- The Agent Roundup
- Posts
- 👾 This $1M AI Agent Team Analyzes Any Stock in Minutes
👾 This $1M AI Agent Team Analyzes Any Stock in Minutes
Baidu & Huawei open-source AI models. Cloudflare sees major shift in search traffic for creators. Breaking news in AI research, code agents, and AI voice tech.
Welcome to The Agent Roundup
This week’s topics:
New open-sourced AI models from China
AI crawling gets blocked from Cloudflare-managed websites
An algorithm that lets AI models collaborate on complex problems
New best-performing coding agent on the market
Open-source text-to-speech for real-time usage
How to build a portfolio research team with AI agents
Plus AI investments, trending AI tools, community highlights, and more
AI Agent News Roundup
💥 Breakthroughs
Baidu Open-Sourced ERNIE 4.5![]() Source: Baidu ERNIE 4.5 is a new family of large-scale multimodal models consisting of 10 distinct variants. The largest 424B parameter model surpasses DeepSeek V3 on 22/28 benchmarks. Development toolkits like ERNIEKit (for fine-tuning and alignment) and FastDeploy (for efficient deployment) are also open-sourced. | Huawei Open-Sourced Pangu![]() Source: Reve / Agent Roundup Huawei announced the open sourcing of its Pangu models (Pangu 7B & Pangu Pro MoE 72B) and model inference technologies. This initiative is a step in Huawei's Ascend ecosystem strategy, aiming to advance research and innovation in large model technologies and accelerate AI application and value creation across industries. |
AI Crawlers Blocked From 20% of the Web?![]() Source: Cloudflare Cloudflare announced a shift in the internet’s business model due to AI. AI systems replace traditional search queries, leading to a significant decline in traffic for content creators. Cloudflare is changing the default to block AI crawlers unless they compensate creators for their content. New features like "pay per crawl" are being introduced to enable content owners to charge AI crawlers. | New Algorithm for AI Models Hints at Multi-Agent Future![]() Source: OpenAI Sakana AI has developed AB-MCTS (Adaptive Branching Monte Carlo Tree Search), a new inference-time scaling algorithm. This algorithm enables AI to perform trial-and-error effectively and allows multiple frontier AI models to cooperate, leveraging their collective intelligence. It solved problems unsolvable by any single LLM through collaboration. The research suggests this could be the next paradigm after model scaling. |
Beating Claude Code: Warp Agents![]() Source: Warp Warp 2.0 is announced as The Agentic Development Environment, designed to be the fastest way to build with multiple AI agents for coding. A core feature is the ability to run multiple agents in parallel for tasks like building features, debugging errors, and deploying projects, with central management and user intervention when needed. Warp provides agents with the necessary context through access to MCP servers, Warp Drive (for centralized knowledge), and rules. | Kyutai Mindblowing Text-to-Speech![]() Source: Kyutai Kyutai has released Kyutai TTS (kyutai/tts-1.6b-en_fr), a text-to-speech model optimized for real-time usage. A key innovation is its ability to stream in text, meaning it doesn't need the whole text in advance. It has a latency of 220ms from receiving the first text token to generating the first chunk of audio. Kyutai is demonstrating strong performance in Word Error Rate and Speaker Similarity. It can also clone voices from a 10-second audio sample. |
📈 Investments
🇺🇸 Meta has launched Meta Superintelligence Labs (MSL), a new AI division aimed at advancing artificial general intelligence (AGI). MSL represents a major strategic push to lead in AI development by consolidating internal efforts and aggressively hiring top talent from competitors, aiming to accelerate breakthroughs in superintelligence and maintain technological leadership.
🇺🇸 Google pre-ordered 200MW of fusion energy from Massachusetts-based Commonwealth Fusion Systems (CFS). This carbon-free energy is expected to be delivered in the early 2030s.
🇺🇸 Apple is considering a major shift in its AI strategy by potentially replacing its proprietary in-house models with AI technology from Anthropic or OpenAI to power the next generation of Siri.
🇺🇸 Amazon deployed its one-millionth robot and launched DeepFleet, a new generative AI model. This AI technology acts as an intelligent traffic management system for its robotic fleet, aiming to improve travel efficiency by 10% to achieve faster delivery times and lower costs for customers.
🇬🇧 The Premier League has partnered with Microsoft to integrate Copilot into its platforms, enhancing personalized fan interactions.
🇺🇸 Grammarly has acquired Superhuman, an AI-powered email efficiency tool, as part of its strategy to expand beyond grammar correction and build a comprehensive AI-driven productivity suite.
