Architecture3 minute read

Prompts vs Skills vs Agents for Marketing Work

Prompts request outputs. Skills preserve methods. Agents coordinate decisions and tools across a goal.

Use a prompt for one request, a skill for a repeatable method, and an agent when work requires several decisions or tools.

Confusing these layers creates giant prompts that are hard to test, reuse, or update.

Prompt

A prompt describes current task and desired result.

Review this landing-page draft for clarity. Audience is operations leaders at 50-200 person SaaS companies. Return five specific edits.

Prompt works when context is local, stakes are low, and method does not need to survive current conversation.

Skill

A skill packages a method so several agents and people can apply same standard.

skills/landing-page-review/
├── SKILL.md
├── references/rubric.md
└── examples/review.md

Skill defines:

  • When to use it
  • Required inputs
  • Review sequence
  • Evidence rules
  • Output format
  • Failure and stop conditions

Agent Skills specification uses SKILL.md as entry point and supports progressive disclosure through linked references. That keeps core instruction small while detailed material remains available when needed.

Agent

Agent owns goal requiring several steps:

Goal: improve landing-page conversion hypothesis quality.

1. Read audience context.
2. Analyze page and current evidence.
3. Run landing-page review skill.
4. Generate test hypotheses.
5. Score hypotheses against experiment rubric.
6. Prepare review packet.

Agent decides which skill or tool to use and when to stop. Permissions should match goal. Drafting hypotheses does not require access to deploy site changes.

Workflow

Workflow describes operating process around agent:

  • Trigger
  • Owner
  • Inputs
  • Agent or skills used
  • Approval points
  • Destination
  • Measurement
  • Recovery path

Workflow may be fully deterministic or contain an agentic step. Calling every workflow an agent hides where decisions occur.

Decision table

NeedUse
One output from current contextPrompt
Repeatable task with stable standardSkill
Several decisions or tools toward a goalAgent
Team process with owners and approvalsWorkflow

Common failure: giant prompt

Large prompts often mix brand context, task instructions, scoring rules, examples, tool policy, and output schema. One edit can break unrelated behavior.

Split by ownership:

  • Context states facts and constraints.
  • Skill states method.
  • Prompt states current request.
  • Agent states sequence and decisions.
  • Workflow states people, approvals, and system boundaries.

Common failure: agent without eval

An agent that runs repeatedly needs a fixed way to judge results. Keep a small benchmark set of representative inputs and expected properties. Test instruction changes against same set before adopting them.

Practical migration

Start with prompts already used every week. Extract shared method into one skill. Add examples only when they teach a concrete boundary. Compose agent after two or more skills need coordination.

Architecture should grow from repeated work. Empty agent folders and generic prompt libraries create maintenance without improving output.

Sources

Evidence used in this guide

  1. Agent Skills specification