2 min read

How to write an agent brief that actually produces useful work

How to write an agent brief that actually produces useful work

Hook

Last month, I watched three different Hermes AI agents fail at the same task. The owners had spent hours setting them up, configuring Docker Compose, and wiring them into Telegram. But the agents gave vague, random answers. One replied with a recipe when asked about a support ticket. Another wrote a poem about downtime logs.

The problem was not the software. It was the agent brief.

Background

One of my clients runs a small operations team inside a Telegram group. They handle internal reminders, ticket assignments, and simple follow-ups. They bought AgentForge to automate those tasks. I installed the Hermes agent, deployed it with Docker Compose, and opened the web UI. The team wrote a brief that said: "Be a helpful assistant for operations."

The agent was useless. It answered every message with generic positivity. No deadlines. No owners. No action items.

Problem

The brief was too vague. The model had no guardrails and no specific context. It did not know what "operations" meant for that team. It did not know what a done task looked like. It did not know that silence after a reminder meant the task was still incomplete.

The team blamed the agent. But the real failure was in how they described the agent's job.

Why it matters

A bad brief wastes time. The team spent fifteen minutes a day reading irrelevant answers. They stopped trusting the agent. They went back to manual work. The cost was not the subscription — it was the lost productivity and the broken workflow.

With a good brief, the same agent could have reduced manual follow-up by seventy percent. The difference was a few lines of text in the configuration.

Solution

I rewrote the agent brief using three rules. First, I declared the agent's role in one sentence. Second, I listed exactly what inputs to expect and what outputs to produce. Third, I gave it a short list of behaviors to always follow.

Here is the rewritten brief for that operations team:

You are an operations coordinator for a team of five. You receive task assignments and reminders via Telegram. For every incoming message, you must reply with a confirmation, a deadline timestamp, and the name of the responsible person. If no person is named, you must ask for one. You must never guess a deadline. You must never give an answer that is not related to task tracking.

I typed that into the web UI. That was it. No prompt engineering tricks. No system prompt templates. Just a clear, concrete description of what the agent should do.

Result

The agent started working immediately. The first message after the change was a ticket about a server restart. The agent replied with the task owner, a deadline in UTC, and a confirmation checkbox. It ignored every off-topic message. It never wrote poems.

The team now uses it daily. They estimate thirty minutes of manual follow-up saved per day. The agent brief took me two minutes to write. That two minutes is the only thing that separates a useful agent from a random text generator.

AgentForge

AI agents for real work. If you want to see how this works in practice, visit our product site and blog.

Share
Serhii

Serhii