The week brought several major announcements showing that agentic AI – artificial intelligence systems that can make decisions and complete tasks independently – is moving from theory into real-world use. The biggest news came when the Linux Foundation announced the creation of the Agentic AI Foundation on December 9th, bringing together three important technologies developed by major companies.

The foundation's founding projects represent key building blocks for AI agents. Anthropic contributed the Model Context Protocol (MCP), which is now the standard way to connect AI models to tools and data. Think of MCP like a translator that helps AI understand how to use all the different programs and information sources a company has. Block (the company behind Square and Cash App) contributed goose, which is an open-source framework that helps developers build AI agents safely on their own computers. OpenAI shared AGENTS.md, a simple format that tells AI agents exactly what they need to know about different software projects so they can work reliably. Together, these tools help create an open ecosystem where different companies can build AI agents without worrying about one company controlling everything.

Technical breakthroughs are making AI agents smarter and easier to create. Microsoft Research released Agent Lightning on December 11th, which is a new method for teaching AI agents to improve themselves. Normally, updating how an AI agent learns requires programmers to rewrite lots of code. Agent Lightning changes this by breaking down each task into smaller steps and giving each step a score for how well it helped reach the goal. This approach makes it much easier for developers to customize how their agents learn and to use different training methods without starting from scratch.

Multiple AI agents working together are proving especially powerful. IBM Research's Project ALICE uses several AI agents that cooperate to find computer bugs faster and fix broken systems. The idea is that different specialized agents can look at problems from different angles, similar to how a team of people with different skills can solve a problem better than one person alone. In another example, researchers found that AI agents actually debate each other to reach better answers on math problems. When agents argue about different solutions and explain their reasoning to each other, they catch more mistakes and arrive at more accurate answers.

Real companies are already putting AI agents to work in various industries. Construction companies are using AI agents to automatically plan projects, watch progress, and update schedules – tasks that used to take humans many hours. Microsoft announced new AI tools called Microsoft Foundry and Azure Copilot that include specialized agents for moving old computer systems to the cloud, finding performance problems, and handling IT operations. Deloitte found that 30% of large organizations are testing AI agents, and 38% are running pilot projects with the technology.

Developers are gaining easier ways to build AI agents without being programming experts. Platforms like Make.com, Zapier, n8n, and Cursor let people create custom AI agents just by describing what they want. Some tools like Lovable AI even generate and test computer code automatically based on written instructions, making it possible for people without coding experience to create applications.

However, experts warn that challenges remain before AI agents can be trusted with important decisions. Fujitsu researchers identified three critical gaps that need solving: Collaboration (helping multiple agents work together smoothly), Memory (allowing agents to remember important information), and Quality (making sure agents give reliable answers). Companies also worry about data security, making sure agents truly understand the situation before making decisions, and deciding what level of independence to give agents. Organizations are building what experts call graduated autonomy – starting with agents that need human approval for every action, then slowly giving them more independence as they prove reliable.

Looking forward, the predictions are bold. Gartner projects that by 2028, AI agents will make 15% of daily work decisions on their own (compared to none today), and one-third of enterprise software will include agentic AI (compared to less than 1% today). The Linux Foundation expects the new Agentic AI Foundation to become the neutral home where developers worldwide can confidently build on shared standards, knowing the technology will stay open and not controlled by any single company. As these systems improve, AI agents are reshaping how work gets done across industries, from technology to construction to finance.