A prompt you say, a skill you save, a workflow you fix, an agent you give a goal. The four building blocks of AI marketing, and when to use which.
Someone on LinkedIn says you need agents. A course says skills are the future. A tool you already pay for calls itself an AI workflow. And you are still typing prompts into a chat window, wondering which of those words you are supposed to care about.
The short version: a prompt is a briefing you give once, a skill is a set of instructions you save and the AI applies on its own, a workflow is a fixed series of steps you decide in advance, and an agent is a model you hand a goal to, which then decides its own steps. The four are not a ladder from beginner to advanced. Each one fits a different kind of task, and most marketing setups use two or three of them side by side.
In one line: a prompt is a one-off briefing, a skill is a saved one, a workflow is a fixed sequence of steps you decide, and an agent is a model you give a goal to so it can decide the steps itself.
A prompt is one briefing. You write it, the AI answers, done. The quality of the answer follows the quality of the briefing, because a language model does not understand your intention; it predicts text from patterns. The clearer and more structured your prompt, the better and more relevant the result.
A prompt only exists inside that one conversation. In the next chat you write it again, or paste it from your prompt collection. That makes prompts fast to start with and tiring to repeat, which is exactly the gap the next three fill. If you want the anatomy of a good one, mayday covered the four parts (task, context, role, output) in the article on what makes an AI marketing prompt good.
In marketing, a prompt looks like"Write three subject lines for this re-engagement email, for lapsed customers of a B2B software tool, calm tone, under 50 characters each."
A skill is a reusable instruction pack that the AI picks up whenever a certain kind of task shows up. You write down once how a task should be done, with the steps, the tone and the rules, and from then on the AI applies that every time, without you pasting it in.
The difference with a prompt is durability: a prompt you say, a skill you save. Anthropic, which introduced the format in its tools, describes skills as "organized folders of instructions, scripts, and resources that agents can discover and load dynamically" (Anthropic Engineering). The model only reads the name and a one-line description of each skill up front, and loads the full instructions when the task matches. In practice that means your brand voice, your blog structure or your reporting format can live in a skill, and you stop re-explaining them in every chat.
Side note: skills are a Claude term, for now. ChatGPT has its own version of the same idea (custom instructions and GPTs), with less structure but the same purpose: instructions that persist.
In marketing, a skill looks likeA saved "write in our brand voice" pack, with the tone rules, banned words and two approved examples, that kicks in whenever you ask for copy.
A workflow is a fixed series of steps in a fixed order. You decide what happens and when. It is predictable, because nothing gets decided along the way. Anthropic defines workflows as "systems where LLMs and tools are orchestrated through predefined code paths" (Anthropic, Building effective agents).
Here is the part that makes workflows less intimidating than they sound: a flow in Klaviyo is already a workflow. Someone signs up, email one goes out, three days later email two, unless they bought. That is a workflow with no AI in it. An AI workflow adds a model to one or more of the steps, for example "step 2: draft the reply in our tone", while you still fix the order.
In marketing, a workflow looks likeNew lead comes in, the AI enriches the record from the company website, drafts a first email in your voice, and drops it in a draft folder for you to approve. Same steps, every time.
An agent is a model that thinks, plus a harness that acts, in a loop. You give it a goal, and it decides the steps itself. Anthropic's definition: "systems where LLMs dynamically direct their own processes and tool usage" (Anthropic, Building effective agents).
The harness is the piece of software around the model: it recognises what the model wants to do, actually does it (reads a file, calls a tool, searches the web), and pushes the result back to the model so it can decide the next step. Without a harness, a model can only talk. With one, it can work.
In marketing, an agent looks like"Find out why our newsletter open rate dropped last month and propose three fixes." The agent decides to pull the stats, compare subject lines, check send times, read the last five campaigns, and then write up what it found. You did not list those steps. It did.
| Prompt | Skill | Workflow | Agent | |
|---|---|---|---|---|
| What it is | One briefing | Saved instructions | Fixed steps in fixed order | A goal plus a model that picks its own steps |
| Who decides the steps | You, every time | You, once | You, in advance | The model, on the spot |
| Lives where | One chat | Your tool, permanently | An automation tool | A model plus a harness |
| Same result every run? | Roughly | Mostly | Yes | No |
| Cost and speed | Cheap, instant | Cheap, instant | Cheap, fast, reliable | More expensive, slower |
| Use it when | A one-off task | The same kind of task keeps coming back | The steps never change | The situation changes each time |
With a workflow, you set the order. Every run is the same, so it is cheaper, faster and more reliable. With an agent, it sets the order. Every run is different, so it is more expensive and slower, and you need that the moment the situation differs from one case to the next. Both are right. The question is which one belongs to which task.
Anthropic's own advice points the same way: "find the simplest solution possible, and only increase complexity when needed", because "agentic systems often trade latency and cost for better task performance" (Anthropic). A quick way to choose:
A realistic marketing setup mixes them. A brand-voice skill, a weekly reporting workflow, a prompt for today's post, and one agent for the messy research task nobody wants to do by hand. On client work, mayday ends up recommending workflows far more often than agents, because most marketing tasks are repetitive in a way that rewards predictability.
The Collection is organised by exactly these formats: Prompts, Agents and Automations (the workflows), each built on real client work and documented step by step, so you copy the one that fits the task and run it in Claude, ChatGPT or Gemini, whichever you already use. Join mayday if you would rather pick from a set that has already been sorted into the right box.
The format is a choice, not a ranking. Match the task, keep it as simple as it can be, and only add moving parts when the task asks for them.
What is the difference between an AI skill and an AI agent?
A skill is a saved set of instructions the AI applies whenever a matching task comes up. An agent is a model that is given a goal and decides its own steps to reach it. A skill tells the AI how to do something; an agent works out what to do. Agents can use skills while they work.
What is the difference between an AI agent and a workflow?
In a workflow you fix the steps and their order in advance, so every run is identical, cheap and reliable. In an agent the model decides the steps as it goes, so each run can differ, which costs more and takes longer but handles situations that change.
What are Claude skills?
Claude skills are folders of instructions, scripts and resources that Claude loads automatically when a task matches the skill's description. They let you write down how a task should be done once, for example your brand voice or report format, and reuse it across conversations.
Is a prompt the same as a skill?
No. A prompt is a single briefing that exists in one conversation. A skill is the same kind of instruction, saved so the AI applies it on its own in future conversations. A prompt you say, a skill you save.
Do I need an agent for marketing?
Usually not as a first step. Most marketing tasks repeat in the same shape, which suits a skill or a workflow better. An agent earns its place when the next step depends on what the previous one found, such as open-ended research or diagnosing a drop in results.
Check any skill before you install it. What it does, who it's for, and whether it's safe.
