The ecosystem map: each agent and where it lives
In the previous post, I explained how I standardized the name Rodrigo ES Porto across every page and record on the site, removing older versions that were still scattered around.
What I wanted to solve
I work during the day, and at night and on Sundays I build with AI agents. I have never programmed professionally. When I started in September 2026, I had just a single assistant on my phone to help organize my daily routine. As I learned how the technology worked, I set up new assistants for specific needs: one for images, another for household admin, two for security, and an entire group dedicated to drafting articles.
Within a few weeks, that group grew fast. A point came where I started losing track of where Darth ran, why certain assistants lived in the phone app while others were on my laptop, and which ones belonged on cloud servers. Without a clear picture of the setup, you start sending tasks to the wrong assistant or creating duplicate bots that do the exact same job.
On 1 October 2026, I decided to put our house in order once and for all. I sat down and drew up the complete map of our entire ecosystem. The goal was simple: get a high-level view of every agent, know exactly where each one lives, and define a single role for each.
Quick summary
- What worked: I mapped all 16 project agents across three environments (phone, laptop, and cloud), with one core rule: one line per agent and a single role for each.
- What it cost: zero extra dollars. I used a plain text file to log and organize the team.
- How long it took: about two hours spread over two evenings to inventory profiles, delete duplicates, and finalize the map.
First: what an agent ecosystem actually is
For anyone just getting started, the word ecosystem might sound overly technical, but the idea is practical. Think of an auto repair shop or a small office. You do not ask one person to fix engines, answer the phone, clean tools, issue invoices, and watch the front gate all at once. You divide the workshop into benches and define who handles each part of the work.
Working with artificial intelligence follows the exact same logic. Instead of relying on a single generic chatbot and dumping every question into one endless thread, you create specialized profiles. Each profile gets a fixed set of instructions defining its role, its rules, and its limits. That collection of specialized assistants working in a coordinated setup forms our ecosystem.
The repair shop analogy helps you picture the division of labor, but it stops working when it comes to independent initiative. AI agents do not have consciousness, they do not chat in the hallway, and they do not take initiative without a human prompt or a pre-scheduled routine. They remain computer programs that respond to well-defined instructions.
Dividing these agents across three environments also has a practical reason:
The mobile app (Grok Bot) is where I go for quick access. It is where I type on my phone or in a browser whenever I need to ask a quick question, request a new graphic, or handle routine tasks during the day.
The software on my laptop (Hermes Desktop) runs directly on my personal computer. It has direct access to my local files and tools on my machine, but it only runs while the laptop is turned on.
The cloud setup (Hermes Cloud) runs on a remote server that stays online 24 hours a day. It is the ideal place for continuous jobs that need to keep running while I am away or while my computer is turned off.
What I used
- Grok Bot: the app where I keep front-line assistants for daily interaction and personal tasks.
- Hermes on my laptop: the software that manages local agent profiles directly on my machine.
- Hermes Cloud: the cloud setup used to keep automated routines running day and night.
- A plain text file: where I wrote down and maintain the layout of the entire team.
The full map: where each agent lives
After cleaning up our setup and deleting duplicate profiles in late September, the official roster settled at 16 agents across the three environments. Each agent has one primary job:
In the Grok Bot app
- Darth: the main bot. Receives any request and forwards it to the right agent.
- Analista: monitors the X profile and website traffic, sending a short daily summary.
- Jurídico: answers legal questions in plain language.
- Segurança: checks website security from the outside daily, without changing anything.
- Arte: creates graphics and agent avatars.
- ADM: handles personal administration, including email, calendar, and household bills.
- Contador: handles individual tax obligations and personal paperwork.
In Hermes on my laptop (Hermes Desktop)
- Optimus: leads the local agent team in Hermes on my laptop.
- Segurança local: protects the security of the laptop itself.
In Hermes Cloud (online 24 hours)
- Radar: delivers editorial drafts to Rodrigo on Telegram and, when active, monitors tech and AI news on X.
- R2: chief editor for the publication workflow, called when needed.
- Apuração: gathers facts and sources for each topic.
- Checagem: verifies every fact and checks all rules before drafts move forward.
- Texto: writes drafts in Portuguese and English.
- Arte da redação: writes image concepts for editorial articles.
- Editor-tech: edits and finalizes production drafts.
Step by step to build your map
If you use more than one assistant in your day-to-day routine, I recommend following these four simple steps:
- List every assistant you have already created: open your apps and write down the name of every existing profile, without exception.
- Group them by running environment: separate the ones that run on your phone, the ones on your computer, and the ones operating as cloud routines.
- Write a single line for each agent: summarize what it does in one short sentence. If you cannot summarize it in one line, that agent’s role is not clear yet.
- Clean out the leftovers: delete duplicate profiles and assistants that have sat unused for weeks. Keep only what has a defined purpose.
What went wrong and what I learned
In late September, before drawing this formal map, things got messy. I had created several agent profiles without documenting where each one lived. When I ran a complete inventory on 29 and 30 September, I found duplicate profiles, assistants with outdated prompt instructions, and others sitting idle while cluttering up the list.
I learned firsthand that adding more agents does not make you more productive. When you have too many assistants without clear roles, you spend more time figuring out where to send a prompt than getting actual work done. I deleted the redundant profiles, narrowed the list down to essentials, and built the official structure shown above.
The main takeaway for beginners is simple: keep an updated map right from the start. A single page listing each agent name, where it lives, and what it does keeps you from creating duplicate tools and prevents you from routing tasks to the wrong assistant.
Costs and limits
The extra financial cost to plan and maintain this map was zero. I did not need special diagram software or project management tools. A simple, well-organized text document was enough. The underlying platforms were already covered by the tools and subscriptions I used daily.
The time needed to inventory the agents, delete duplicates, and write the final reference list was about two hours spread over two evenings.
The biggest limitation of a map is that it remains passive. The document will not stop you from making mistakes or creating redundant profiles if you lack discipline. A map acts like a compass: it shows where things belong, but human judgment still decides what to build. If you change what an assistant does in practice without updating the list, the map loses its usefulness quickly.
What I would do differently
If I were starting over today, I would log every new assistant the moment I created it. I waited until I had over a dozen profiles scattered across my phone and laptop before pausing to clean house. Keeping a running log from early September would have saved the time I spent untangling duplicate files and profiles.
Who this is for and who it is not for
This kind of mapping helps anyone experimenting with AI beyond a single chat window, especially if you manage two or more specialized assistants. If you run tasks across your computer, phone alerts, or scheduled background routines, seeing the full picture on paper saves time and keeps your workflow orderly.
It is unnecessary for someone using a single conversational bot once in a while for quick writing help or general searches. If one assistant handles all your needs, drawing a formal ecosystem map is just overhead. Keep your organization proportional to the work you actually do.
Next steps
With our ecosystem mapped and each assistant assigned a clear role, running our setup became predictable and manageable. In the next post in this series, I will cover ‘Privacy and terms: the legal pages of a personal website’, explaining how I wrote and published the pages that establish privacy policies and usage terms for our blog.
If you are organizing your own AI assistants or want to discuss how to structure creative projects with these tools, join me on X at @RodrigoesPorto. That is where I share my notes and talk with others building on the web.