# My agents are not tied to one model

By Niyaz Irsaliev, 2026-10-23

Today I switched to a new coding agent and barely noticed. Here is why.

For eight months I have done almost everything through AI agents. For most of that time they ran on Anthropic models. I also work with Codex, Grok Bot, Hermes and Muse. Each one is good at something different, so I use each for what it does best and switch between them when the task calls for it.

Today, on my friend Askhat's advice, I tried Claude Code again. A while ago it felt inconvenient to me. It has been reworked a lot, and I am starting to use it.

The interesting part is how little the switch cost me. I barely noticed it.

## Why the switch was easy

Every service my agents use sits behind its own MCP server. MCP is a standard way to plug a tool into any AI agent. Passwords, Cloudflare, video production, transcribing my voice notes, my knowledge base: each one is wrapped the same way.

So a new agent does not have to learn anything. It connects to the same servers and gets the same tools. Codex, Claude Code or Grok Bot, the plumbing stays the same.

Passwords come through Bitwarden. Cloudflare is another good example. Each bot gets its own access, and I never have to log in there myself or learn a new API.

## The real strength

I think the strength of anyone building with agents today is not the model they use. It is how fast they can move to any other bot or model, including local ones, without rebuilding everything.

Models keep changing. If your services are tied to one tool, every switch is a rebuild. If they all speak the same standard, a switch takes an afternoon.

## Where I keep my data

I keep my data locally on my own machines, not in GitHub or the cloud. Hacks happen, a private repo can be made public by mistake, and clouds go down.

If you go this way, keep two copies in two places far apart, and set up a proper RAID on your storage.

This is the first post in a series. Next I will go through my services one by one.