Levels of AI: Prompt, Skill, Connector, Plugin and Cowork

Abstract
This guide explains the five levels of AI, from chatbots to AI that runs an organisation, and the working parts a photographer or filmmaker meets on the way: the prompt, the skill, the connector, the plugin, the custom GPT, the design system, and the three ways of working called Chat, Code and Cowork. It says what each one is, what it is for and how they fit together in one workflow. The tools change often, so the guide teaches the ideas and names tools only as examples.
Keywords: Levels of AI; chatbots; reasoners; agents; prompt; skill; connector; plugin; GPTs; design system; Cowork
What are the five levels of AI?
The five levels of AI are a map of how far AI has come: Level I chatbots, Level II reasoners, Level III agents, Level IV innovators and Level V organisations.
A five-level scale of AI progress was reported in 2024 as an internal framework at OpenAI, and it is now widely used to explain where AI stands. This guide uses it as a map, not as a ranking of products.
- Level I: Chatbots. AI that converses in everyday language. You ask, it answers.
- Level II: Reasoners. AI that solves problems at a human level, such as PhD-level science questions and coding problems. You give a hard problem, it thinks it through.
- Level III: Agents. AI that takes actions. It maps a goal, plans a multi-step workflow, uses outside software and manages its own errors without a person at each step.
- Level IV: Innovators. AI that helps with invention. It could discover new scientific knowledge, new frameworks and new technology.
- Level V: Organisations. AI that could do the work of a whole organisation, running a workflow from top to bottom.
An example for each level, from a photographer's week.
- Level I, chatbot: you ask, "What aperture suits a portrait?" and it answers in a few lines.
- Level II, reasoner: you give a hard problem, "My indoor photographs look blue under tube lights. Work out why and what to change," and it reasons through white balance and gives a fix.
- Level III, agent: you say, "Sort these forty photographs into chapter folders and rename them," and it plans the steps, does them, and corrects its own mistakes.
- Level IV, innovator: an AI that proposes a new method of measuring light that nobody has published, and proves that it works. This is a goal for the future, so it is only an illustration.
- Level V, organisation: an AI that runs a whole photo studio, with bookings, shoots, editing and billing, from top to bottom. This is also a goal for the future.
Where a photographer works. Day-to-day work sits at Levels I to III: you chat, you ask it to reason, and you let an agent do multi-step jobs on your files, such as sorting a shoot into folders. Levels IV and V describe where AI may go, and they are not part of everyday photographic work today.
Why we do this. Knowing the level tells you how much to hand over. A chatbot needs every instruction. An agent needs a clear goal and clear limits.
The higher the level, the more the AI does by itself, and the more careful your limits must be.
- List the AI tools you use.
- Write which level each belongs to.
- Note what you hand over to each and what you keep.
What is a prompt?
A prompt is the written instruction you give an AI tool. It lasts one chat, and it is the first part of every other tool in this guide.
A good prompt gives the tool the same brief an editor would give you: a role, a task, steps in order, limits and rules. A prompt lasts one chat. When you want an instruction to last, you save it as a skill. How to build a prompt, step by step, is taught in Prompt Creation for Photographers.
Example. "List ten shots for a two-hour harbour shoot at dawn. Give each shot a one-line reason." That one instruction is a prompt.
A prompt is the instruction. Everything else in this guide is a way to keep, share or extend it.
- Find one prompt you type again and again.
- Rewrite it with a role, a task, steps, limits and rules.
- Test it once and note what to keep for a skill.
What is a skill?
A skill is a saved playbook, a folder of instructions with optional examples or scripts, that tells the AI how to do one kind of task. The AI loads it whenever a request fits.
A skill is a folder of instructions, and it can include example files or small scripts. You write it once. After that the AI loads it when your request matches, so you do not paste the same long prompt every time.
Prompt and skill. A prompt is typed or pasted for one chat. A skill is saved and used again and again.
Example. A skill for client emails holds your tone, your price rules and your sign-off. Every time you ask for a reply, the AI writes it in the same way.
Why we do this. A skill keeps your method the same each time and saves credits, because the instruction is not retyped.
Example in use. You save a "client email" skill that holds your tone, your price rules and your sign-off. A month later you type, "Reply to this client," and the AI loads the skill and writes the reply your way, with no long prompt pasted again.
Make a skill from a prompt you have already tested. Never save an untested prompt.
- Pick one prompt you have already tested.
- Write what its skill would be called and when it should load.
- List any example files it would need.
What is a connector?
A connector is a live link between the AI and an outside app, such as your files, calendar, email or website. It lets the AI read from that app, and sometimes act in it, with your permission.
Many connectors are built on the Model Context Protocol, called MCP, an open standard for linking AI tools to apps and data. A skill tells the AI how to do the job. A connector gives it access to the place where the job is done.
- What it is for. Reading a document in your drive, or putting a draft into a page, without you copying and pasting.
- Safety. Connect only the apps you need. Know what the AI may read and what it may change. Ask it to show you before it sends, publishes or deletes anything.
Why we do this. A connector saves the carrying of files between apps. Your control over permissions keeps your work safe.
Example in use. You connect your drive. You type, "Read the file Harbour-Shoot-Plan and make a shot list," and the AI opens the file itself. You do not upload or paste anything.
Give an AI the least access that does the job. You can always give more later.
- List the apps you would connect first.
- Write what the AI may do in each: read only, or read and act.
- Choose the least access that does each job.
What is a plugin?
A plugin is an installable package that bundles skills, connectors and commands, so a person or a team can add a whole way of working at once.
A skill supplies the how, a connector supplies the access, and a plugin is the packaging that delivers both. A plugin can carry a set of skills for one job, the connectors those skills need and shortcuts to start them.
