Published May 11, 2026
The market loves specialized agents: a mail agent that reads your inbox, a code agent that writes your functions, a news agent that summarizes your feeds. Each one is trained for a narrow task.
It sounds efficient, and for a fixed, well-defined context it can work well. But when your context keeps changing, with new projects, shifting priorities and evolving relationships across multiple businesses, specialization starts to crack.
A mail agent reading your business emails doesn't understand your business. It processes text, but it doesn't know your vendors, your margins or your ongoing negotiations. A code agent doesn't know why you're building what you're building. It writes functions in isolation and misses context that would have changed the approach entirely. Each agent sees a sliver, and none of them sees the whole picture.
I found that if you give a capable AI the right context and clear rules, it can handle almost anything. You don't necessarily need a different agent for each task. You need one AI with deep vertical knowledge and well-defined norms.
Think about how you'd work with a trusted person in a company. You don't have a "spreadsheet person", an "email person" and a "meeting person". You have someone who understands the business and can handle whatever comes up. That person is effective because they have the full picture, not because they're specialized.
EIDARA treats agents as documents, not as programs. A file called agent-librarian.md contains the complete librarian protocol. Any AI that reads it becomes the librarian, whether it's Claude, GPT or DeepSeek.
To share an agent, you share a file. To modify an agent, you edit a file. To create a new one, you write a file. The intelligence lives in the document, not in whoever executes it. That removes the platform dependency and lets any AI step into any role, as long as it has the right context.
I manage 4 businesses with one AI that has complete context, not six agents that each see a sliver. It can answer questions about operations, review a contract, draft content or summarize a strategy document, because it knows the full picture.
In my experience, specialization is often a workaround for missing context. If you want to test this, take one task you'd hand to a specialized agent, write the context and rules into a single file, and give that file to the general AI you already use.
Javier
In EIDARA v2 an agent is a file any AI can read. It’s free to download.
EIDARA is the open-source project behind this approach. See it on GitHub or on the Website.