Payment Infrastructure for AI Agents

Today’s top AI Highlights:

1. The first financial infrastructure for AI agents to transact real money

2. Opensource framework to run AI agents in Kubernetes

3. NVIDIA brings GB300 Superchip to your desktop

4. AI voice model that can clone voices with just 3 seconds of audio

5. Build and deploy MCP servers without writing a single line of code

& so much more!

Read time: 3 mins

AI Tutorials

In this tutorial, we'll show you how to create your own powerful Deep Research Agent that performs in minutes what might take human researchers hours or even days—all without the hefty subscription fees. Using OpenAI's Agents SDK and Firecrawl, you'll build a multi-agent system that searches the web, extracts content, and synthesizes comprehensive reports through a clean Streamlit interface.

OpenAI's Agents SDK is a lightweight framework for building AI applications with specialized agents that work together. It provides primitives like agents, handoffs, and guardrails that make it easy to coordinate tasks between multiple AI assistants.

Firecrawl’s new deep-research endpoint enables our agent to autonomously explore the web, gather relevant information, and synthesize findings into comprehensive insights.

We share hands-on tutorials like this 2-3 times a week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.

Don’t forget to share this newsletter on your social channels and tag Unwind AI (X, LinkedIn, Threads, Facebook) to support us!

Latest Developments

Build AI Agents that Can Autonomously Handle Money 🦾 💰

AI agents can now handle real-world money, securely. A new startup, Payman AI, has launched a financial infrastructure platform built specifically for AI agents to transact with real-world money (USD and USDC). This platform addresses the growing need for secure and controlled financial capabilities as AI agents take on more complex tasks.

Payman AI allows you to give your AI agents financial autonomy, but with built-in safeguards and human oversight to prevent errors or misuse. The system features no-contract wallets and robust APIs, creating an efficient path to integrating payment capabilities directly into AI workflows.

Key Highlights:

1. Dedicated AI Agent Wallets & Accounts - Payman AI provides separate financial accounts specifically for AI agents, keeping them isolated from your primary business accounts. This allows for precise control over spending and provides clear transaction tracking for each agent.

2. Payee Protection and Human-in-the-Loop - You can pre-approve payees, preventing unauthorized transactions by AI agents. Customizable approval workflows ensure human oversight for all or specific transactions, adding a critical layer of security.

3. Secure Fund Orchestration - Payman AI manages funding for AI agent accounts predictively, ensuring they have the necessary capital without ever directly accessing your main funding source. This separation minimizes risk and maintains financial security.

4. SDKs and API Integration - NodeJS and Python SDKs are available for immediate integration, allowing you to quickly implement payment functionality into their AI agent applications. The API is designed specifically for AI payment operations, with built-in safeguards.

Kubernetes-Native Framework for Building AI Agents ☁️🤖

Kagent is a new open-source framework that brings AI agents directly into Kubernetes environments, enabling DevOps and platform engineers to automate complex operations without leaving their clusters. Kubernetes is the most popular orchestration platform for running workloads, and kagent makes it easy to build, deploy, and manage AI agents in Kubernetes.

These agents can reason through multi-step problems, troubleshoot issues across service boundaries, and transform insights into concrete actions within your cloud-native infrastructure. The framework uses a flexible architecture built on Microsoft's AutoGen to create autonomous systems that can tackle everything from connectivity issues to progressive rollouts.

Key Highlights:

1. Tool-rich Environment - Kagent comes with dozens of pre-built tools for Kubernetes, Prometheus, Istio, Helm, Argo, and Grafana, allowing agents to interact with your entire cloud-native stack. You can extend capabilities further through Model Context Protocol (MCP) support or by annotating OpenAPI-compliant services for automatic tool discovery.

2. Autonomous Problem-solving - Agents can chain multiple operations together to handle complex tasks like diagnosing connectivity issues across service hops, debugging gateway configurations, or automating alert generation from metrics. Each agent can work independently or as part of a team where a planner assigns subtasks to specialized members.

