👾 AI Nuggets #19: Agents Evolve, Models Shrink, Prices Tumble

Welcome to edition #19 of the AI Nuggets. In this issue:

• OpenAI releases new tools for building agents

• AI agent systems could be standardized

• Baidu releases super cheap AI models

• Google releases small AI models running on laptops and phones

• US and Chinese AI stock news

• Edit images with Gemini Flash 2.0 via text prompts

• 4 new AI tools added to tool collection

Latest News

👨🏼‍💻 OpenAI Releases New Agents SDK with Enhanced Tools

OpenAI has released a new Agents SDK that lets developers build AI agents with advanced capabilities. The SDK includes tools for web searches, file retrieval, computer automation, and function calling. It also allows agents to work together in teams.

What's New: The OpenAI Agents SDK comes with three types of tools. Hosted tools run on OpenAI's servers and let agents search the web, find information in files, and use computers. Function tools turn any Python function into a tool for agents. The third type lets agents call other agents. You can now build a main agent that works with specialist agents for different tasks.

Business Impact: You can connect your business data and processes to AI agents that work together. A customer service agent could search your product database, calculate shipping costs, and translate messages - all in one conversation. The SDK's automatic tool setup cuts development time from weeks to days. This means faster deployment and lower costs.

Our Take: OpenAI also stepping into the AI agent space shows once again the relevance. The Agents SDK might become a stable framework for building reliable agents. We’re testing it right now.

🤖 AGNTCY Launches to Standardize Multi-Agent AI Systems

AGNTCY, a new open collective formed by Galileo, LangChain, and Cisco, aims to solve the standardization challenges in multi-agent AI systems. The initiative addresses critical infrastructure gaps in how AI agents collaborate, communicate, and operate at scale. Their approach focuses on creating open standards and protocols that allow developers to build interconnected AI systems across different frameworks and vendors.

What's New: AGNTCY has released its first set of specifications targeting two critical needs in multi-agent systems. The Open Agent Schema Framework (OASF) provides a standard format for describing agent capabilities, making it easier for developers to discover and evaluate agents for specific tasks. The Agent Connect Protocol (ACP) enables standardized communication between agents built on different frameworks, handling complex challenges like message passing and context sharing. These tools form the foundation of what AGNTCY calls the "Internet of Agents," where AI systems can seamlessly collaborate regardless of who built them or what framework they use. The initial specifications are already available on GitHub, with code implementations coming soon.

Business Impact: AGNTCY's work promises to dramatically lower the barriers to advanced AI automation. Today, building complex AI workflows requires specialized expertise and often results in siloed systems that don't communicate well. The standardization AGNTCY offers will allow companies to mix and match specialized agents from different vendors or open-source projects without rebuilding integration layers.

🅱️ Baidu Launches Ernie 4.5 and X1 AI Models

Baidu has released two new AI models: Ernie 4.5, its latest native multimodal foundation model, and Ernie X1, a specialized reasoning model. Both models are now available free to individual users through the Ernie Bot website.

What's New: Ernie 4.5 brings major improvements to Baidu's AI platform. It can process text, images, audio, and video together naturally. The model shows better skills in understanding context, preventing hallucinations, and logical reasoning. Ernie X1 focuses on deep reasoning tasks. It excels at planning, reflection, and using multiple tools. These tools include advanced search, document analysis, image understanding, and code interpretation. Baidu claims Ernie X1 delivers performance comparable to DeepSeek R1 at half the price. Both models are available to individual users for free, while businesses can access them through Baidu's cloud platform at prices starting as low as RMB 0.004 per thousand tokens for Ernie 4.5 and RMB 0.002 for Ernie X1.

Business Impact: The pricing structure makes high-performance AI accessible to organizations with limited budgets. Businesses can now implement sophisticated AI capabilities for a fraction of what they would pay for Western alternatives. Baidu's integration of these models across its ecosystem will also expand their reach. Companies operating in Chinese markets gain access to AI that excels at understanding Chinese language and cultural context. This release intensifies competition in the global AI market, potentially driving down prices while pushing innovation forward.

Our Take: The AI model business is a very competitive niche. OpenAI is not even profitable while the Chinese competition can offer similar models for a fraction of the price.

🥉 Google Releases Gemma 3 Models

Google DeepMind has introduced Gemma 3, a new collection of open AI models built from the same technology that powers Gemini 2.0. These lightweight models run on single GPUs or TPUs while delivering performance that rivals much larger models. Gemma 3 comes in four sizes (1B, 4B, 12B, and 27B) and includes advanced capabilities for text, image, and video processing across 140 languages.

