From autonomous agents to vibe coding, 2025 was the year generative AI stopped being theoretical and started doing real work—with a little fun along the way. Our readers gravitated toward features and tutorials that explored how to move AI into production software and reshape developer workflows, and to columnists who forced uncomfortable (and sometimes amusing) questions about the role of humans in the AI-driven workplace. Here’s a look back at some of InfoWorld’s most popular AI coverage this year.

The year agents took off

2025 may be remembered, among other things, as the year AI agents moved beyond research concepts and toy demos to drive real-world applications and platforms. Agents can now handle everyday software tasks, integrate into developer workflows, and are embedded into large-scale enterprise infrastructure. Some of the year’s most popular articles looked at how AI agents were being used in production:

• Agentic coding with Google Jules
  Software developers are among AI’s most enthusiastic fans, and Google Jules is an agentic coding assistant with real heft. It fixes bugs, adds documentation, and integrates with your GitHub repos.
• How LinkedIn built an agentic AI platform
  The careers behemoth built an enterprise-scale agent AI deployment, using an agentic platform that leverages distributed application techniques. Here’s a candid look at the real architectural decisions and practical engineering patterns used for agentic systems at scale.
• Multi-agent AI workflows: The next evolution of AI coding
  Now multi-agent systems are emerging, with coordinated workflows capable of completing complex coding tasks. Agents are starting to interoperate in real development contexts by sharing state, governance, and human-in-the-loop control mechanisms.
• How AI agents will transform the future of work
  AI agents are already reengineering software development, business processes, and customer experiences. What’s next?

Multi-agent systems? New protocols make it possible

As autonomous agents are embedded in real workflows, the next challenge is getting them to talk to each other and the tools they depend on. This year, open standards like the Model Context Protocol moved from experimental specs to practical infrastructure, enabling agents to share context, invoke external services, and participate in coordinated multi-agent workflows across environments:

• A developer’s guide to AI protocols: MCP, A2A, and ACP
  The big three emerging agent communication standards—MCP for tool and data access, Agent-to-Agent (A2A) for peer collaboration, and ACP for messaging—help agents interoperate and operate in real systems. Here’s a guide to all three.
• AWS’ Serverless MCP Server to aid agentic development of managed applications
  AWS released a serverless implementation of the Model Context Protocol, giving AI agents real, context-aware access to cloud tools and data to help design, deploy, and troubleshoot applications, bringing MCP into the world of practical engineering workflows.
• 10 MCP servers for devops
  MCP is also being integrated into devops tooling, including server offerings from GitHub, AWS, Grafana, and Akuity.

Why code when you can vibe?

If AI agents are increasingly doing the heavy lifting of writing and coordinating code, it’s fair to ask what’s left for the rest of us to do. Enter vibe coding—a playful, almost rebellious approach to coding with AI. Some of the year’s most popular reads captured the excitement, the absurdity, and the potential dangers of working with AI-generated code:

• Vibe code or retire
  Like it or not, vibe coding is here, and developers need to take it seriously. This article takes the stance that embracing AI-driven code generation isn’t optional; it’s a survival skill for modern developers.
• Writing code is so over
  Nick Hodges takes it a step further, declaring that traditional coding is becoming obsolete, just as hand-coded assembly vanished after reliable compilers became available. Will developers soon command systems in spoken English rather than hand-typed code?
• Is vibe coding the new gateway to technical debt?
  This article strikes a nerve, addressing vibe coding’s potential to amass more bad code, particularly when used as a substitute for real learning and experience.