Rupesh Panda

Building applied AI systems for the enterprise. Sharing what works, what doesn’t, and what I’m learning about agentic AI, SAP, and enterprise transformation.

[Rupesh Panda @Rupesh_Panda]

A defining moment in AI history. Jensen personally delivering OpenAI’s first GPU box in 2016 gave the team the much-needed compute to push the early deep learning experiments forward.

Hard to believe how much of today’s AI revolution traces back to moments like this.

[Elon Musk @elonmusk]

That was 10 years ago

Imagine 10 years from now

[Rupesh Panda @Rupesh_Panda]

@vasuman Maybe the goal is not to make every employee an AI power user. The real win is when AI gets embedded into the work itself, almost invisible to the employee.

[vas @vasuman]

AI adoption is fundamentally misunderstood, and this is the reason why it will keep failing.

[Rupesh Panda @Rupesh_Panda]

AI learning from AI sounds efficient.

Over time, it can also lead to model collapse.

Intelligence needs to stay connected to reality.

[Jeff Dean @JeffDean]

Announcing Discovery Loop!

I am very excited to announce that, along with my longtime friends and collaborators @Sanjay_Ghemawat, @OriolVinyalsML and @quocleix, we are founding Discovery Loop (@DiscoLoopAI), a Public Benefit Corporation whose mission is to automate machine learning, science, and engineering to accelerate discoveries and progress. The four of us have worked together for 14 to 30 years, and have helped build some of the world’s most used products, infrastructure and AI models, and we’re excited to turn our attention to this ambitious endeavor.

♾

Learn more at: https://t.co/Rv3LMdLluK

[Shaun Maguire @shaunmmaguire]

Many of my close friends and loved ones don't understand this yet

[Rupesh Panda @Rupesh_Panda]

Markdown files aren’t just documentation.

They capture human judgment.

Great agents don’t load everything.

They load the right context aka .md files at the right time.

[Rupesh Panda @Rupesh_Panda]

People underestimate Markdown files because they look simple.

But a great .md file captures human judgment, principles, and workflows and that context can steer an AI agent more than thousands of lines of code.

[Rupesh Panda @Rupesh_Panda]

Where will AI’s economic value accrue?

• Models provide intelligence

• Harnesses turn that intelligence into reliable business outcomes

• Applications are how businesses consume AI

Agentic AI = Model + Harness

The winners won’t optimize 1 layer. They’ll orchestrate all 3

[Rupesh Panda @Rupesh_Panda]

Same model, different harness, up to a 3.7x cost gap. As agentic workloads grow, harness efficiency becomes a major competitive advantage.

[Rupesh Panda @Rupesh_Panda]

One day you’ll die. That’s not depressing, it’s liberating. Life is precious. Stop worrying what people think.

They’re too busy thinking about themselves. Take the risk. Build. Create. Love. Live. The biggest regret isn’t failure. It’s never fully living.

[Rupesh Panda @Rupesh_Panda]

If AI reaches singularity, history may remember backpropagation as one of the algorithms that made it possible.

It didn’t create superintelligence by itself, but it gave machines the ability to improve from experience, and every subsequent breakthrough has compounded on that.

[NVIDIA @nvidia]

AI security advances when the industry builds in the open, together.

We're introducing the Open Secure AI Alliance with industry leaders to develop new techniques and tools to safeguard software and agents.

By sharing models, tooling and research in the open, we can broaden the community of defenders.

Learn more about the founding members’ contributions: https://t.co/A16oqxs5Ty

[Rupesh Panda @Rupesh_Panda]

Model intelligence is becoming abundant. Turning that intelligence into reliable business outcomes remains difficult.

Frontier labs are building intelligence. The applied AI ecosystem will turn that intelligence into work.

That is where the next wave of opportunity lies.

[Rupesh Panda @Rupesh_Panda]

The supply chain of INTELLIGENCE:

Energy → Data centers → Chips and GPUs → Compute clusters → Training infrastructure → Frontier & Open weight models → APIs and open weights → Applied AI → Business outcomes

[Rupesh Panda @Rupesh_Panda]

We’re spending a lot of time comparing token prices. I think we’ll eventually compare intelligence per dollar. That’s the metric that will matter in long term.

[Rupesh Panda @Rupesh_Panda]

Three equations that simplify Applied AI:

Application = Applied AI = Agentic AI

Agentic AI = Model + Harness

Harness = Context + Tools + Memory + Evals + Guardrails + Workflows

My hypothesis: as models increasingly commoditize, the competitive advantage shifts to the harness.

[Rupesh Panda @Rupesh_Panda]

One AI stack for execution. Model-agnostic enterprise knowledge on top. That’s a simple architecture I think we’ll see more of as enterprise AI matures.

[Rupesh Panda @Rupesh_Panda]

Enterprise AI succeeds when it becomes part of the operating layer, not another application employees have to remember to use.

[Rupesh Panda @Rupesh_Panda]

Directionally, I think enterprise AI is moving beyond reporting.

The real opportunity is AI that can reason across both structured data and decades of unstructured enterprise knowledge.

That’s where the most valuable insights will come from.