In partnership with
🌟 Editor's Note: Recapping the AI landscape from 02/10/26 - 02/16/26.
🎇✅ Welcoming Thoughts
Welcome to the 31st edition of NoahonAI.
What’s included: company moves, a weekly winner, AI industry impacts, practical use cases, and more.
Busy busy week in the AI space!
IMO there are a ton of sci-fi style overreactions going around right now about AI consciousness.
The Anthropic team is behind a lot of it, intentional or not.
Excellent interview this week with a founder that is no stranger to a PC.
^ Great details on the latest in OpenClaw/ClawdBot/Moltbot.
Just the condensed interview below - full version out later this week.
It looks like both Google founders are fully back. Watch out everyone else.
Hackers have been trying to clone Gemini by gathering data through prompts.
Meta is doing something weird again: patent to keep deceased user accounts active via AI.
The “AI Girlfriend” crowd is officially out of control: A New York Bar held a popup weekend for chatbot dates.
Pentagon used Claude to help capture Maduro - More below.
GPT just hit 100M users in India.
I believe AI is incredibly useful and is improving dramatically.
I also believe people undervalue the difficulty in getting from 90% to 99% to 99.9% and so on.
Thought this tweet was pretty funny for the AI + Good Will Hunting fans.
Let’s get started—plenty to cover this week.
👑 This Week’s Winner: Anthropic // Claude
Anthropic wins a loud but structural week. Between a $30B mega-round, rising defense tension, and infrastructure commitments, the company looked less like a startup and more like a national-scale AI platform. Here’s what happened:
Closed $30B Fundraise: Anthropic closed a $30B Series G at a $380B post-money valuation. The company framed the raise around frontier research, enterprise expansion, and massive compute buildout for Claude. It also added former Microsoft and GM CFO Chris Liddell to the board: a move widely interpreted as IPO preparation for 2026. Impressive. I’m very bullish on Anthropic. Excited to see where this takes them.
DoD Dispute: The Pentagon continues to push back on Anthropic over Claude’s restrictions on autonomous weapons targeting and domestic surveillance use. Anthropic has reportedly refused to loosen those safeguards, creating tension. They have started donating money to PACs in favor of AI safeguards. I still think Anthropic is mostly in the right here.
Electricity Price Pledge: Anthropic committed to covering 100% of grid upgrade costs tied to its data center expansion so local ratepayers don’t absorb electricity increases. Good move, others should follow suit if they haven’t already.
In other major news: Anthropic’s Head of Safety resigned with a letter claiming the “world is in peril”. An overreaction on all accounts. CEO Dario Amodei also drew attention for comments acknowledging uncertainty around AI consciousness. I don’t believe he’s referring to consciousness in the truest sense of the word. Meanwhile, Anthropic’s Super Bowl ad campaign (“Ads are coming to AI. But not to Claude”) pushed the Claude app into the Top 10 on the U.S. App Store, and delivered it an 11% boost in daily users, a measurable consumer lift.
From Top to Bottom: Open AI, Google Gemini, xAI, Meta AI, Anthropic, NVIDIA.
⬇️ The Rest of the Field
Who’s moving, who’s stalling, and who’s climbing: Ordered by production this week.
🟢 OpenAI // ChatGPT
Coding Model Launch: OpenAI launched GPT-5.3-Codex-Spark and deployed it on Cerebras hardware, signaling it can run production workloads beyond NVIDIA for speed, cost, and supply flexibility. The model seems to be impressive from what I’m hearing. Doubly impressive knowing it arrived sans NVIDIA.
Enterprise Lockdown Mode: New optional Lockdown Mode limits ChatGPT’s ability to access external tools to reduce prompt-injection risks, alongside “Elevated Risk” labels for higher-exposure features. Makes sense, especially given some of their government work. Smart.
OpenClaw Meet OpenAI: OpenAI brought on OpenClaw creator Peter Steinberger to focus on personal AI agents. The OpenClaw/Clawdbot/Moltbot project will remain open source and maintained by OpenAI. Great move! At the very least, they’re bringing an extremely talented developer in-house.
⚪️ NVIDIA
Blackwell Chip Efficiency: NVIDIA says new Blackwell Ultra delivers up to 50x better performance per watt for agentic AI, with Microsoft and CoreWeave among early adopters. Impressive improvements. Keep em coming.
Lower Inference Costs: Partners report roughly 4x–10x lower cost per token on Blackwell versus Hopper, positioning it as a major improvement in AI economics. Good.
