Demystifying AI Agents with Python Code (PyOhio 2025)

Transcript

1. Demystifying AI Agents with Python Code William Horton
2. Everyone’s talking about AI Agents
4. But what are AI Agents?
10. Is it all just marketing hype?
12. My opinion: “AI Agents” are currently overhyped, but at the core there is useful functionality that is worth exploring.
13. Who am I and why do I care about AI Agents?
14. Who am I? ML Engineer applying GenAI in Healthcare Twitter/X: @hortonhearsafoo BlueSky: @horton.hearsa.foo
15. Maven Clinic Maven is the world's largest virtual clinic for women and families, supporting our members through fertility, pregnancy, parenting, and menopause. Maven uses Python! We are hiring remotely. Many different roles: AI/ML, Backend, Data, Infra, Analytics
16. Thank you PyOhio! PyOhio 2018 was my first time ever giving a conference talk
17. In this talk I hope to show you how to translate the concept of “AI Agent” from the high-level terms into concrete Python code that you can understand.
18. Code https://bit.ly/agents-pyohio
19. Real-World Example Coding Agents: There is no “magic” to something like Cursor Behind the scenes it is an LLM agent like what I demonstrated, with a lot of development dedicated to the tools. Coding Agents usually have tools that mimic command-line tools like: - ls - grep (or more advanced search techniques) - cat/tail/head - cd
20. Practical tips The agent can only understand the functions you give it as far as you’ve: - given the function and parameters clear names - written an informative docstring - added parameter types (and descriptions as needed, see typing.Annotated) If the agent has trouble, you can give it examples of correct “trajectories” of using the different tools Performance tends to decline as number of tools increases
21. Thank you! Twitter/X: @hortonhearsafoo BlueSky: @horton.hearsa.foo https://bit.ly/agents-pyohio