Example. A photography-class plugin could carry a photo-sorting skill and a client-email skill, and a connector to the class folder, ready for every student to install.
Why we do this. A plugin means every student works the same way without building it from nothing.
Example in use. The teacher installs a "Photography Class" plugin once. Thirty students then have the same photo-sorting skill and the same class-folder connector, with no setup of their own.
Build the skill first and the plugin last. A package is only as good as what is inside it.
- Choose one job you repeat often.
- List the skills and connectors a plugin for it would carry.
- Write one shortcut command it would offer.
What are GPTs?
A custom GPT is a version of ChatGPT that holds your instructions and reference files, so it works your way every time for you or anyone you share it with.
A custom GPT saves a set of instructions, reference files and sometimes links to outside services inside ChatGPT. You or anyone you share it with can open it and start working at once. It is ChatGPT's way of keeping a prompt and its material together.
Other tools keep the same idea under other names, such as Gems in Gemini and Projects in Claude. The names differ, and the idea is the same: saved instructions and files, ready to use.
Why we do this. A custom assistant saves you from pasting instructions and files again and again.
Example in use. You build a GPT called "Harbour Shoot Planner" that holds your shot-list rules and your notes on the harbour. You send the link to your students, and each of them gets the same planner.
Test a custom assistant on three photographs before you share it.
- Write the instructions your own assistant would hold.
- List the reference files it would need.
- Test it on three photographs before you share it.
What is a design system?
A design system is a fixed set of rules for how everything you make looks: colours, fonts, spacing and standard parts such as buttons and cards, with when to use each.
A design system names the colours, the fonts, the spacing and the standard parts, such as buttons and cards, and says when to use each. A person can follow it, and an AI can follow it too.
In AI work. You give the AI your design system once. After that, every page, poster or album it makes matches your look without you repeating the rules.
Why we do this. Without a design system, every output looks different. With one, the work looks like one hand made it.
Example. Your design system says: charcoal text, one ember accent colour, one serif font for headings and one plain font for text, and rounded cards. You ask the AI for a poster for your exhibition, and the poster comes in those colours, fonts and cards.
Write the rules before you ask the AI to design. The AI follows rules better than moods.
- Write your own colours.
- Choose two fonts and say where each is used.
- Describe the three parts you use most, such as a button or a card.
What are Chat, Code and Cowork?
Chat, Code and Cowork are three ways of working with AI: Chat is conversation, Code builds software, and Cowork does multi-step work on your files and apps.
- Chat. You write, it answers. It works at Level I and II. Use it to ask, to plan, to draft and to check.
- Code. An agent that writes and runs software for you, such as a script or a website. It works at Level III, with the tools of a programmer. Claude Code and OpenAI's Codex are examples.
- Cowork. An agent that works on the files and apps on your computer with your permission. It sorts photographs, renames files, makes folders and builds documents. It works at Level III, and it needs no coding.
An example for each.
- Chat: "Explain f-stops in simple words." You get an answer and you ask more.
- Code: "Build a small web page that shows my ten best photographs in a grid." The agent writes the code, runs it and fixes errors, and gives you a working page.
- Cowork: "Sort the forty photographs in the Working-Copy folder." The agent renames the files and puts them into chapter folders on your computer.
Code and Codex are not the same word. "Code" is a kind of tool: an AI that writes and runs software. "Codex" is the name of one product of that kind, made by OpenAI. "Claude Code" is another product of the same kind, made by Anthropic. Think of a camera and a Canon R5. A camera is the kind of tool, and the Canon R5 is one product. Both Codex and Claude Code do the same job, so learn the kind first and the product names after.
The same is true of the other two. Chat is the kind, and Claude, ChatGPT, Gemini and Grok are products of it. Cowork is the name Claude uses for working on your files. Other tools offer similar modes under other names.
Several AI companies offer these modes under different names, and the features change often, so check your tool's help page.
Why we do this. Choosing the mode is choosing how much you hand over. Chat suggests, Code builds, Cowork does.
Start in Chat. Move to Cowork when the job is the same steps on many files.
- List your next three AI tasks.
- Choose Chat, Code or Cowork for each.
- Write one line on why for each choice.
How do they fit together in one workflow?
In one workflow, the prompt says what to do, the skill saves how, the connector gives access, the plugin packages them, the design system keeps the look, and the agent does the steps.
- You sort the shoot with an agent in Cowork. See Sorting Your Photographs with AI.
- You draft captions, posts and client replies with a prompt in Chat. See Prompt Creation for Photographers.
- You save the tested prompts as skills, so the method is the same every time.
- A connector brings your photographs and your drafts from your drive to the AI, and puts finished pages back.
- A design system keeps the look of your album or page.
- A plugin packs all of it for a class.
You check each stage. The AI drafts and you decide. To learn how to build a prompt, read Prompt Creation for Photographers.
Example of one full run. A harbour shoot. In Chat you plan ten shots. You shoot, then copy and back up your files. In Cowork the agent sorts the forty photographs into folders. In Chat you write the captions from your facts. A saved skill writes the client email in your tone, a connector fetches the plan from your drive, and the design system keeps the album page in your look. You check every step.
Build from the bottom: prompt, then skill, then connector, then plugin.
- Draw your own workflow in six boxes.
- Label the tool used in each box.
- Mark which box you would build first.
How to cite this guide
Alex, A., & Joseph, A. (2026). Levels of AI: Prompt, skill, connector, plugin and Cowork. Creative Hut Academy. https://creativehut.org/library/levels-of-ai/
Make this your craft at Creative Hut Academy
Creative Hut Academy in Mattakkara, Kottayam, Keralam teaches photography, cinematography, editing, sound and AI filmmaking in one-year residential diplomas. Students come from across India.
Want to talk? Message us on WhatsApp, or ask Cham, our AI mentor, a study doubt.