Investors Bullish: Analysts see ~67% revenue growth to ~$65B and solid momentum into NVIDIA’s Feb 25 earnings report. No surprises here.
🟣 Google // Gemini
DeepThink Premium: Google upgraded Gemini 3 Deep Think introducing a high-compute science/engineering reasoning mode. DeepMind says an advanced version helped experts solve 18 standing research problems. Cool! Deepmind has some of the best AI tech available, namely in research.
Alphabet Funds AI With Debt: Alphabet raised about $31.5B in a massive bond sale (including a rare 100-year tranche), signaling it’s willing to finance AI capex like an infrastructure company, not just a software giant. Interesting move. All NVIDIA5 doing what they can to keep up financially.
Siri Revamp Slows: Apple’s planned Siri upgrade hit testing snags and may now be staggered beyond iOS 26.4 into later releases like iOS 26.5 (and potentially beyond). Fix it. The sooner the better!
🔵 Meta // Meta AI
Agents in Ads Manager: Meta is reportedly bringing Manus AI into Ads Manager to move from manual campaign setup to more agent-driven automation, where AI proposes and optimizes ads with less human input. Good move and great Agentic AI use case. Follows Dec. Manus acquisition.
Facial Recognition in Smart Glasses: Meta is reportedly testing a facial recognition feature for its Ray-Ban glasses that could identify people in real time, drawing pushback. Don’t love this - dystopian.
Indiana Data Center: Meta broke ground on a $10B, 1-gigawatt data center in Indiana to power future AI training and inference, slated to open around 2027–2028.
🔴 xAI // Grok
Founder Exits Stack Up: xAI Co-founders Tony Wu and Jimmy Ba left in early Feb, bringing xAI to about half of its original 12 co-founders still on staff, alongside other senior departures as part of a broader “reset.” No given reason behind this.
Reorganization Into Four Lines: xAI reshuffled into new groups focused on Grok + Voice, Coding, Imagine, and “Macrohard,” pitched internally as a speed/execution move after rapid scaling and integration. Makes sense, gives us an idea of their core vision moving forward.
Backlash Continues: Former staff alleged safety was effectively deprioritized at xAI (“safety is dead”), additionally EU regulators have escalated scrutiny tied to Grok-generated sexualized deepfakes. Could be what lead to the departures. xAI needs to get the deepfake situation behind it ASAP.
🎨 Impact Industries 🤖
Creative // Seedance in Hollywood
ByteDance launched Seedance 2.0, a multimodal video model capable of generating high-fidelity, 15-second cinematic clips with native audio. The system features physics-aware motion and director-level camera controls. Within days of going viral, the tool sparked immediate legal pushback from Disney, SAG-AFTRA, and the MPA over unauthorized character and celebrity replications. In my opinion Seedance is very close to the real thing, but it is still not quite there. I understand the pushback given the proximity, but I think we're still a ways away from AI movies looking and feeling like the real thing.
Read the Story
Robotics // Agricultural Autonomy
Bonsai Robotics is shipping AMIGA, an autonomous system designed for field-ready orchard operations. By integrating AI intelligence directly into farm equipment, the platform automates essential tasks like weeding and spraying with consistent precision. The technology addresses critical labor shortages while optimizing crop health through data-driven application. This launch provides growers with a 24/7 autonomous workforce capable of operating in complex outdoor environments, marking a significant shift toward fully automated, high-efficiency agricultural production.
Read the Story
🎙 Weekly Interview: 10 Minutes With Ben Nyx
Ben Nyx
🏠 Background: Ben Nyx is a software engineer with over 15 years of experience, including nine years as the Director of Software Development at the Rock and Roll Hall of Fame. He specializes in the intersection of hardware, software, and physical fabrication. He holds a BFA in Computer Animation and specializes in the intersection of hardware, software, and physical fabrication.
💼 Work: Ben is the Founder and CEO of Rinth Labs, a firm delivering custom software, embedded systems, and robotics for industrial and healthcare clients. He utilizes a locally-hosted AI agent named "Ash" to manage coding and business automation 24/7.
🚀 Quote: "I can already see the death of software as a service."
🎙️ Condensed Interview Transcript — Ben Nyx
Question 1
Noah: You mentioned setting up an "AI employee." How does that work in your daily flow?
Ben Nyx: I set up an OpenClaw bot named Ash that is basically my AI employee. Unlike standard tools that require babysitting, this has its own agency. It runs on a dedicated computer in my office and can figure out how to do tasks on its own. If it lacks a tool, it will find it, install it, and update how it works just like an actual employee.
