Varun Panesar

Leading product design for the future of OTT and video experiences. I also run a passion project called Forth. IG:

[Varun Panesar @varunpanesar]

I build components and screens for our design system without opening Figma. Describe the intent, get system-compliant output. Took a year and Claude to get here. Worth it.

[Varun Panesar @varunpanesar]

There was a wave of "bootcamp" designers who did bootcamps for a few weeks and started off as product designers. I have nothing against them and some of them did produce good design, but majority of them just had output equivalent to what we call as today's AI Slop.

Such designer's jobs will be taken by AI.
Learning the craft is even more important now.

[Varun Panesar @varunpanesar]

Figma's problem isn't AI. It's that the file is expensive to read.

A machine has to take in a lot of it to work out what you saw instantly.

Close that and the tool has years in it. Don't, and the deliverable moves on without it.

[Varun Panesar @varunpanesar]

The best argument for Figma has almost nothing to do with design.

It's the only room where someone who isn't a designer can point at something and say no, while changing it still costs nothing.

[Varun Panesar @varunpanesar]

I don't think we have reached a stage where a developer says that they have "Vibe Coded the whole thing" and it can be entirely trusted. Especially enterprise level stuff where security is paramount.

Human in the loop is the most critical factor for trust and quality.

[Varun Panesar @varunpanesar]

Figma is still the source of truth for what a component actually is.

Not because it's the best possible format. Because it's the one everyone already agreed on.

That's worth more than it sounds.

[Varun Panesar @varunpanesar]

With Meta Muse, the bigger shift might be voice.

Lots of people already send WhatsApp voice notes without anyone teaching them. If talking to an AI feels the same, there's no reason it stays on phones. TV feels like the obvious next place.

(Can't try Muse yet. US-only, I'm in Budapest.)

[Varun Panesar @varunpanesar]

If the design system underneath is in good shape, AI builds a feature flow in Figma that's genuinely usable.

If you're inventing the interaction, it isn't. You have to look at it. Figma is still where looking happens.

[Varun Panesar @varunpanesar]

Figma's competition was never another design tool.

It's the build. Once a working version is cheap enough to make, a picture of one stops being a deliverable.

It's still the cheapest place to be wrong, and that's what keeps it.

[Varun Panesar @varunpanesar]

Meta Muse is quite interesting because for most people, AI is still something you chat with.

An AI that actually books the flight or sends the email is what gets their attention. Give it a face and a voice, and it stops feeling like software.

[Varun Panesar @varunpanesar]

Figma is still where I decide. It stopped being where I hand over.

Those used to be the same act. That's the entire change.

[Varun Panesar @varunpanesar]

Meta just gave its new AI agent, Muse, a face.

A bit Clippy, sure (remember Ms Word?). But for people who've never used AI, a character is much easier to start with than the sparkle icon every AI product uses. Duolingo already showed that with its owl.

[Varun Panesar @varunpanesar]

The moment the decisions are made and the file is still the deliverable, it's a picture of a product being passed around as if it were the product.

The tool was never the argument. Being wrong cheaply, before it costs anything, is the argument.

[Varun Panesar @varunpanesar]

What's interesting about tools like Penpot isn't that they're open source.

It's that the output is code rather than a picture of code.

Layout kept as the rule that produced it. Values with names instead of numbers. Less to read, more meaning per line.

[Varun Panesar @varunpanesar]

A design file mostly records where things sit.

Hand that to a machine and you pass over a great deal of detail that says very little. It reads all of it to work out what was obvious to anyone looking.

That's the handoff cost nobody prices in.

[Varun Panesar @varunpanesar]

@Google @Android Now Android users have another thing to argue about with iOS users. 😄

I am looking forward to see what this is capable of.

[Varun Panesar @varunpanesar]

AI can put a screen together from parts that have already been decided, faster than I can.

What it can't hand me is a cheap place to be wrong.

A new interaction, a flow nobody has walked. I can't type my way to that. I have to see it and change my mind three times.

[Varun Panesar @varunpanesar]

"Figma is on its way out" is half true.

It isn't losing its job. It's losing one of the two jobs it was holding at once.

Assembly is close to solved. A cheap place to be wrong is not.

[Varun Panesar @varunpanesar]

Feature design does not need to start on a canvas anymore.

Codified design rules turn PRDs directly into staging pull requests.

The interface gets tested in code instead of debated in mockups.

[Varun Panesar @varunpanesar]

After designing for Video/OTT, I have understood

AI can instantly match conversational intent to an entire film catalog. The recommendation is accurate.

Users still won't press play until the poster artwork or a 3-second preview does its job.

Matching intent is solved. Convincing someone to watch is still visual merchandising.