In the last post we built a skill together, and a question came to me. Why create a multi-agent architecture when I can have one agent and multiple skills?
I have been thinking about this, because the whole industry is pushing us toward the multi-agent layer by default. But when is it actually necessary, and when is it not?
Let’s look at this with my own examples.
My single agent does more than I expected
I have a personal assistant running 24/7 based on OpenClaw. That single agent does different kinds of tasks, like reading my AI newsletter every morning and sending me a summary over Telegram, or accessing my calendar to know what will happen during the day. None of this needs a second agent, only tools (email, calendar, Telegram) and a few skills that tell it what to do and when.
For a personal user, I think a single, well-equipped agent goes further than we think. It’s a question of giving it the right tools and skills.
3 reasons to only use one agent
- Easier to debug. If something breaks, there is only one place to look. With multiple agents passing work to each other, a bug can hide in the gaps between them.
- It’s cheaper, because every new agent is a new API call. But this one is debatable. A single agent caarries all its tools and context in every call, so each call is heavier. But in general, fewer agents means a smaller bill.
- Keeps everything in one context, and you don’t have to pass any information around.
3 reasons to add a second agent
- You need different models. For example, you can have a very simple agent that classifies information with a cheaper model.
- The context does not fit and involves thousands of documents. In this case it’s better to split the information between different agents.
- There is real parallelism, and it makes sense to have two agents working at the same time.
Skills and agents solve different problems
This is the idea that made everything click for me. A skill is a new ability you give to one agent. A second agent is a new brain, with its own tools and its own context.
So the real question is not one agent or many agents. The real question is: do I need a new ability, or a new brain? And most of the time, the answer is a new ability. And a new ability is a skill.
My conclusion
Start with one agent and give it good tools and good skills. Only add a second agent when you really hit one of the three reasons mentioned above.
The industry will keep selling you multi-agent. But for a personal user like me, one agent with the right skills has been enough.
Related reading: what an agent harness is.