I handle growth for a small AI product. To keep up with LinkedIn, X, SEO articles and outreach, I set up a separate AI agent for each job. One writes LinkedIn posts, one writes replies on X, one plans blog articles, one helps with outreach, and one plans my day.
On paper this sounds like a small team. In practice, I spend a lot of my day doing one thing: explaining the same context again and again.
The problem I kept running into
Every agent starts with a blank head. Each time I open a new chat, I have to tell it:
- What the product does
- Who we are trying to reach
- Which words we never use
- What we posted yesterday
- What changed in the plan this week
The LinkedIn agent does not know what the X agent wrote this morning. The planning agent does not know that a launch date moved. If I forget to mention something, the output looks fine but is wrong in small ways.
What I tried
1. One giant prompt with everything in it.
It worked for about a week. Then the plan changed, and I had to update the same text in five places. Half the time I forgot one.
2. Shared documents inside a project.
Better. Each agent could read the same master doc. But the docs got long, some parts went stale, and agents still had no idea what the other agents had done.
3. Writing a short "today" note by hand.
This helped the most, but it is manual work every morning. It also depends on me remembering everything.
What I learned
After a few weeks, I noticed the agents themselves were rarely the problem. The writing quality was fine. The problem was context:
- Context lives in many places. The plan is in one doc, the rules in another, yesterday's posts somewhere else.
- Context goes stale. A static brief from last week gives confident but outdated answers.
- More context is not always better. When I gave an agent everything, it pulled in details that had nothing to do with the task.
- Agents do not share what they know. Each one works alone, even when they serve the same goal.
What I think a better setup needs
If I could design the ideal system, it would have four things:
- Selected sources. I choose what each agent can see.
- A clear trail. When an agent uses a fact, I can see where it came from.
- Fresh context. When the plan changes once, every agent sees the change.
- Shared goals. Agents working on the same campaign see the same current picture, while private notes stay private.
Where I am now
I still use project docs and a daily note, and it works well enough for one person. But the more agents I add, the more time goes into keeping them in sync instead of doing actual work.
This problem is also the reason the team I work with is building Pryveo, a context and memory layer for people and AI tools.
If you use more than one AI agent or assistant in your work, how do you keep them all working from the same context?
This post was written with help from AI for drafting and editing. The experience and setup described are my own.
Top comments (0)