Shivam Bhatnagar

Intelligence Designer. Designing how AI agents decide, fail, and recover.

[Shivam Bhatnagar @MrShivamB]

@Arnavmj TOOL

[Shivam Bhatnagar @MrShivamB]

@soleio So true. People who have moved beyond just learning AI have forgotten the anxiety, they are actually building. The future belongs to the builders as you always say.

[Shivam Bhatnagar @MrShivamB]

@jietang @Zai_org It’s definitely vision

[Shivam Bhatnagar @MrShivamB]

Making is now free. Being used is not.

A study of 100,000+ developers tracked three generations of AI coding tools. Commits went up 180%, more apps started showing up on the Apple App Store, Google Play, Chrome Web Store, and SourceForge.

Usage did not move. Zero. Across all four stores. The code pile grew, but nobody is using more of it.

All the value moved to the gap between making something and someone actually using it.

The judgment, trust and the question of who it is even for, Almost nobody crosses that gap.

One thing a real person uses beats a hundred things sitting in a repo.

https://t.co/zK2QK06eki

[Shivam Bhatnagar @MrShivamB]

When agents record a video of their work..

[Shivam Bhatnagar @MrShivamB]

The rise of agentic products isn’t hype, it’s AI finally finishing the job instead of just chatting about it.

[Greg Brockman @gdb]

agents make for a surprisingly great product

[Greg Brockman @gdb]

agents make for a surprisingly great product

[Shivam Bhatnagar @MrShivamB]

@Kangwook_Lee The bottleneck was never the scaffolding. It was always knowing what you want the agent to be. Better models just make that bottleneck impossible to hide.

[Shivam Bhatnagar @MrShivamB]

@harshilmathur Right, and that permission model is the unsolved problem. Who decides what cross-app agentic actions are allowed?

[Shivam Bhatnagar @MrShivamB]

@everestchris6 Guide

[Harshil Mathur @harshilmathur]

Hot take: in a few years, apps will assume you have a local LLM with memory
just like they assume always-on internet today.

Every app becomes personal, intelligent and a little different for everyone.

And software as we know it completely changes.

[Shivam Bhatnagar @MrShivamB]

@karrisaarinen Upstream is right, but the genuinely scarce layer is what agents do between the problem and the output. That one's still unnamed.

[Shivam Bhatnagar @MrShivamB]

@charmiekapoor I think that the designers who’ll matter in the agentic era aren’t the ones who adapted, they’re the ones who got curious enough to build.

[Karri Saarinen @karrisaarinen]

@jstn Designers will pm, PMs will code, engineers will design. This is all in the name of progress

[Shivam Bhatnagar @MrShivamB]

@chalaska It’s also knowing whether it should exist.

[Shivam Bhatnagar @MrShivamB]

The hard part of design was never the screens.

AI generates form now, fast, plausible, convincing. Karri Saarinen said it his recent article (Output isn’t design): the real risk isn't bad output, it's mistaking generated output for a solved problem. Form without fit.

Teams have measured design by delivery for years. Screens, specs, handoffs. Visible work. Not the actual work.

The actual work is holding the problem long enough to know whether what got built actually fits. To say "this shouldn't exist yet" and mean it.

That judgment didn't become important when AI arrived.

AI just made it the only thing left.

If your team is shipping faster than ever — how are you deciding what shouldn't get built?

[Shivam Bhatnagar @MrShivamB]

@HeyGen HyperFrames

[HeyGen @HeyGen]

We built our launch video in Claude Code using HyperFrames.

Now it's yours.

Open source, agent-native framework. HTML to MP4.

$ npx skills add heygen-com/hyperframes

RT + Comment "HyperFrames" to get the full source code of this launch video (must follow)

[Shivam Bhatnagar @MrShivamB]

@marianoaavila @OpenAI Congrats