3. AutoGen-powered Engine - Kagent's core engine is built on Microsoft's AutoGen framework, taking advantage of its flexible and extensible architecture. It enables powerful conversation loops while supporting custom Teams, Agents, and Tools tailored for cloud-native environments. It is a solid foundation that you can easily extend for your specific operational needs.

4. Multiple Interfaces - The framework offers both a dashboard UI and CLI for interacting with agents, making it accessible for different workflows and preferences. The architecture includes a Kubernetes controller for handling custom CRDs, along with a Python-based engine that powers the agent's conversation loops.

Quick Bites

Cartesia has released Sonic-2, a groundbreaking AI voice model that can clone voices with just 3 seconds of audio input. It offers precise emotion control and allows you to create ultra-realistic voices with adjustable speaking styles, emotions, and speeds. Sonic-2 supports 15 languages, including English, Mandarin, Japanese, and Hindi, and offers significantly lower latency than previous models.

LG AI Research has released EXAONE Deep, a series of reasoning LLMs (32B, 7.8B, and 2.4B) that delivers impressive benchmark results despite its relatively modest size. The 32B parameter model matches DeepSeek-R1's performance (671B parameters) on mathematical reasoning tasks while being just 5% of its size, while the smaller 7.8B model outperforms many larger competitors across math, science, and coding benchmarks. All the models are available on Hugging Face.

Anthropic is developing a voice mode for Claude, to give a more natural user interface. It is confirmed that they are internally developing and prototyping voice capabilities.

The much-anticipated keynote by Jensen Huang wrapped up at the NVIDIA GTC 2025 yesterday. Though it lacked the usual revolutionary punch, NVIDIA gave a sneak (only) at its upcoming hardware designed for AI workloads. Blackwell Ultra GB300 is an enhanced iteration of the Blackwell architecture arriving in the second half of 2025, alongside a preview of the next-generation Vera Rubin architecture slated for 2026.

• Blackwell Ultra: While Ultra retains Blackwell's core performance, it provides a significant memory boost of 50% over the original Blackwell. One rack of Blackwell Ultra with NVL72 delivers 1.1 exaflops of computing.

• Vera Rubin: Just like the Grace Blackwell architecture, Vera Rubin superchip combines 1 Vera CPU and 2 GPUs, with each GPU consisting of 2 Rubin chips. Aims for 50 petaflops of compute.

• Rubin Ultra and NVL576: The Rubin Ultra, arriving in the second half of 2027, will feature a dual-GPU design. One rack with NVL576 targets 15 exaflops of FP4 inference performance.

But here’s a big one: NVIDIA unveiled the new DGX Station powered by the GB300 Grace Blackwell Ultra Desktop Superchip. It delivers up to 20 petaFLOPS of AI performance with a massive 784 GB of unified memory. It can handle AI models with over 200B parameters for inference and fine-tune models up to 70B parameters (that’s insane!!)

Tools of the Trade

1. Firecrawl MCP Server: Connects MCP clients like Claude and Cursor to Firecrawl to enable them to do web scraping, crawling, and content extraction with features like JavaScript rendering, batch processing, and structured data extraction.

2. MCPify AI: Build and deploy MCP servers without writing a single line of code. Just tell it what you want in simple language > it creates an MCP server with various tools and gives your own MCP Server URL. You can prompt it to add/remove/modify these tools. Works with all MCP clients.

3. Agentic Radar: Open-source security scanner for AI agents to identify potential vulnerabilities by generating comprehensive security reports with dependency graphs and OWASP framework integration. Supports multiple frameworks like LangGraph and CrewAI.

4. Awesome LLM Apps: Build awesome LLM apps with RAG, AI agents, and more to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos, and automate complex work.

Hot Takes

1. The simplest way to bring back manufacturing to the US
   Win The Robotics Race!
   The government should drop $100B on robotics and AI research
   Lets accelerate the breakthroughs 🚀🚀 ~
   Bindu Reddy
   
   

2. AI-free experiences might soon become something we start advertising and selling.
   A "handmade" of sorts. It might even be a huge selling point for many. ~
   Santiago

That’s all for today! See you tomorrow with more such AI-filled content.

Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!

Unwind AI - X | LinkedIn | Threads | Facebook

Awesome LLM Apps | Sponsor Us

PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