One job.
Then it stops.
An agent is a worker with one specific task, a guard around what it can touch, and a person who signs off. A prompt in front of a web API is not an agent. We do not let them run loose.
A job. A guard. A person.
-
1
The job
One task, named. Quote follow-up. The close exceptions. Not "go run the company."
-
2
The guard
The tools it may use, the data it may see, and the actions it may not take. Sending and posting wait.
-
3
The stop
When judgment is required, it waits. A person is named on the work, and that person decides.
A prompt is not
an agent.
The market is calling a web API an agent because a prompt sits in front of it. It can call a system and return text. It has no job, no guard, and no one who has to say yes. That is the thing people are afraid of, and it should not be released.
- ChatbotIt answers
- API wrapperIt calls and returns
- ScriptIt always runs
Marquis Agents stay on the job.
Each one is paired with a person and trained around one task. It does the work around that job. It waits when a decision is required. It does not roam.
Short answers.
If it calls our API, isn't that an agent?
Calling an API is a tool. A prompt in front of that API can look busy and still have no job, no guard, and no one accountable. That is an imposter. An agent has one job, a limit on what it may touch, and a person who signs off.
Isn't every chatbot an agent now?
No. A chatbot answers the person in front of it. An agent is given a job, does the work around that job, and stops when a person has to decide.
Will agents take over?
Not these. The fear is an agent with no guard and no person, turned loose on the business. We do not build that. A Marquis agent does one specific task and waits for a sign-off. It does not roam.
Why do the guards matter so much?
Because the agent can touch real systems. Without a guard it can see what it should not, send what no one approved, and leave no record. The guard is what makes a worker safe to put on a job.
Why not one agent for the whole company?
That is how they get loose. One agent, one job, one person. A different job gets its own agent, with its own guard.
What is an agent?
A digital worker with one specific job, a short list of allowed tools, and a person who signs off before anything that matters goes out.
The agent does the work. You make the call.
One job. One person. A guard around both.
