KPAI Weekly: Issue 34

Let’s get started—plenty to cover this week.

NVIDIA is back and stretching the lead. Between a trillion-dollar demand projection, one of the biggest compute partnerships in AI, and a telecom deal that pushes AI closer to the edge, NVIDIA had another week that reinforced just how much of the stack runs through them. Here’s the recap:

NVIDIA also kicked off its annual developer conference, GTC 2026, by unveiling DLSS 5, which it described as its biggest gaming graphics breakthrough since real-time ray tracing. The company also introduced NemoClaw, an open-source stack designed to help developers run always-on autonomous agents more safely.

Who’s moving, who’s stalling, and who’s climbing: Ordered by production this week.

Interview Outline: This interview explores the "efficiency of atoms" and how AI is being used to solve real-world friction in the physical food and logistics industries. Kalanick discusses the operational challenges of "Cloud Kitchens," focusing on predictive cooking and minimizing "dwell time" to ensure food arrives hot and fresh. He emphasizes that the biggest AI opportunity isn't in flashy chatbots but in the "un-glamorous efficiency" of massive physical industries.

About the Interviewee: Travis Kalanick is the CEO of Atoms and the co-founder of Uber. He specializes in building "real-world" technology companies that use software to optimize physical infrastructure, delivery, and logistics at a global scale.

Interesting Quote: "The edge isn't the model; it’s the data from your own operations."

My Thoughts: We’ll dive a bit more into Travis Kalanicks new startup in the spotlight below but I think this was an interesting representation of AI in the physical world. While ‘Atoms’ doesn’t refer to molecular science, Travis breaks down how AI is used in everyday tasks such as food preparation and delivery. The above quote is relevant across industries and an important one to remember.

Difficulty: Basic

If you listen to or read the noahonai interview this week, you’ll get a taste of how powerful AI’s deep research functionality can be when it’s executed right.

Deep research works by pulling in hundreds to thousands of primary sources—news articles, research papers, blog posts, forum responses—and then running dozens to hundreds of follow-up queries across those sources to fill in gaps, summarize perspectives, or extend threads. It’s not just a summary of the top 10 search results. It’s full-scale, multi-threaded investigation.

What makes deep research so powerful is the concept of time. You could have the smartest, most efficient person in the world sitting next to you, and it would take them days to match what deep research can do in a few minutes. It’s an interesting framework when thinking about AI use cases, some are about productivity, some are about creativity, but almost all of them save us the one thing we can’t get back: time.

Sergey Brin, cofounder of Google, stepped away from his company for a few years and rejoined for the AI race in 2023. Here’s what he said about deep research:

“By default, when you use some of our AI systems... it'll suck down whatever top 10 search results... I could do that myself... But if it sucks down the top, you know, thousand results and then does follow-on searches for each of those and reads them deeply—that’s, you know, a week of work for me. Like, I can’t do that.”

Effectiveness by Company:

Grok and Perplexity are definitely the fastest, they’ll do the job in 3-4 minutes, but it often feels surface-level, or some details are askew. Gemini is very good, but sometimes it’s almost too much, and I find myself needing to sift through to make sense of it all.

GPT generally hits the sweet spot for me. It seems to pull in 90% of what Gemini does, but it presents it in a cleaner, more focused way, especially if you’re trying to build a tight argument or summary. For shorter projects, I’d recommend ChatGPT, for longer projects I’d recommend a combination of GPT and Gemini deep research.

Issue 34 Update: A lot of people know about deep research now, but it is still one of the top AI use cases for me. What I like to do these days is run 4 DR sessions simultaneously (Generally Claude, Gemini, GPT, and Grok), then combine all of that in to one document. Place that document back into Claude, and ask it to put together a concise new version with an emphasis on the data that was similar across the different searches. Claude is my top Deep Researcher, Gemini 2, GPT 3, and Grok 4. GPT changed their output structure and I don’t like it. Still waiting on the specialized deep research API from Claude and Grok in order to fully automate this process.

Atoms — Gainfully Employed Robots for the Physical World.

The Problem: While AI has mastered digital tasks like writing and coding, the "physical world" remains largely un-digitized. Industries like mining, heavy transport, and large-scale food production still rely on manual labor or "dumb" machines, creating a massive bottleneck in global productivity and safety.

The Solution: Atoms builds specialized, task-oriented robots rather than general-purpose humanoids. Their strategy revolves around a "wheelbase for robots,” a standardized mobility platform that can be outfitted for different industrial jobs. The goal is to create machines that are "gainfully employed," meaning they perform high-cycle, productive work in mines, kitchens, and warehouses to drive down the cost of physical goods.

The Backstory: Founded by Uber co-founder Travis Kalanick, Atoms is the rebranded evolution of his holding company, City Storage Systems. The company operated in total stealth for eight years with thousands of employees who weren't even allowed to list the company on LinkedIn. It has now officially absorbed CloudKitchens (food infrastructure) and Pronto (autonomous mining/transport) into a single robotics powerhouse with major backing from Uber and the Saudi PIF.

My Thoughts: Found this very interesting not just because of the nature of the company but also the founder backstory. On the website, Travis details how investors ripped Uber away from him at a vulnerable time in his life. Atoms is on par with the movement of AI into the physical world and I’m excited to follow the company. I think this is the way to go with robotics, as we’re already seeing it in warehouses. Rather than a one size fits all humanoid bot, having robots that can perform highly specialized tasks that will be automated over a long period of time. I have no doubt it will end up being a success.