Continuing education in AI: choose what to learn next
By Fabrice PAYETPublished Updated
An AI learning plan works best when you can name the task you cannot yet do. In 2026, that might be evaluating a retrieval system, deploying a model safely or explaining uncertainty to a product team. Start with that task. You do not need to collect every new course or framework.
Audit a task, not your entire career
Choose one responsibility from your current role or a target job. Write down what competent delivery would look like: inputs, expected output, constraints and a review method. For a classification task, that could mean selecting a baseline, preventing leakage and explaining the cost of false positives.
Compare those requirements with work you can already demonstrate. Separate a missing foundation, such as statistics or Python, from an unfamiliar library. A new framework may need a short documentation exercise; weak evaluation skills deserve sustained practice.
Choose the learning format that closes the gap
Use official documentation for a specific tool question. Choose a structured course when you need a sequence, exercises and feedback. Consider a degree when its depth, supervision or qualification requirements match your target role, and compare the time and cost before committing.
Google’s Machine Learning Crash Course covers foundational modeling, data preparation and production considerations. Treat any syllabus as a starting point: inspect prerequisites, exercise quality and the provider’s revision history. Reading a lesson is not evidence that you can apply it.
Make the project your assessment
For a suggested retrieval project, create a small set of questions with expected answers, include cases with no relevant document, and compare against a simpler search baseline. Keep a record of data access, evaluation decisions and failures. These are proposed exercises, not reported Dataaxy experiments.
Write down what changed after each iteration. Explain whether a result improved because of the model, the input data or the test setup. Reproduce the work from a clean environment and ask another person to review the assumptions.
Keep a manageable study cycle
Our suggested cycle is one focused topic, one practical exercise and one review before choosing the next topic. Reserve a repeatable study slot that fits your work. If the exercise reveals a foundational gap, return to that gap rather than adding another course.
For credentials, check the official exam page before paying. Exam availability, requirements and retirement dates change. A past credential can remain part of your history, but an unavailable exam should not become your next learning objective.
Reassess after you can demonstrate the skill
Save a concise account of the task, method, result and remaining limitations. Use it to decide whether to go deeper or move on. NIST’s AI risk framework can help identify evaluation and governance questions beyond model performance. Learning should improve a decision or a deliverable you can explain.
Sources and next steps
Sources checked on October 2, 2026. Follow the linked documentation for current requirements.
Google Machine Learning Crash Course
scikit-learn: leakage and consistent preprocessing