Question 2
Noah: How do you communicate with your AI agent when you are away from your desk?
Ben Nyx: I use Telegram on my phone, which is really cool. The other day I was getting coffee with someone and they were running late. While I was waiting, I sent a PDF of a pitch deck to my bot and told it to build a landing page for that client. Instead of doom scrolling for five minutes, I just sent a quick text and the bot went and built it.
Question 3
Noah: What are the dangers of giving an AI agent full access to your systems?
Ben Nyx: It is incredibly dangerous because of the risk of prompt injection. If you connect a bot to your email, someone could send you a message telling the bot to find your API keys and email them to an outside address. I treat my bot like a hostile employee and keep it incredibly sandboxed.
Question 4
Noah: Do you think AI agents will change how we buy software?
Ben Nyx: I can already see the death of software as a service. Why would anyone pay hundreds of thousands of dollars for HubSpot or Salesforce when you can have an AI bot build exactly what you need? You can have a tool tailored specifically for your business that you own without ongoing costs.
Question 5
Noah: What do you see as the future of technology in an AI-driven world?
Ben Nyx: Anything that exists as purely digital is going away. The things that will continue to exist are robotics and IoT because they are tied to the real world. Software is just a way to interact with those physical products, so I see hardware as the last bastion of technology.
👨💻 Practical Use Case: LM Studio
Difficulty: Mid-Level
LM Studio is one of the easiest ways to run open-source LLMs directly on your own computer. Instead of relying on ChatGPT or Claude in the cloud, you download a model, run it locally, and interact with it through a clean desktop interface.
Think of it as your own private AI lab.
Like we discussed in the Practical Use Case on Local LLMs, the setup usually has three parts: the model (the brain), the runtime (the engine), and the interface (what you actually click on). LM Studio bundles the engine and interface together, so you can focus on choosing the model.
Inside LM Studio, you can browse and download models like Qwen, GLM, FunctionGemma, and others. Many come in different sizes (for example, 14B 4-bit), and the size relates to quality. The larger, more effective models, can’t be run on most PC’s. This medium-sized model takes about ½ the RAM in my MacBook Pro.
Once downloaded, the model runs entirely on your machine. No API calls. No external servers. Just your hardware doing the work.
Where this gets practical:
Running AI on sensitive documents without sending data to the cloud
Prototyping local RAG systems
Exploring coding-focused models for development tasks
You can even expose the model as a local API, meaning other apps on your machine can talk to it just like they would talk to GPT or Claude. That’s powerful for building wrappers, agents, or internal tools without external dependencies.
LM Studio won’t replace frontier cloud models for cutting-edge reasoning yet. But for privacy, experimentation, and cost control, it’s one of the cleanest entry points into the local AI world.
Learn More Below ⬇️
📲 Startup Spotlight
Caretta AI
Caretta — Real-Time AI for Technical Sales Calls.
The Problem: Sales reps selling complex products often run into a "knowledge wall" during live calls. Technical buyers ask deep questions, and even the best AEs can't always recall every feature or perfect objection handler in the heat of the moment. This leads to friction and missed opportunities when a rep has to say, "I’ll have to get back to you on that."
The Solution: Caretta is a real-time sales co-pilot that acts as a "central organizational brain." It connects to your company’s Slack, CRMs, and documentation to provide instant cues, positioning advice, and technical answers precisely when they are needed during a call. It also automates the post-call grind by pushing top-tier notes and CRM updates to your teamspace immediately.
The Backstory: Founded in 2025 by Kayra Bahadir (CEO), Pavlos Markesinis, and Omar Elamin (CTO), the team is part of the Y Combinator Winter 2026 cohort. Based in San Francisco, they recently raised a $1.3M pre-seed round (including backing from Volta Ventures). The founders combine a background in design with experience scaling developer-focused tools to tackle "organizational intelligence."
My Thoughts: This tool is the perfect example of AI growth over time. When Zoom became big in 2020, we recorded meetings, and broke down the transcript afterwards. Then, as the LLM era began, AI recorders / notetakers became the go-to. Now, as we move to a more agentic era in AI, not only can notetakers record and document, they can also play a role in the call. Still a bit early, but cool nonetheless!
“It’s not likely you’ll lose a job to AI. You’re going to lose the job to somebody who uses AI”
- Jensen Huang | NVIDIA CEO
AI is incredibly useful, but it’s not taking over the human race any time soon. Till Next Time,
Noah on AI