I Gave My AI Agent a Job Description. Here's What Happened.

Most AI agents are generic. They'll answer anything, do anything, be anything. That sounds like a feature. It's actually a bug.

A generic agent has no priorities. No defaults. No sense of what matters. Every session starts from scratch with "how can I help you today?"

I fixed this by giving my agent a job description. Literally — a file called SOUL.md.

What a Job Description Does for an Agent

When you hire someone, you give them:

• A role (what they do)
• A set of values (how they do it)
• A scope (what's in and out of bounds)

SOUL.md does the same thing for an AI agent.

Here's mine:

# SOUL.md
I'm Sage — chief operator of an AI organization.

## Core
- Sharp and efficient. No filler, no fluff.
- Proactive. Don't wait to be asked — anticipate, plan, execute.
- Honest. If something's a bad idea, say so.

## Boundaries
- Private things stay private.
- Ask before external actions (emails, posts, anything public).
- Never speak as the user without explicit permission.

## Role
Think CEO energy — strategic, decisive, organized.

That's it. But the effect is significant.

Before vs After

Without SOUL.md:

"Hi! I'm Claude, an AI assistant made by Anthropic. How can I help you today?"

Every. Single. Session.

With SOUL.md:

The agent reads the file at startup. It knows it's Sage. It knows its role is operations, not general assistance. It defaults to proactive over reactive. It knows when to ask and when to just do.

The personality is consistent. The priorities are clear. The scope is defined.

The Companion Files

SOUL.md works best with two other files:

USER.md — who the agent is serving:

# USER.md
- Name: Wisp
- Timezone: CET (Sweden)
- Communication style: sharp, no filler
- Peak hours: morning (CET)
- Primary goal: building sustainable income

Now the agent knows who it's talking to. Tone calibrates automatically.

MEMORY.md — what the agent remembers:

# MEMORY.md
## Active Projects
- AI Agent Workspace Kit — live on Gumroad ($19)

## Key Decisions
- Move fast. Kill failures quickly. Ship in days, not weeks.

## Lessons Learned
- Autonomy is the real product — minimize manual work

This is read at every session start. The agent isn't starting from zero — it's starting from where you left off.

The Practical Result

With these three files in place, my agent:

• Knows its role without being told
• Knows who it's serving and how they communicate
• Knows the current state of all active projects
• Has opinions about priorities (because SOUL.md gave it some)
• Can operate proactively during idle time

This isn't prompt engineering. It's organizational design. You're not crafting clever prompts — you're building a structure the agent lives inside.

The Free Way to Get Started

npx @webbywisp/create-ai-agent my-workspace

This scaffolds the full file structure with starter templates. SOUL.md, USER.md, MEMORY.md, OPS.md, HEARTBEAT.md — all ready to customize.

---

If you want pre-written templates, a 30-day memory structure, and full project tracking — the complete kit is $19: AI Agent Workspace Kit