Cleaning house: fewer agents, more clarity

What I wanted to fix

In the previous post, I covered the company meeting: we defined who does what and what the human never gives up.

Once those boundaries were set and I decided to build my personal website, another practical problem became obvious in my daily routine. I had collected too many agents. Between early September and the end of that month, every new idea felt like an excuse to open the terminal and set up another assistant. If I wanted help writing copy, I created a profile. If I wanted to review data or test commands in the terminal, I booted another one. If I read about a different prompt format, I duplicated an existing assistant just to see how it handled two extra sentences.

That initial excitement carried a real cost. I work during the day, and at night and on Sundays I build with AI agents. My free time in front of the computer is limited, and it needs to count. Instead of sitting down and getting things done, I found myself scrolling through long lists of profiles, trying to remember which version was up to date and which was an abandoned draft. Even worse, when I needed to assign a quick task, I hesitated because I had no idea which window to use.

Having too many assistants scattered my focus instead of helping. On September 29 and 30, 2026, I stopped running new experiments and cleaned house. The plan was direct: run a strict inventory of everything I had built, delete the clutter, and keep only the assistants that genuinely helped in daily practice.

Quick summary

• In the previous post, I covered how I aligned roles between myself and my assistants during the company meeting.
• On September 29 and 30, 2026, I ran a full inventory of my assistants to cut out what was unnecessary.
• I found duplicate profiles, forgotten test setups, and assistants with vague duties that only slowed down my routine.
• I deleted the excess and kept a lean laptop environment, naming Optimus as the leader for local agents.
• Cleaning up cost nothing in cash, took two evenings of focused work, and gave me back the speed I needed.

First: what an agent profile is and what an inventory means

If you are new to this topic, the word “profile” inside a system like Hermes Agent points to something very concrete. Each profile is an independent assistant with its own name, folder, and instructions file detailing its role, rules, and limits.

You can picture this setup like a hanging file folder in an office desk. Each folder holds a sheet for a specific role, describing what files belong inside and what duties fall to it. The comparison helps visualize the separation, but it stops working at one practical point: in the physical world, creating a new folder costs money, paper, and physical space in a cabinet. On a computer, creating an AI profile takes under a minute and takes up virtually no physical room. Because it is so easy, the digital filing cabinet turns into a mess before you know it.

Taking an inventory simply means pulling every folder out of the drawer and laying them on the desk. You write down the name of each agent, where it runs, what it is supposed to do, and when you last opened it. If you cannot describe an agent’s purpose in a single clean sentence, that agent does not belong on your machine.

What I used

During this cleanup on September 29 and 30, 2026, I relied on three things:

• Hermes Agent installed on my laptop (Hermes Desktop), which held my local assistants and temporary test setups.
• Hermes Cloud, the remote environment running background jobs that need to continue when my laptop is closed.
• A plain text document to list every profile and mark what to keep, merge, or delete.

How I cleaned house

I tackled the cleanup across four practical steps over two evenings.

1. List everything without exception. I went through the Hermes directories on both my laptop and the cloud server, writing down every profile in a single list. The result caught me off guard: I had over a dozen profiles, many with vague names like “test,” “helper,” or numbered copies of the same assistant.
2. Run the utility test. For every name on the list, I asked an honest question: did this assistant produce a useful result in the last seven days? If the answer was no, or if I took longer than ten seconds to recall what it was built for, it went straight onto the discard pile.
3. Delete duplicates and test drafts. Without sentimentality, I erased entire folders that were only copies with slight prompt variations. If two roles could be handled by the same assistant with clear guidelines, I merged the instructions and deleted the duplicate.
4. Pick a local leader for the laptop. On my personal machine, I organized the surviving agents under clear leadership: Optimus took charge of coordinating local actions in Hermes Desktop, working alongside Segurança local, the agent in charge of watching over the laptop itself.

By the end of the second evening, the local profile list had shrunk by more than half. The main program menu fit on a small screen again without any scrolling.

What went wrong and what I learned

My biggest struggle was letting go of old test files. When you spend thirty minutes writing detailed instructions for an assistant, you feel an urge to keep that file around forever, telling yourself it might come in handy someday. In reality, that archived text never gets reopened. It only clutters your search results and adds mental noise.

Another beginner mistake was believing that a larger roster of agents makes a system more capable. In practice, a bloated menu only slows you down. Every idle agent adds another option on screen, another outdated instruction file, and another pause when you need help fast. A system with twenty confused assistants loses badly to a system with three well tuned assistants that know their jobs.

What became obvious over those two evenings was that keeping things simple takes daily discipline. Keeping a small set of tools with clear boundaries is what allows a regular person to sit down at night, finish a project, and avoid getting lost in configuration menus.

Costs and limits

In financial terms, cleaning house cost exactly $0. Deleting folders, revising instruction files, and reorganizing Hermes required only my own time and judgment, with no new software purchases or service upgrades.

In terms of time, the cleanup took two evenings of focused work on September 29 and 30, 2026, right after my normal workday.

The practical limits of this cleanup were also clear:

• A cleanup does not sustain itself automatically. If I lose discipline, the natural pull over the following weeks will be to spawn disposable profiles and build up clutter all over again.
• It requires regular audits. Every few weeks, you have to run through the inventory again to see if the assistants you kept still carry their weight.
• Leaner agents cannot fix poorly written prompts. An assistant only performs well if the initial instructions saved inside it are sharp and direct.

What I would do differently

If I had known this in early September, I would not have created dedicated profiles for quick experiments.

Now, whenever I want to test a prompt idea or a response structure, I run a temporary conversation or sketch a quick draft. I only create a permanent profile in Hermes when a chore repeats across multiple days and justifies keeping dedicated memory and guidelines. Following that rule from day one would have saved me hours of typing and later pruning.

Who this is for and who it is not for

Running a strict inventory and trimming your agent roster is useful for:

• Anyone who started experimenting with AI tools and already feels overwhelmed by too many open windows, menus, and assistants.
• People who only have a few hours each week to study or build, and cannot afford to waste time on digital clutter.
• Anyone who wants full visibility into what every program is doing on their machine, without relying on forgotten scripts.

This practice is not for:

• People who measure progress by the quantity of bots they build rather than the real problems they solve.
• Casual users who open an assistant once a month and do not care if their files are organized.
• Environments running dozens of parallel experimental prototypes with no intention of putting them to practical use.

Wrapping up

Cleaning house on September 29 and 30, 2026, brought needed calm to my daily routine. Pruning the agent roster down to what mattered, removing duplicates, and establishing Optimus as the laptop leader turned a chaotic pile of experiments into a dependable working tool. Having fewer options on screen leaves more mental clarity to build.

In the next post, I will cover how I built the newsroom: research, fact-checking, writing, art, and editing.

If you find yourself hoarding assistants and shortcuts you never open, try a simple exercise today: open your tools, list what you have, and delete whatever you have not touched this past week. Drop by on X and let me know how it felt: @RodrigoesPorto.

Sources and links

• Official Hermes Agent documentation: https://hermes-agent.nousresearch.com/docs
• My official profile on X: https://x.com/RodrigoesPorto