AI Startup Stories: How To Build The Future: Aravind Srinivas(Perplexity CEO)

AI Resources: What is Model Context Protocol

MCP (Model Context Protocol) is an open standard designed to simplify how AI assistants and Large Language Models (LLMs) connect to and interact with various data sources and tools. Acting as a universal interface, MCP streamlines AI integration, reducing the need for multiple custom API connections.

Key Benefits of MCP:

• Simplified Integration: One standardized protocol allows AI to access multiple tools with minimal development effort.

• Real-time Communication: Enables persistent two-way interaction, letting AI retrieve and trigger actions dynamically.

• Dynamic Discovery: AI can identify and use new tools without hard-coded integrations.

• Enhanced Context Awareness: Provides real-time, relevant information for better AI responses.

• Scalability & Flexibility: New tools can be added easily, and AI models can be switched without major reconfiguration.

• Improved Security: Built-in security practices ensure consistent access control across tools.

An example is Perplexity's Sonar API, which uses MCP to allow AI models to perform real-time web research.

In summary, MCP offers a unified framework for AI connectivity, enhancing flexibility, automation, and context-awareness, though traditional APIs may still be useful for highly controlled scenarios.

What is Model Context Protocol (MCP)?

Integrating MCP with Perplexity's Sonar API

Building Agents with Model Context Protocol

Introducing the Model Context Protocol

OpenAI New Tools for Building Agents with API

OpenAI has introduced new tools and APIs to simplify the development of AI agents that can independently accomplish tasks. Key releases include:

1. Responses API – A new API that combines Chat Completions' simplicity with the Assistants API’s tool-use capabilities. It includes built-in tools like web search, file search, and computer use, making it easier to build intelligent applications.

2. Built-in Tools:

   • Web Search – Provides up-to-date answers with citations.

   • File Search – Allows retrieval of relevant information from large document sets.

   • Computer Use – Enables agents to complete tasks on computers, though it remains a research preview.

3. Agents SDK – An open-source toolkit to orchestrate multi-agent workflows with features like tracing, safety checks, and automated handoffs.

4. API Changes:

   • The Chat Completions API remains for general use, but new integrations should start with the Responses API.

   • The Assistants API will be phased out by mid-2026, with a migration plan to the Responses API.

OpenAI envisions AI agents becoming core to various industries, enhancing productivity.
New tools for building agents

Gemma 3: The Most Powerful Model for a Single GPU or TPU

Google introduces Gemma 3, a new collection of open AI models designed for speed, efficiency, and accessibility across various devices. With sizes ranging from 1B to 27B parameters, Gemma 3 delivers state-of-the-art performance, outperforming competitors in human evaluations.

Key Features

• Best single-accelerator performance: Runs efficiently on a single GPU or TPU.

• Multilingual capabilities: Supports 140+ languages.

• Advanced reasoning: Handles text, images, and short videos.

• Large context window: Processes 128k tokens for complex tasks.

• Function calling: Enables AI-driven automation.

• Optimized for efficiency: Includes quantized versions for faster performance.

Safety & Responsible AI

Gemma 3 was developed with rigorous safety protocols, including fine-tuning for risk mitigation and enhanced evaluations for STEM-related safety risks.

ShieldGemma 2, a 4B image safety model, is also introduced to detect harmful content (e.g., violence, explicit material).

Integration & Accessibility

• Works with popular tools: Hugging Face, PyTorch, JAX, TensorFlow, Google AI Edge, NVIDIA GPUs, AMD ROCm.

• Deployment options: Google Cloud (Vertex AI, Cloud Run), local setups.

• Optimized for NVIDIA GPUs & Google TPUs.

Expanding the "Gemmaverse"

• Over 100M downloads and 60,000 community-created variants.

• Academic Program: Offers $10K in Google Cloud credits for researchers.

Introducing Gemma 3

Adobe Shares Drop 13% Amid AI Growth Concerns

Adobe's stock fell 13% after its earnings report, as concerns over AI monetization overshadowed better-than-expected financial results. The company posted $5.08 EPS and $5.71B revenue, beating analyst estimates.

Key Points:

• AI Monetization Concerns: Investors worry Adobe is losing ground in generative AI despite its $125M AI revenue, expected to double by fiscal year-end.

• Growth & Forecast: Adobe forecasts $5.77B–$5.82B revenue for Q2, in line with Wall Street expectations. FY 2025 guidance projects 9% growth.

• Market Competition: Analysts say Adobe must prove AI will add new revenue, not replace existing streams.

• AI Strategy: The company is integrating AI into Photoshop and other products to differentiate from competitors.

Despite strong earnings and AI investments, investors seek clearer monetization plans, contributing to the stock decline.