I was looking at a competitor’s docs the other day—I won’t say which one—and I counted. They had 15 named agents. Here are a few of them: a code review agent, a testing agent, a documentation agent, a deployment agent, and a “planning” agent. Each had its own configuration, its own persona, and its own set of hardcoded instructions.
And I thought, “Isn’t this just 15 config files wearing trench coats?”
The Agent Explosion Is a Confession
When a product ships 15 specialized agents, what it’s actually telling you is, “Our AI can’t figure out the right approach from context, so we need you to pick the right mode.”
Think about what a “review agent” does differently from a “coding agent.” It reads code. It reasons about it. It produces output. The difference isn’t capability—it’s which files it reads, what it’s looking for, and what format the output takes. These are context differences, not capability differences.
A model that knows your team’s review standards, testing conventions, and deployment process doesn’t need a mode switch. You say, “Review this PR,” and it reviews it the way your team reviews PRs—because it knows how your team reviews PRs. You say, “Write tests,” and it writes tests using your testing patterns—because it’s seen how your team writes tests.
The agent for each task only exists because the AI doesn’t know enough to handle the task with its general intelligence. It’s a crutch for missing knowledge.
Slash Commands Are a Step Backward
This one really gets me. We spent 50 years building natural language understanding, and the industry’s answer to “make AI tools easier” is ... a command-line interface?
All of a sudden, you’re confronted with /review, /test, / commit, /explain. You’re learning a new vocabulary, so you can tell the AI what to do in a format that’s less expressive than English. The whole promise of AI-powered development is “describe what you want in natural language.” Slash commands are the opposite: Memorize our taxonomy, and invoke capabilities by name.
The other day, I watched a junior engineer try to use a tool with slash commands. She typed, “Can you review this code?” The tool didn’t do anything—it was waiting for /review. The engineer didn’t know the command existed. So she Googled it, found the docs, learned the syntax, and then typed the same request in a less natural format.
That’s not AI-augmented development. That’s a chatbot with macros.
The Maintenance Tax Nobody Mentions
Here’s the thing about specialized agents that nobody talks about: Someone has to maintain them. Think of it as a maintenance tax.
The review agent has instructions about your code review standards. Great. Your standards change: You adopt a new testing framework; you change your PR size policy; you start requiring security review for certain paths. Who updates the review agent? The same person who was supposed to update the CLAUDE.md. The same person who was supposed to update the wiki.
And now you have N agents times M rules, instead of one config file with all the rules. The maintenance burden didn’t decrease—it multiplied. Each agent is a separate surface area for staleness.