whose upside is unbounded. Joining a startup that folds often teaches you more than it costs. Sending a thoughtful email to someone you admire invites nothing worse than silence.

That asymmetry will never tilt further in your favor than it does right after graduation, when you have the least to lose, the most time to recover, and the longest runway for good decisions to compound.

Taking risks also improves your luck. Luck is a collision, an unplanned moment when a person, idea, or opportunity crosses your path. These collisions aren't completely random: you can engineer far more of them than you think.

A few ways to do this:

Work for people who attract talent, because their orbit will become yours. Share what you're learning in public to draw like-minded people toward you. Be useful before you're asked, so you build a reputation for taking initiative. And follow up after you've been ignored: a surprising number of connections happen on the second, third, or even fourth attempt.

Each of these habits puts you in the path of more collisions. Over time, that widens the surface area of your luck.

[ashu garg @ashugarg]

A few more pieces of advice for new grads here: https://t.co/ryKHJhrTLs

[Anshu Sharma 🌶 @anshublog]

Today, we are once again on the Inc best workplaces list.

I love this one because Inc actually conducts anonymous employee surveys to pick Best Workplaces list.

[Dipanjan Dey @Dipanjan_Dey]

Introducing Kombai 2.0 - the first AI design engineer.

We keep hearing that AGI is almost here. Still, we’re stuck with coding agents that don’t have taste and design tools that don’t understand our codebase. Both are artifacts of a world where design and engineering were two different jobs, with a handoff in between.

That world is changing fast. Today, designers ship code, engineers want to escape handoffs, and everyone wants to build tasteful UX.

Kombai is built for this new world.

Design and engineering, finally on the same side.

[Kombai @Kombaico]

The first AI design engineer is here.

Craft tasteful designs on canvas. Code with your stack. Refine in the browser. Keep all synced in your repo and IDE.

A massive thank you to all the users who gave early feedback!

[ashu garg @ashugarg]

https://t.co/DBjBSJpFTu

[Anshu Sharma 🌶 @anshublog]

You can not trust an agent without

a runtime inline policy-enforcement layer that intercepts and transforms data based on identity and intent.

[ashu garg @ashugarg]

Super proud of all the Indian origin founders in SV. Grateful to the US for embracing immigrants. Call 1-800-ashugarg for seed capital😜

[Dan Kopf @dkopf]

Our series on the rise of Indian Americans in the Bay Area has a lot of illuminating charts. This one beats them all. cc @ycombinator @paulg
https://t.co/Gf1zSSezQl

[Vivek Sen @Vivek4real_]

LARRY ELLISON: AI IS RAPIDLY COMMODITIZING BECAUSE MOST MODELS ARE TRAINED ON THE SAME PUBLIC INTERNET DATA.

THE REAL COMPETITIVE EDGE ISN’T THE MODEL ANYMORE — IT’S ACCESS TO EXCLUSIVE, PROPRIETARY DATASETS.

THAT MAY BE THE ONLY MOAT LEFT.

[Tanvir Bhathal @BhathalTanvir0]

Super excited to release OpenJarvis!

Start using it for the most functional, complete, secure, and ipw-efficient personal agent experience!

[ashu garg @ashugarg]

@vassallo @andrewdfeldman @toly Amazing and so well deserved. Congrats 🍾🎈🎉

[ashu garg @ashugarg]

@FDavidsonT Count me in

[ashu garg @ashugarg]

Security and governance concerns are a big reason why AI pilots stall.

You need to have visibility into what the hundreds of agents at your organization are doing, and what they might unintentionally have access to.

This is compounded by the fact that every legacy enterprise application was designed around a human interface.

When agents replace that interface, the architecture of enterprise software has to be rethought from the ground up—permissions, security, even UX.

I’m excited to see where that leads us.

More thoughts in my newsletter: https://t.co/RBS4p72oNS

[ashu garg @ashugarg]

Super proud of my partner @vassallo 🍾🎉🎈

[Steve Vassallo @vassallo]

Proud to be on this year’s Forbes Midas List 100.

But the list isn't really measuring what people think it's measuring.

- It measures @andrewdfeldman , building wafer-scale silicon when the industry said it couldn't be done.

- It measures @toly, betting on a new way to build blockchains.

- It measures Pierre-Damien Vaujour and Alex Greenberg, going all-in on satellite buses before they had clarity on the future.

- It measures Lynn Jurich and Edward Fenster, rebuilding Sunrun overnight when the 2008 GFC broke the model.

These are the people who move the world from what is to what ought to be.

And I feel lucky to play a supporting role.

I love what I do and I love who I get to do it with too.

[ashu garg @ashugarg]

“A procurement decision about a software is actually a group decision,” shares Sumit Johar, CIO at @BlackLine.

When selling software, get familiar with a company’s procurement process so you know which stakeholders need to get on board. It might be more people than you think.

See what other CIOs—including Karthik Chakkarapani (@Zuora) and Saket Srivastava (@Asana)—have to say about the selling tactics that actually move the needle: https://t.co/XnZlYMRR8H

[ashu garg @ashugarg]

Full essay here: https://t.co/iNVAYoTbLX

[ashu garg @ashugarg]

Long-horizon autonomy is the next major inflection point in AI.

The longest task an AI agent can complete autonomously has expanded from ~2 hours a year ago to ~12 hours today.

We’re moving from chatbots that respond when prompted to systems that can plan, act, recover from failure, and persist until a job is done.

That changes both how software is used and where the moat lives.

Products need to be built for agents, not just humans clicking through dashboards. Software that isn't legible to an agent is built for the old world.

At the same time, moats are shifting toward the context graph: the decisions, exceptions, and organizational judgment that accumulate over time as a company operates.

Most enterprises don't have this yet. Making an organization’s decision-making substrate legible to agents is one of the more interesting unsolved problems in enterprise AI.

More of my notes in the latest edition of B2BaCEO: