To Build or Not? Multi-Agent AI Systems

Today’s top AI Highlights:

1. This Research Agent costs $0.10 while OpenAI charges $200/month

2. Anthropic vs Cognition: To build or not to build multi-agent systems

3. Meta is investing $14.3B in Scale AI to build a Superintelligence lab

4. Google Cloud crashed 3 days back, so devs built an AI outage bot

5. Opensource alternative to Clay by Firecrawl

& so much more!

Read time: 3 mins

AI Tutorial

Building good research tools is hard. When you're trying to create something that can actually find useful information and deliver it in a meaningful way, you're usually stuck cobbling together different search APIs, prompt engineering for hours, and then figuring out how to get the results into a shareable format. It's a headache, and the results are often inconsistent.

In this tutorial, we'll build an AI Domain Deep Research Agent that does all the heavy lifting for you. This app uses three specialized agents that are built using the Agno, Qwen 3 235B model via Together AI, and use tools via Composio to generate targeted questions, search across multiple platforms, and compile professional reports.

What makes this deep research app different from other tools out there is its unique approach: it automatically breaks down topics into specific yes/no research questions, combines results from both Tavily and Perplexity AI for better coverage, and formats everything into a McKinsey-style report that's automatically saved to Google Docs.

We share hands-on tutorials like this every 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) to support us!

Latest Developments

Opensource, Fast, and Cheap Deep Research Stack 🔓⏳💸

Big Tech just flooded the market with their "Deep Research" features - and they're all expensive, locked-down, and limited to their specific ecosystems. OpenAI’s version costs $200/month, Google's requires Gemini Advanced subscriptions, and even Perplexity limits free users to just 5 queries per day.

GPT Researcher offers the antidote to this vendor lock-in by being a fully opensource research agent that you can actually own, customize, and deploy however you want.

It deploys multiple AI agents that work in parallel to gather information from 20+ sources, then synthesizes everything into comprehensive, cited reports. It even completes these thorough investigations in just 3 minutes for around $0.1 per report.

Key Highlights:

1. Multi-Agent - Uses specialized "planner" and "execution" agents that work together to generate research questions, scrape relevant sources, and synthesize findings into objective conclusions.

2. Deep Research - The new recursive research feature explores topics with tree-like depth and breadth, taking about 5 minutes and costing $0.4 to deliver unprecedented comprehensive analysis.

3. Scalable Performance - Completes standard research tasks in 3 minutes for $0.1, while the new "Deep Research" mode delivers comprehensive analysis in 5 minutes for $0.4.

4. Source Coverage - Aggregates information from 20+ sources and breaks through LLM token limitations to generate reports exceeding 2,000 words with proper citations.

5. Integrations - Works with any LLM provider, search engine, or document source, plus includes MCP server integration and multi-format export options.

News you’re not getting—until now.

Join 4M+ professionals who start their day with Morning Brew—the free newsletter that makes business news quick, clear, and actually enjoyable.

Each morning, it breaks down the biggest stories in business, tech, and finance with a touch of wit to keep things smart and interesting.

Check it out

Multi-Agents Are Both Amazing and Terrible 🔥💀

Two AI giants just served up the perfect example of why building production AI agents is still an art, not a science.

Cognition released "Don't Build Multi-Agents" arguing that parallel agent systems are too fragile for real-world use, while Anthropic simultaneously published their deep dive into building the exact multi-agent system that powers Claude's new Research capabilities.

The timing couldn't be more perfect or more revealing about how different teams solve the same fundamental problems. Both teams have served us with what works and what doesn't. And developers building any kind of AI agent system need to study both to understand the real trade-offs.

Here's a broad comparison of how both teams tackle the core engineering decisions that make or break production agent systems:

Plug and Play RAG Stack That’s Ready for Production 🔌📦🚀

RAG is everywhere, but most stacks are either bloated or messy - half-baked scripts, duct-taped SDKs, and zero structure.

Ragbits is a modular type-safe opensource Python package that gives the essential “bits” for building RAG apps: fundamental tools for working with LLMs and vector databases, abstractions for building agentic systems, full-stack infrastructure for conversational apps, and more.

It lets you swap between 100+ LLMs, and connect to any vector store. The framework includes everything from document ingestion pipelines that handle 20+ file formats to a complete chat UI with persistence and user feedback.

Key Highlights:

1. Plug and play components - Switch between 100+ LLMs via LiteLLM, connect to any vector store (Qdrant, pgvector, Chroma), and use your preferred document parser (Unstructured or Docling) with consistent APIs that just work.

2. Type-safe - Python generics enforce strict type safety in model calls, catching errors at development time rather than in production when your RAG pipeline decides to return unexpected data structures.

3. Production-ready - Built-in observability with OpenTelemetry, CLI tools for managing vector stores and testing prompts, plus Ray-based parallel processing for scaling document ingestion without performance headaches.

4. Quick prototyping - Start with uvx create-ragbits-app, pick your stack components, and get a working RAG system with a React UI in minutes, then scale seamlessly as your requirements grow.

Quick Bites

After Llama 4 fizzled and “Behemoth” went missing, Meta is turning to Scale AI and paying $14.3B for it. They’re buying speed, talent, and a working data pipeline by bringing in Alexandr Wang and his team. Wang will now head a fresh Superintelligence division at Meta focusing and agentic and autonomous capabilties, with a clear goal: catch up or get left behind.

The timing is telling - while Google, OpenAI, and Anthropic ship new capabilities weekly, Meta's been stuck iterating on models that can't quite handle the agentic AI everyone's racing toward. The real question is when you already have FAIR and GenAI teams, will yet another AI team fix what’s broken?

Google has released some major upgrades to its AI coding agent, Gemini Code Assist. The agent is now powered by Google’s best Gemini 2.5 Pro model for better code generation, reasoning across chats, and PR reviews. The personalization features steal the show though - you can now set up custom commands for scaffolding and establish project-wide rules that guide the AI's output to match your conventions. It’s free to try via VS Code or JetBrains.

Tools of the Trade

1. Fire Enrich: Opensource alternative to Clay. Turn a simple list of emails into a rich dataset. Upload a CSV and AI agents automatically fill in missing data like decision makers, company size, and more. Free to try.

2. Down Detector Bot: 3 days ago, Google Cloud went down and dragged half the internet with it - OpenAI, Anthropic, Spotify, even smart homes froze. This CLI bot checks AWS, GCP, Cloudflare, and Azure every hour, uses GPT to spot real outages, and pings your Slack only when something actually breaks.

3. MEOW: Opensource image format that bakes AI-relevant metadata directly into the image, like edge maps, attention zones, and object relationships. It works like a PNG but adds extra info that helps models understand images faster and better.

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

Hot Takes

1. mcp is really useful but there is a 0% chance the average user goes through the headache of setting up custom ones ~
   Sully Omarr
   
   

2. it's bizarre that being rude to chatgpt, claude etc somehow unlocks AI god mode

   “are you kidding me? this is AWFUL. make it way better or else” ~
   GREG ISENBERG

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

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 😉