Shantanu | Product x AI

Building things that matters.

[Shantanu | Product x AI @Shan0o]

@iamnirman @Microsoft Great stuff!! Congrats

[Shantanu | Product x AI @Shan0o]

@simonw I remember there used to be Netiquette, now we need Aitiquette

[Shantanu | Product x AI @Shan0o]

I wish there was a switch account feature in @claudeai app

[Shantanu | Product x AI @Shan0o]

@trq212 This is great, I saw it in your CWC workshop.

I happened to ask, how do you share these while collaborating with larger team?

Drop’em on slack?

[Shantanu | Product x AI @Shan0o]

I was making the same mistake until I learned about caffeinate

[Cormac @cormachayden_]

software engineers before vs after agents

[Shantanu | Product x AI @Shan0o]

Vibe coding discourse feels novel?
We had this exact argument when compilers took over from hand-written assembly.

Compilers had bugs. Zero-days. Optimization gaps.
We didn't abandon them. We built better ones, added tooling, and evolved our trust model.

The question isn't "should we trust AI-generated code?"
It's "how do we build the compiler layer on top of it and abstract ourselves from the code itself as we did with assembly?"

[Shantanu | Product x AI @Shan0o]

https://t.co/nHTPiUkq3z — “The Reasoning Trap” by Yin, Sha, Cui & Meng, ICLR 2026.

[Shantanu | Product x AI @Shan0o]

The smarter the model, the worse it knows its own limits.

ICLR 2026 showed it — more reasoning training increases tool hallucination in lockstep with task performance.

Capability is easy to benchmark. Restraint isn’t.

What does “knows when to stop” even look like in agent evals?

[Shantanu | Product x AI @Shan0o]

Redwood City at @StartupGrind!
If you're here grinding with 5,000+ other founders, builders & investors, let's grab coffee.
#StartupGrind

[Shantanu | Product x AI @Shan0o]

@potterlab Is it still available?

[Shantanu | Product x AI @Shan0o]

@martin_casado API key? No, Account number and routing number

[Shantanu | Product x AI @Shan0o]

@ky__zo ✌️ check on both

[Shantanu | Product x AI @Shan0o]

This is amazing, this week I was teaching my dad (anesthesiologist) ai and finding a great starter,

I wouldn't have gotten anything better than this. Thanks @yourgirlhils

[Lenny Rachitsky @lennysan]

"I keep meaning to really dig into AI, but I can't seem to find the time. I'm just so busy."

If that sounds like you, @yourgirlhils and I made something just for you.

Introducing the Couch-to-5K for AI—a free, fun, 30-day program that finally turns "I should get better at AI" into a daily habit.

Fewer than 10 minutes a day. Builds your AI practice progressively.

This will accelerate your AI learning more than anything else you've tried.

Learn more: https://t.co/riBr61bDrA

Get started now: https://t.co/bwD1CBx3Tr

[Shantanu | Product x AI @Shan0o]

The stack is:

Model → Chip → Electricity

Everyone invested in the top layer.

Same pattern as cloud.

AWS was not the best cloud app in 2010.

[Shantanu | Product x AI @Shan0o]

The AI race isn’t being won in a lab.
It’s being won on the power grid.

Let me explain the stack very few are talking about. 🧵

[Shantanu | Product x AI @Shan0o]

Chips need electricity.

One LLM query uses 10x the electricity of a Google search.

We scaled that to a billion users.

[Shantanu | Product x AI @Shan0o]

Thrilled to attend @AnthropicAI code w/ Claude event on 7th May

[Shantanu | Product x AI @Shan0o]

@GoogleDeepMind just released https://t.co/gFPjg5CW03

Excited to see this coming live…