What's New: The models outperform competitors like Llama3-405B and DeepSeek-V3 in human preference tests while requiring far less computing hardware. The 27B version can run on a single GPU yet delivers performance close to models needing 32 GPUs. Gemma 3 handles text and images together and works with short videos. It processes large amounts of information with a 128K token context window. The models support function calling for building AI agents and automated workflows. Google also released ShieldGemma 2, a 4B image safety checker that screens for dangerous, explicit, or violent content. Developers can use Gemma 3 with popular tools like Hugging Face, PyTorch, and Ollama. The models come with optimized versions for different hardware including NVIDIA GPUs, Google Cloud TPUs, AMD GPUs, and even CPUs.

Business Impact: Businesses can now run advanced AI on standard workstations instead of expensive cloud services. The models work in 140 languages, opening global markets for startups with minimal translation costs. The inclusion of ShieldGemma 2 helps companies deploy AI safely without building custom safety systems. Businesses can start small with the 1B model and scale up as needs grow, all using the same framework.

Our Take: AI models are getting more efficient. Hardware is getting more powerful. Soon it’ll become mainstream to run advanced AI models on laptops and phones.

AI Investments - What to Watch¹

🇺🇸 Nasdaq AI Stock News

Tesla: (NASDAQ: TSLA) plans to launch its robotaxi service in Texas this June, with Goldman Sachs highlighting potential for attractive profits in the autonomous vehicle sector.

Nvidia: (NASDAQ: NVDA): demonstrates strong financial performance with 77.94% revenue growth, significantly outpacing the industry average of 14.97%, while maintaining a healthy debt-to-equity ratio of 0.13.

Apple: (NASDAQ: AAPL) faces significant pressure over AI development delays, particularly regarding Siri improvements. Warren Buffet’s Berkshire Hathaway has dramatically reduced its Apple position by approximately 70%.

Meta: (NASDAQ: META) demonstrates superior financial performance versus industry peers, with strong Return on Equity, EBITDA, and revenue growth, supported by a notably low debt-to-equity ratio of 0.27.

Amazon: (NASDAQ: AMZN) trading at PE ratio of 35.8x (below industry average), with strong EBITDA of $38.55B and gross profit of $37.37B, though revenue growth of 10.49% lags industry average of 27.94%.

Microsoft: (NASDAQ: MSFT) maintains a robust financial position with a notably low debt-to-equity ratio of 0.21, indicating strong balance sheet health compared to top industry peers.

Alphabet: (NASDAQ: GOOG) Waymo maintains clear leadership in the autonomous vehicle market with over 2 million paid robotaxi rides across multiple U.S. cities, significantly ahead of competitors in commercialization

🇨🇳 Hang Seng AI Stock News

Baidu: (NASDAQ: BIDU) shares jumped over 8% Monday following the company's launch of two advanced AI models (ERNIE 4.5 and X1). The stock's positive performance reflects investor confidence in Baidu's AI strategy amid intensifying competition in China's tech sector.

AI Use Cases and Tools

🪄 Edit Images with Text Prompts

Google published a new experimental feature for their Gemini 2.0 Flash model. Here’s how to test it:

1. Log in to Google AI Studio

2. Select the Gemini 2.0 Flash (Image Generation) Experimental model

3. Upload an image or let AI generate it

4. Prompt your changes and see how AI modifies your image

MadebyAgents Updates

👾 Curated Tool Collection with Tested AI Tools

We rebuilt the backend for our tool collection, enabling advanced filtering and room for new features. Instead of scraping every AI tool out there and loading it into our database, we decided to add only tools we’ve tested. This week we added:

✂️ Capcut: Transform text and images into professional videos instantly with this AI-powered editor.
🎥 Opus Clip: Extract and repurpose the most engaging moments from long-form content into short clips for multiple platforms.
🪿 Codename Goose: Turn your ideas into functional software with this locally running AI agent that connects to MCP servers
👄 LipDub AI: Synchronize any audio-to-video footage with this AI lip-syncing tool that creates realistic mouth movements for dubbing in any language.

That’s It for This Week

✨ Before You Go:

Share The AI Nuggets: This newsletter is 100% free. To support our work, consider sharing The AI Nuggets with your friends and colleagues.

We’d love to hear what you think. Please share your opinion.

See you next time!

Tobias from MadebyAgents