Lakmal Rupasinghe

Founder - Nirva

[Lakmal Rupasinghe @lakmalr]

These are the best youtube channels to learn Cyber Security. Enojoy. @NahamSec @TCMSecurity @LiveOverflow @stokfredrik @Hak5 @hak5darren , Hak5 is the oldest and coolest. nothing beats hak5 #learn #Hacking

[Lakmal Rupasinghe @lakmalr]

Three more checks decide who scales:
⚙️ automate the costliest job, not the loudest
🧰 two systems you own > seven you don’t
🔒 know & secure every automated thing on your floor

[Lakmal Rupasinghe @lakmalr]

🏭 A robot helped build 30,000+ cars at BMW. Toyota and Amazon are running them too.

So why do 80% of US factories still run ZERO automation — and why do most robot pilots never make it past the demo?

40 years on the plant floor already taught us why 🧵https://t.co/r2J00p4NUz via @LinkedIn

[Lakmal Rupasinghe @lakmalr]

The fix is boring on purpose:
• a clear stop a human can hit
• every action safe to repeat
• limited access
• it talks to the line

The robot you can still switch off is the one still running next year.

[Lakmal Rupasinghe @lakmalr]

When an automation project dies, the autopsy almost never says “the robot couldn’t do the job.”

It says the robot couldn’t talk to anything else. No work orders in. No records out. No alarm when it broke.

The robot is an island.

[Lakmal Rupasinghe @lakmalr]

The tech is here. Automate 2026 this week is packed with “physical AI” — robots that see, decide and act on a real line. Robot density has doubled in 7 years.

The machine isn’t the bottleneck. Everything around it is.

[Lakmal Rupasinghe @lakmalr]

Check out the latest article in my newsletter: Four days, three words: the week the best model vanished https://t.co/5XiJh7hZnL via @LinkedIn

[Lakmal Rupasinghe @lakmalr]

The production gap is stark: 4 in 5 enterprises have AI agents in some form. Only 1 in 9 runs them unattended in product…

[Lakmal Rupasinghe @lakmalr]

The panic is the bug.

[Lakmal Rupasinghe @lakmalr]

Check out the latest article in my newsletter: Your Firewall Wasn’t Hacked. It Was Logged Into. https://t.co/BSzsFlNG1A via @LinkedIn

[Lakmal Rupasinghe @lakmalr]

Check out the latest article in my newsletter: The IPO Edition — Three IPOs, Six Months, $3.65 Trillion https://t.co/ILJAkcCLDI via @LinkedIn

[Lakmal Rupasinghe @lakmalr]

The most fundable AI idea sitting unbuilt right now:

SOC 2 for AI agents.

Every deployer is liable for what their agents do. EU AI Act + state rules require risk management.
No incumbent owns the category.

Market: $7.6B → $183B by 2033.

Someone's going to build this.

[Lakmal Rupasinghe @lakmalr]

This is a fascinating direction from @karpathy.

The idea of autonomous research loops where:

• Humans iterate on prompts
• Agents iterate on code
• Training runs continuously searching for better architectures
This is essentially “self-improving research infrastructure.”
The real opportunity is building systems where AI agents become research collaborators — discovering architectures, optimizers, and datasets faster than human teams.

I’m increasingly convinced the next generation of startups will look like AI research factories, not traditional software companies.

#AI #Agents #ResearchEngineering

[Andrej Karpathy @karpathy]

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then:

- the human iterates on the prompt (.md)
- the AI agent iterates on the training code (.py)

The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc.

https://t.co/YCvOwwjOzF
Part code, part sci-fi, and a pinch of psychosis :)

[Andrej Karpathy @karpathy]

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then:

- the human iterates on the prompt (.md)
- the AI agent iterates on the training code (.py)

The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc.

https://t.co/YCvOwwjOzF
Part code, part sci-fi, and a pinch of psychosis :)

[Lakmal Rupasinghe @lakmalr]

🏀 Check out the latest article in my newsletter: Industry Automation Coach - Episode 2 https://t.co/iGG15ibiW4 via @LinkedIn

[Lakmal Rupasinghe @lakmalr]

@aureliengeron Fantastic work. The visualization of the dimensions ($B, L, d_m$) throughout the block effectively demystifies the Mamba architecture. It’s rare to find such high-quality educational material on these newer State-Space Models made freely available. Much appreciated!

[Lakmal Rupasinghe @lakmalr]

https://t.co/efdnEwmYhc

[Lakmal Rupasinghe @lakmalr]

The "Paradox" Angle (High Impact)The 2025 AI Paradox: 📉 Data Supply: Depleted by ~2028 📈 Capital: $13T GDP boost by 2030

We can't just "read" more internet. We have to "think" deeper. The shift from Big Data ➡️ Big Compute is here. 🧠⚡️

#AI #TechTrends #GenAI

[Lakmal Rupasinghe @lakmalr]

Check out the latest article in my newsletter: AICoach Newsletter - Episode 3 https://t.co/xMwSKwcCjV via @LinkedIn

[Lakmal Rupasinghe @lakmalr]

@doolaHQ @DanielBellCEO @ArjunMahadevan @corylgill @tomasalonso_me @subscriptiondoc 📣 Growth & Creative: @dtc_alchemist, @lorenzo_pravata, @TheAdBrownie, @EmailEngineers

[Lakmal Rupasinghe @lakmalr]

You become the average of the 5 people you spend the most time with.
I’m upgrading my timeline with world-class operators:

🏢 Ops & Structure: @doolaHQ, @DanielBellCEO, @ArjunMahadevan 🛍️ DTC & Ecomm: @corylgill, @tomasalonso_me, @subscriptiondoc