Over the past few years, the use of artificial intelligence has become commonplace both at work and in education, yet we still use it in very different ways. Some are figuring out how to write better prompts, while others dream of “taking over the world” with AI agents. This short video will help you discover what level you are currently at and how to move on to the next one: The 7 Levels of AI User (and how to level up)
If you look at what the frontrunners in the AI revolution are doing, they are talking about how to give AI system instructions, or skills (aka magic spells), that allow artificial intelligence to complete specific tasks faster, more accurately, and more consistently. Here is a video overview of what you should know before starting to build agents, so you can stay safe in the process, and what the seven new skills of agent builders are.
AI skills (aka magic spells) – the next step in using artificial intelligence
In addition to human skills, there are also AI skills, or “AI skills,” which are essentially specialized abilities or workflows taught to artificial intelligence that determine how it solves a specific type of problem. While a conventional chatbot can answer general questions, a skill can teach it, for example, to check APA references, prepare thesis reviews, analyze cybersecurity logs, or create learning outcomes. A skill typically includes instructions, examples, knowledge, and, where needed, access to external tools and data sources.
In practice, skills are increasingly used in AI agents. For example, an AI assistant working at a university may use separate skills to answer learners’ questions, analyze course feedback, or evaluate research articles. In companies, skills are used to automate customer support, process documents, analyze security incidents, and support project management.
Skills can also be created without programming knowledge. Often, a well-thought-out instruction, or prompt, is enough, along with the necessary files, rules, and examples. In more complex solutions, such as AI agent platforms or workflow automation systems, skills can be connected to email, calendars, databases, and other information systems. In this way, artificial intelligence becomes not just a conversation partner but a digital work assistant capable of carrying out entire work processes. According to experts, AI skills are one of the most important development trends in the application of artificial intelligence in 2026. If until now the focus has been on asking questions, in the future it will become increasingly important to know how to create and use specialized AI skills that help automate repetitive tasks and support knowledge work in education, research, and business.
One place where a TalTech employee may encounter skills is, for example, OpenAI’s new Codex desktop application, which is essentially a workspace for AI agents for developers and, increasingly, for researchers, analysts, and other knowledge workers as well. The program works with files on your computer or on a TalTech server.
How to use Codex?
1. Install the Codex application.
2. Sign in with your ChatGPT account.
3. Open the project folder or Git repository that contains the required source files.
4. Write Codex a task, for example:
4.1 Consolidate the partners’ activity reports into one standardized project report, or analyze the partners’ descriptions and create a table with competencies, work packages, and possible areas of collaboration.
4.2 Create a promotional poster or an introductory presentation for the course using the course materials as the source content.
4.3 Compare the programme learning outcomes with the course learning outcomes and identify potential gaps or overlaps.
4.4 Take a Teams or Zoom meeting transcript and generate a summary of decisions, action items, deadlines, and responsible persons.
4.5 Check APA 7 references in theses, identify errors, and suggest corrections.
5. Codex performs the work in a separate worktree, shows the proposed changes, and allows you to review them before approving and committing them.
To ensure consistent quality every time, Codex needs to be taught how you want it to work. For example, you may want one essential AI skill to be checking APA references, something you do not want to explain to the application every time, but instead trigger like a spell in Harry Potter.
With Codex, you can use skills by writing the skill description as a file and asking Codex to build the code based on it. Some key terms are in English so that the application can better understand your intentions.
Example: you want to create a skill called “APA reference checker.”
Give Codex the following instruction:
1. Write it as a skill file
Skill name: APA Checker
Goal:
Check whether the references in the text conform to APA 7 style.
Input:
text or a list of references
Steps:
Find all references.
Check the author, year, title, and source.
Mark errors.
Provide a corrected version.
Output:
Table: original, problem, correction.
2. Codex builds the required skill and installs it in the Skill Library.
3. Once the skill is installed in your Skill Library, you usually do not need to “run” it manually. ChatGPT will invoke it automatically when your query matches the skill description.
For example:
“Check whether these references conform to APA 7 style.”
“Analyze the references in the following article according to APA.”
“Find APA errors in this bibliography.”
“Check the in-text citations and the reference list.”
Ideas for creating skills:
Here are skills that a lecturer or researcher might genuinely need — in other words, instructions that can be used systematically, so they do not have to be explained to AI every time again:
- apa-checker — APA 7 reference checker
- paper-summarizer — scientific article summaries
- literature-mapper — mapping literature by topic
- research-idea-refiner — refining the research question and hypothesis
- grant-proposal-helper — drafting a grant application text and project idea
- ethics-review-checker — ethics compliance checker
- methodology-reviewer — identifying weaknesses in methodology
- survey-builder — creating a survey
- interview-guide-maker — creating interview questions
- coding-rubric-builder — building an assessment rubric
- exam-question-generator — creating test or exam questions
- lecture-outline-builder — creating a lecture structure
- lab-instruction-writer — writing a lab manual
- csv-analyzer — quick analysis of a data file
- stats-report-writer — writing results in an academic style
- figure-caption-writer — captions for figures and tables
- peer-review-responder — assistant for responding to reviews
- course-syllabus-builder — creating a course syllabus
- moodle-content-creator — creating e-learning content
- plagiarism-risk-checker — flagging risks of textual overlap
At the same time, I also have a question for the university community: could these skills be created collectively, so that we agree on what level of quality a given task should be carried out to? Then anyone who wants to use a university-created base skill could download it from the university’s internal skill library. I imagine that would already save a lot of time, because then everyone would not have to reinvent the same collection of skills, whose quality varies depending on each person’s ability to tame the “kratt.”