Jonathan Peacher

PyTexas 2025

Day 1

Keynote: Mariatta

❤️

Reducing the “Oops Factor”: Pipelines for Securing your Python Development Lifecycle on a budget

• More projects are having security issues, especially now with AI and vibe coding.
• A “swish cheese” approach to securing your codebase, most of which can be integrated into your CI pipeline.
  • Secret Detection: gitleaks
  • Dependency Scanning: pip-audit
  • SAST (Static Analytis Security Testing): bandit
  • DAST (Dynamic Analysis Security Testing): Dastardly / Tenable
  • YOU

The Pythonic Ideal in The Age of Generative AI

• AI ← LLM ← Coding Assistant
• An experienced engineer is defined by their experiences, not by the duration of their role.
• Pros/Cons of using a coding assistants:

• If you are a beginner, write code first, then use AI to explore when stuck, but make sure to understand the generated code.
• If you are experienced, let AI write the boring stuff (boilerplate, consistent code, etc.), and level up by prompting for code beyond your comfort zone (ask for options).
• If you are a teacher, use AI from the start. Ask it to suggest alternatives to existing code.
• Will AI replace you? Maybe. It will if you refuse to evolve with it.
• Is pythonic code dead? Maybe. But be kind to future developers (including yourself).
• Takeaways:
  • LLMs are just fancy autocomplete
  • If we over-rely on AI when first learning, we run the risk of not gaining the experience needed to grow.
  • AI is a tool, nothing more.
• Advice:
  • Don’t panic
  • Learn your craft
  • Embrace AI
  • Teach the next generation
  • Keep it pythonic

Generators: The Unsung Hero of Async Programming

• The GeneratorExit exception can be to cleanup generators.
• You can send data back into a generator using .send(x) and x = yield y.

Demystifying AI Agents with Python Code

• What are AI agents?

“agent = llm + memory + planning + tools + while loop” – Tweet

• You can give the ChatGPT API memory by including the previous messages in subsequent API requests.
• You can tell the ChatGPT API to use tools (functions), to do things like searching the web, that returns structured data.
  • You can use .signature() to generate the function specs for these tools.
• Tools like crewAI abstract all this away.

Python Untethered: Building Robust Embedded Systems

• What are embedded systems?

“…[an] embedded system is a device that contains a computer that the average person would not consider being a computer.” – Reddit

Lighting Talks

• pip install interrogate for docstring coverage.

Day 2

Keynote: Jay Miller

• Contributing to open source and community are similar.
  • You have an idea and want to share it with others.
  • When you give away your ideas, they only tend to get better.
• “Creating a new open source package (and community) is like adopting a puppy.” - Pamela Fox
• Let people know what help you need, and vice versa.
• Autocorrect “LGTM” to “Looks good to me, thank you!” and don’t say “just x”.
• There is always more that you could have done, but if you are doing something, it is helpful.
• Respect everyone’s capacity.
• Ambition + Inspration = Scope Creep

Building a Test Framework From Scratch (Or Not)!

• What is a testing framework? A programming structure for coding automated test cases, executing them as a suite, and reporting their results.
• Just use pytest.

Place-making and Productivity: Build Maintainable Broad-scale Tools With a Small Team

• pain = n_teams * size_team * n_tools * n_migrations * (1 + r_turnover)

Verbs, Not Nouns: Writing Documentation Users Want to Read

• Two types: Reference / Tutorial
• Reference (Nowns): Comprehensive, focuses on the product, quick retrieval
  • e.g. man pages, api reference, code coments, readme files.
• Tutorial (Verbs): Not comprehensive, focus on the user, and read beginning to end
  • e.g. tutorials, how-tos, walkthroughs, use cases
• Outline your doc first:
  • What do you want user to do?
  • Ask the users, if you can
  • Write down everything
  • Define discrete steps
  • Prioritize the use cases
• Advice
  • Look for the verbs
  • Focus on what the users want to do
  • Move the verbs up front

Lighting Talks

• LLMs are like APIs. You make a request, and they return data, but with “moods”.
• Micromentor with SLQAR: Show up, listen, question, act, recharge.
• How to say thanks to Python People:
  • Level 0: Actually say thanks
  • Level 1: Say thanks in public
  • Level 2: Show your support
  • Level 3: Contribute/volunteer
  • Level 4: Contribute money
  • Level 5: Nominate them for awards
• Be kind to your screen sharing partner. Just share the window or tab.
• “Make friends, not connections” - Kojo