One general agent with a skill library beats 47 custom agents
Teams cram every procedure into monolithic agent prompts. Agent Skills replace them with one general agent and a composable library that loads knowledge in layers.
Written by
-
Benedikt Stemmildt Founder & Co-CEO
Published: 27 November 2025 Updated: 2 min read
Your AI agent’s context window is collapsing under the weight of your organization’s procedures.
Every team is building 47 custom agents. Nobody knows what each one does. Knowledge is trapped in bloated system prompts that cannot scale.
Agents are still monoliths
We keep thinking about AI agents as monoliths. One agent. One massive prompt. Every procedure jammed in. This is how we did software 40 years ago, before we learned libraries and composition. Most teams building agents today are still in monolith mode.
What Agent Skills are
Anthropic just released Agent Skills, an elegant answer to this exact problem.
The concept: instead of 47 custom agents, you have one general agent and a composable library of skills. A skill is a folder containing a SKILL.md file that packages organizational knowledge: instructions, scripts, resources. The agent discovers skills, loads them dynamically and applies them when needed.
Progressive disclosure in four levels
Context loads in layers:
- Metadata: name and description. The agent knows its options without loading everything.
- Core instructions: the full SKILL.md with procedures and guidance.
- Deep references: additional files, loaded only for edge cases.
- Executable code: scripts for deterministic operations.
Building your skill library
- Start with gaps. Where does your agent struggle? That is a skill.
- Structure for scale. When a SKILL.md gets unwieldy, split it.
- Think like Claude. Monitor usage. Skill names and descriptions decide when a skill triggers.
- Iterate with Claude. Codify actual patterns, not anticipated ones.
Two paths for teams
One path: 47 custom agents, tribal knowledge, fragmented capabilities. The other: composable skill libraries with discoverable capabilities that keep scaling.
Learning and teaching
Last week I wrote about context progression: how agents improve their own playbooks over time (ACE). This week is about context composition: how agents compose organizational knowledge into specialized capabilities.
The two complement each other. ACE is how agents learn. Skills are how organizations teach. Together, they answer the real question: how do we build AI agents that actually understand our business?
Are you still building custom agents for every use case, or are you architecting a composable skill library?
Further reading
Written by
-
Benedikt Stemmildt · Founder & Co-CEO
More on Coding agent setup
- Shallow agents fail at step 20, deep agents plan and remember The prompt, tool, result loop works for 5 to 15 steps and collapses on real work. Deep agents add explicit planning, delegation to sub-agents, persistent memory and heavy context engineering.
- Use subagents as researchers, not implementers Subagents that implement directly lose context between sessions. Subagents that research, write their plans to the file system and leave the implementation to the main agent work in production.
- Generic coding agents hit a limit, custom agents go past it Claude Code, Cursor and Gemini CLI are built for the average codebase. Teams that scale with agents move from better agents to more agents to custom agents built for their own domain.
Scaling needs named owners
Coding agents scale across teams when named people own the platform, the team practice, the guardrails, and the measurement. Rollout & scale is where that ownership is built and handed to internal owners, in your stack.
Training on this topic: Agentic Engineering for Developers and QAWant to check one team first? The self-check: twelve yes or no questions, three minutes, instant result, no account. Take the self-check