AI agents are officially out of the lab.

This week, the world of AI felt like it hit the gas pedal—and then some. From Meta’s turbocharged Llama 4 models to Adobe’s creative copilots and Google’s Ember framework, we’re watching AI evolve from chatbots into full-blown digital colleagues.

Meta’s Llama 4: The Controversy

Meta dropped not one, but two new models: Llama 4 Scout and Maverick. Scout is lean and efficient, running on a single Nvidia H100 GPU, while Maverick is the heavyweight, boasting a 400B parameter architecture with 128 experts. Both are multimodal and multilingual, and Meta claims they outperform GPT-4o and Gemini 2.0 Flash on benchmarks. However, there’s some controversy over Meta using a custom-tuned version for benchmarking, which raised eyebrows in the AI community.

Google’s Ember: Fixing AI’s Overthinking Problem

Nvidia, Google, and Foundry introduced Ember, an open-source framework designed to tackle AI models’ tendency to “overthink.” Ember orchestrates multiple models with different strengths to handle complex prompts, aiming to improve accuracy and efficiency. This approach could redefine how AI systems process tasks, moving away from relying on a single model to a more collaborative model ecosystem.

Adobe’s AI Agents: Creative Assistants in the Making

Adobe is enhancing its suite of creative apps with advanced AI agents. These agents aim to streamline complex design processes, making tools like Photoshop and Adobe Express more accessible to users without in-depth knowledge. A more advanced Photoshop agent with Action panel functionalities is expected to be revealed at Adobe Max in London.

Google Cloud’s Agentic AI Push

At its 2025 Cloud Next conference, Google announced significant advancements, including the Ironwood TPU, which offers over ten times the performance of its predecessor. Google is also integrating autonomous, goal-driven AI agents into its Vertex AI platform, enabling tasks beyond conversational capabilities. These developments are part of Google’s broader strategy to support agentic AI in enterprise environments.

UiPath’s Transition to Agentic AI

UiPath is shifting from traditional robotic process automation to agentic AI. CEO Daniel Dines emphasizes integrating deterministic software automation with non-deterministic AI capabilities like large language models. This approach aims to enhance enterprise workflows by combining AI agents, human workers, and traditional automation.

Salesforce’s APIGen-MT: Advancing Multi-Turn AI Interactions

Salesforce introduced APIGen-MT, a framework for generating high-quality, multi-turn agent data through simulated human-agent interactions. Their xLAM-2-fc-r models, ranging from 1B to 70B parameters, reportedly outperform GPT-4o and Claude 3.5 in multi-turn settings. These models and datasets are open-sourced, contributing to the development of more reliable and capable AI agents.

Deloitte’s Zora AI: Enterprise-Ready Agentic Platform

Deloitte unveiled Zora AI, a new agentic AI platform offering a portfolio of digital agents for various enterprise functions, including finance, human capital, supply chain, procurement, sales, marketing, and customer service. This platform aims to enhance enterprise operations through intelligent automation.

Google’s A2A Protocol: Teaching AI Agents to Talk

Google introduced the Agent2Agent (A2A) protocol, an open standard designed to enable seamless communication and collaboration between AI agents across various enterprise platforms and applications. A2A allows AI agents to communicate with each other, securely exchange information, and coordinate actions on top of various enterprise platforms or applications. This initiative aims to break down silos between different AI systems, allowing them to work together more effectively.

Google’s Agent Development Kit (ADK): Building Smarter AI Agents

Google unveiled the Agent Development Kit (ADK), an open-source Python toolkit designed to simplify the creation of sophisticated AI agents. ADK provides developers with the tools to build, evaluate, and deploy agents with greater flexibility and control. It supports integration with Google’s Gemini models and offers features like modular agent design, built-in evaluation frameworks, and seamless deployment options through platforms like Vertex AI. This initiative aims to empower developers to create more reliable and capable AI agents efficiently.

The Bottom Line

Agentic AI is no longer a futuristic concept; it’s becoming an integral part of various industries. From creative tools to enterprise solutions, AI agents are poised to transform how we work and interact with technology. Let’s see what next week brings.