Perform I/O, Error Handling and Validation
By the end of this lesson, you'll be able to
- Read and write files safely with with, choosing the right method (.read(), .readline(), direct iteration, .seek()) for a given situation
- Parse and write both JSON and CSV data, and handle the specific ways each format can fail to parse
- Define your own exception hierarchy, and use raise/raise ... from to re-raise or translate a low-level failure into a clearer one without losing the original cause
- Replace print()-based debugging with the logging module — levels, formatted output, file-based logs, and logger.exception() for errors specifically
- Combine all of the above into one realistic pipeline: reading external data, validating it against a Pydantic model, and handling every way that can go wrong as one coherent, well-logged failure
Why it matters
An agent’s entire relationship with the outside world runs through I/O — reading a config, loading eval data, writing results, calling a tool that might fail. None of that data arrives guaranteed to be well-formed, and none of those operations are guaranteed to succeed. This lesson is about building the habit of treating “the file might not exist,” “the data might be malformed,” and “something needs to log this before it fails” as the normal case to design for, not an edge case to bolt on afterward — exactly the posture real agent code running against real, messy external data actually needs.