The AI tool that we are about to cover ran on a server that has never been switched off, regardless of a storm, a grid spike, or a regional outage.

The way Google guarantees that is 100 million lithium-ion cells, one for nearly every rack in its global data center fleet, acting as instant backup the moment grid power flickers.

They switched from lead-acid to lithium in 2015 because it holds twice the power in half the space. A decade later, that number is still climbing.

This is what lithium demand actually looks like at scale before you even get to EVs. Morgan Stanley projects an 80,000-ton shortfall this year.

By 2040, demand is expected to be 5x what it is today.

EnergyX has patented technology that recovers up to 3x more lithium than traditional methods, 500x faster than conventional evaporation ponds.

General Motors has already backed them, and you can invest at $11/share before their price increases after 2/26.

Replit launched an animation tool this week.

You describe a video, and it builds it.

Under the hood, it’s all React and animation libraries but you will never see any of it.

We just get a looping preview, which can be refined through conversation, and the final output can be exported as an MP4.

This is the same bet Remotion made: motion design can now be done via code.

Replit just made that bet more accessible. So I gave it a spin.

Before we jump in, let's catch up on AI this week:

1. Docura Health

An AI-native med-legal platform that automates workers’ compensation documentation tasks, turning thousands of pages of medical records into compliant reports and summaries in minutes rather than hours.

2. Gemma

An AI personal assistant that actually calls you with reminders instead of just sending notifications, syncing with your calendar and letting you reply by voice.

3. Return Signals

It offers proactive, high-touch customer support for e-commerce brands by combining data-driven signals with live agent workflows to catch issues early and improve shopper experiences.

Go to replit.com, sign in (free tier is enough) and select Animation from the dropdown under the prompt box.

Type what you want, the agent builds it, shows a looping preview which you can refine through conversation, then export when it looks right.

Export options include video quality (720p or 1080p), frame rate (30 or 60fps), and it downloads as MP4.

You can also click directly on elements in the preview and edit by hand.

It's powered by Gemini 3.1 Pro on the backend.

I ran three tests plus a bonus one.

Test 1: Cold Prompt

Prompt: 
Add Bengaluru visuals as backgrounds behind the text.
Scene 1 - aerial view of Bengaluru's tech skyline at dusk.
Scene 2 - lush green Cubbon Park with Vidhana Soudha in the background.
Scene 3 - neon-lit Bengaluru street scene at night.
Keep the same text and transitions, just layer city imagery underneath.

Result: This is a decent baseline.

It gave me a skyline at dusk, Vidhana Soudha in frame, Kannada signboards on a neon street and even added "THE HEART OF INNOVATION" as a subhead.

One catch: that neon street looked more like night-time Tokyo. The wider roads, warmer lighting, and cyberpunk vibes looks great in a promo but it doesn’t look like Bengaluru at all.

Test 2: Reference-Fed Prompt

This is the biggest quality lever I found, and Replit doesn't mention it anywhere in their docs.

I found a kinetic typography video on YouTube I liked, screenshotted the key frames, uploaded them to ChatGPT and asked it to describe the motion graphics style: font weights, animation techniques, and pacing.

Then I condensed that into a focused prompt.

Prompt: 
Recreate the following motion graphics style for a 10-second Bengaluru city promo: Stomp typography with bold condensed sans-serif typeface, all-caps throughout.
Mix of font weights — hero words in ultra-bold filling the frame, supporting text smaller and lighter.
Text slams into frame with hard, sudden entrances, no easing. Sequential word-by-word reveal with physical momentum. Alternates between solid dark backgrounds for text-only frames and city skyline imagery behind text. High contrast, percussive rhythm, hard cuts between scenes.
Content — three scenes:
Scene 1 - 'BENGALURU' as massive frame-filling kinetic typography slamming into frame over aerial tech skyline at dusk.
Scene 2 - 'Where 90,000+ Startups Call Home' with stomp text reveal over Cubbon Park with Vidhana Soudha.
Scene 3 - 'The Pulse of India's Tech Future' slamming in over neon-lit night street scene. Color palette: deep navy (#1B2A4A), warm amber (#F5A623), off-white (#FAF9F6).
Give each scene time to breathe.

Result: The stomp style came through immediately.

BENGALURU slammed in with 3D depth shadows, frame-filling, the word-by-word cascade matched the rhythm and the street scene actually looked more like real Bengaluru this time.

But the scene compositions were less polished than Test 1's iterated version.

Test 3: Say what you want

After two structured tests, I dropped all the prompt engineering and just... talked to it.

Prompt:
Create a cinematic promo to showcase Bengaluru (city in India) as the next Silicon Valley.

Result: Miles better than Test 1.

When you stop over-engineering the prompt, the tool starts making creative decisions.

Things like transitions I wouldn't have thought to specify, timing choices that just felt right and design sense that exceeded what I would have specified if left to my own.

I'm not a motion designer and I don't know the nitty-gritties of easing functions and scene composition, but the tool clearly does.

And when you get out of its way, it shows.

Test 4: The Workshop (Bonus)

Our engineering masterclass is running this weekend actually, if you're reading this, it might be starting today.

I used Replit Animation to generate the promo for it, and this one came out pretty decent. The same promo I'd normally wait a week for now done in 10 minutes, how cool is that.

Speaking of the workshop

It’s something worth considering if you're a professional working in tech or want to get into it.

Two days, live, and hands-on.

You walk out knowing how to actually build with AI and work with agents, automation, real workflows

Remotion first made the bet of creating videos simply through code.

If it's code, it can be versioned, templated, automated, and plugged into a pipeline.

Your brand guidelines become a config file and your weekly product update becomes a cron job, but it requires you to write React, which most of us don't.

Replit just got rid of that issue altogether.

You describe what you want, the agent writes the React, and you get your motion graphics back.

Hit reply with what you make, I am curious what comes out when you throw your own brief at it.

Until next time,
Vaibhav 🤝