Concurrency and Async

By the end of this lesson, you'll be able to

  • Explain why sequential code wastes time on I/O-bound work, and what "concurrency" actually means as a fix for that specific problem
  • Write async def coroutines, use await, and avoid the "called without await" gotcha — a coroutine object that never actually runs
  • Use asyncio.gather() to run multiple independent operations concurrently, and asyncio.create_task() when you need to start something before you're ready to wait for its result
  • Avoid the specific trap of a blocking call (like time.sleep()) silently freezing an entire event loop from inside a coroutine
  • Handle errors in async code — both a normal try/except around await, and gather()'s return_exceptions=True for collecting partial failures instead of losing every result to the first one
  • Recognize when async is the wrong tool entirely — CPU-bound work needs a different answer (multiprocessing), not async/await

Why it matters

An agent that calls three tools, or three different LLM providers, or fans a single request out to several data sources, is doing exactly the kind of independent, I/O-bound waiting this lesson is built around. Writing that sequentially — waiting for each one to fully finish before starting the next — is correct but slow, in a way that compounds badly as an agent calls more tools per turn. Everything in this lesson, from gather() to return_exceptions=True, is aimed at the same real target: agent code that does several things at once, and handles it gracefully when one of them doesn’t come back cleanly.