Agent Skills

After reading Jaya Gupta's post about Context Graphs, I have not been able to stop thinking about it. For me, it did something personal: it gave a name to the architectural pattern I have been circling around in the agentic infrastructure discussions on this blog for the past year. Gupta's thesis is simple but profound—the last generation of enterprise software created trillion dollar companies by becoming systems of record. The question now is whether a new layer will emerge on top of them: a system of record for decisions.

If 2024 was the year of AI experimentation, 2025 was the year of industrialization. The speculative boom around generative AI has rapidly matured into the fastest-scaling software category in history, with autonomous agents moving from the lab to the core of enterprise operations.

The enterprise AI landscape is at a critical juncture. We have powerful general-purpose models and a growing ecosystem of tools. But we are missing a crucial piece: a standardized, portable way to equip agents with the procedural knowledge and organizational context they need to perform real work.

Anthropic has introduced two powerful but distinct approaches to AI customization: Claude Skills and the Model Context Protocol (MCP). While both aim to make AI more useful and integrated into our workflows, they operate on fundamentally different principles. This post explores their differences, synergies, and the exciting future they represent.