🇺🇸 Thinking Machine Labs is aggressively attracting elite AI talent by offering exceptionally high base salaries, with some technical roles reaching up to $500,000 annually.
🇺🇸 Anthropic has reached a $4B annualized revenue by mid-2025, which is nearly 4x higher than its projections at the start of the year. This rapid growth is driven mainly by enterprise and startup API usage, particularly through its Claude AI models, with code generation being the primary revenue source.
🇸🇪 Lovable is raising over $150M in a new funding round, which would value the Swedish company at nearly $2B. Founded in 2023, Lovable builds entire web applications from text prompts, and its CEO stated it hit $50 million in annual recurring revenue (ARR) within six months.
Unlock the Power of AI With the Complete Marketing Automation Playbook
Discover how to scale smarter with AI-driven workflows that actually work. This playbook delivers:
A detailed audit framework for your current marketing workflows
Step-by-step guidance for choosing the right AI-powered automations
Pro tips for improving personalization without losing the human touch
Built to help you automate the busy work and focus on work that actually makes an impact.
Building a Portfolio Research Multi-Agent AI System
The Problem: Information Overload in Modern Investment Research
Every day, investors are bombarded with thousands of data points, market signals, and conflicting opinions. Traditional investment research faces three critical challenges:
Too much data to track: Economic indicators, news articles, and analyst reports.
Siloed expertise: Fundamental analysts, quantitative researchers, macroeconomists, and portfolio managers are usually not the same person.
Inconsistent quality: Manual research is error-prone, quality varies between researchers, time constraints, and a lack of standardized frameworks.
The Solution: A Multi-Agent AI Research Orchestra
What if you could have a team of AI specialists working 24/7, each expert in their domain, collaborating seamlessly to produce institutional-quality investment research? That's exactly what I've prototyped.
The Portfolio Research Multi-Agent System creates a virtual research team that mirrors the structure of elite hedge funds:
🎯 Head Portfolio Manager Agent
Orchestrates the entire research workflow
Synthesizes insights from specialist agents
Challenges conventional thinking with contrarian perspectives
Ensures alignment with a firm investment philosophy
📊 Quantitative Analysis Agent
Performs statistical analysis and backtesting
Builds predictive models using historical data
Conducts risk analysis and portfolio optimization
Generates charts, correlations, and technical indicators
🏢 Fundamental Analysis Agent
Evaluates company financials and business models
Analyzes competitive positioning and market dynamics
Assesses management quality and strategic direction
Identifies catalysts and value drivers
🌍 Macro Analysis Agent
Monitors economic indicators and policy changes
Analyzes sector rotation and market cycles
Evaluates geopolitical risks and opportunities
Connects macro trends to specific investments
✍️ Report Editor Agent
Compiles analysis into professional investment reports
Ensures consistent formatting and structure
Generates both Markdown and PDF outputs
Maintains institutional-quality presentation standards

Flowchart of the Agent System
Key Principles
Parallel Execution: All specialist agents run simultaneously, not sequentially
Tool Integration: Each agent has access to specialized data sources and analysis tools
Quality Control: The Portfolio Manager reviews and challenges each analysis
Iterative Refinement: Agents can request additional analysis if gaps are identified
Professional Output: Final reports match institutional research standards
The Tech Stack
Core Framework: OpenAI Agents SDK, Python, Asyncio, Pydantic
Data Sources & APIs: Yahoo Finance MCP, FRED API, Web Search, Code Interpreter
Analysis Tools: Pandas/NumPy, Matplotlib/Seaborn, SciPy, CVXPY
User Interfaces: Streamlit, CLI, Markdown/PDF
*Tutorial will be published tomorrow on my YouTube channel.
Tool Spotlight
🗺️ Explorium MCP – Connect Your Agent to B2B Data

Source: Explorium
The Explorium MCP server is a powerful AI data integration platform designed to seamlessly connect live B2B company and contact data with LLMs and AI agents through a simple, unified API.
It enables AI-driven applications, agents, and workflows to access real-time, high-quality firmographic, technographic, and compliance data from over 150M companies and 800M professionals worldwide, empowering smarter account research, prospecting, and personalized outreach.
💻 Cursor launched apps on web and mobile: It allows running tasks in the background, access from anywhere, and enhanced collaboration.
👾 Replit released Dynamic Intelligence: It enhances the agentic coding tool by extending thinking, higher power models, and web search.
More Resources
Blog: In-depth articles on AI workflows and practical strategies for growth
AI Tool Collection: Discover and compare validated AI solutions
Consultancy: Explore AI potential or make your team AI-fit
Agency: AI implementation services to scale your business
See you next time!
Tobias from The Agent Roundup