Learning with AI: one small task instead of ten courses
By Oliver Schmalzl · Written: · Archive discovery date:
21 May 2025 · Archive date
New tools quickly come with a whole reading list: videos, courses, manuals and collections of examples. That can be motivating. But it can also mean a lot gets saved and little gets tried.
A manageable start begins with a single question. For example: can an AI sum up a factual text you wrote yourself in three clear bullet points, without leaving out an important limitation?
An exercise with a visible result
The sample text describes an invented paper plane and states clearly that its flight distance has not been measured yet. The summary should reflect its design and purpose, but keep this open point.
Now the result can be compared: which statements were kept? Which have disappeared? Has something been added that is not in the original text? These questions reveal more about how it works than a general call to "become more productive with AI".
For the next exercise, only one thing is changed. The text gets longer, or it contains two contradictory details. That way it stays clear which extra difficulty is being examined.
Read documentation that fits the question
A public learning resource is helpful when it makes the concrete task easier to understand. You do not have to finish the whole course series before the first small attempt. A vendor's guide should also match the version you actually use.
A learning diary needs no private data for this. The task, your own sample text, the observed result and one open question are enough. This way you slowly build your own understanding, instead of just a long list of well-known product names.
The next step follows from what you observe. If the summary works reliably, another task can follow. If not, it is worth understanding the error better.
Original finds
These links were the starting point for this post. The assessment in the text is my own; the linked posts are not my own tests.
- Public learning resources for AI
- An AI agent in Python as a learning project
- Understanding Claude Code even without programming experience
- Education, creativity and AI as an opinion piece
- An author's first experiences with Codex
- Collection of public AI projects, including n8n, LangChain and DeepSeek.
- Post arguing that AI itself is not the most important future skill; exact argument in the video still open.
- Post about using Codex as a programming agent, with several claimed use cases.
- Andrew Ambrosino describes a big ChatGPT update with Work, a desktop app, plugins and browser features.
- Overview of personal settings, data controls and deliberate configuration of ChatGPT.
About the date: This post was written on 11 October 2026. The archive date is the date of the earliest related find in the collection, not of a publication at that time. Technical assessments and sources added later refer to the time of writing.