Chaotic Commits | Software, AI, and Uncomfortable Truths

Joanne Skiles

The podcast that says the thing everyone in the room is thinking but nobody puts in the post-mortem. Chaotic Commits covers software engineering, AI, cloud architecture, and engineering leadership (not that sanitized LinkedIn version, the real one). Architecture decisions that almost broke everything. Management lessons learned the hard way. The bigger questions about who gets to build the future and what that actually means. New episodes every week. Hosted by Dr. Joanne Skiles, AWS Community Builder, engineering leader, and speaker at AWS re:Invent, AWS Summits, GHC, and more.

Episodes

  1. feat: I know they're just pixels, but can't you be nice?

    3d ago

    feat: I know they're just pixels, but can't you be nice?

    My seven-year-old was trash-talking a Pokémon trainer. The way a seven-year-old does. And I stopped him: "I know they're just pixels, but can't you be nice?" That simple moment led to a much bigger question: What are we practicing when we interact with AI systems, chatbots, NPCs, voice assistants, and virtual characters? In this episode of Chaotic Commits, we explore what Tamagotchis, ELIZA, Roombas, Alexa, ChatGPT, and modern AI systems can teach us about human behavior, AI ethics, transparency, and the habits we build when interacting with technology. The question isn't whether AI has feelings. It's what we're practicing when we interact with increasingly human-like systems. Topics: AI ethics, artificial intelligence, ChatGPT, human-AI interaction, AI transparency, anthropomorphism, responsible AI, human-in-the-loop systems, software engineering, engineering leadership, AI design, and behavioral psychology.

    27 min
  2. raise e: the line that logged everything

    Jun 5

    raise e: the line that logged everything

    It's 8:30 PM, and I'm lying in my four-year-old's bed playing Spiderman when the text comes in. "yoo, we are logging all of tenant config in prod logs has secrets in plain text." Not test secrets. Not staging. All of the secrets. Production. In the logs. This episode is the story of that night: what raise e actually does in Python, why a valid exception handler became a production security incident, and what "log scrubbing" actually means when your logs are in a SIEM, a compliance archive, and systems your vendor manages. We cover secret rotation when you're two incidents deep in forty-eight hours, and your customers are involved. The thing I keep coming back to: logging is an architectural decision, not a default. Most teams find out they never made that decision when they're staring at a screen at 10 PM, calculating how long tenant API keys have been sitting in plain text. If you've been on-call, worked through a security incident, or built anything that handles customer secrets, this one is for you. What you log is a decision. Own it before someone else finds it. Topics: production incident, secrets in logs, Python exception handling, raise e, incident response, log scrubbing, API key leak, SIEM, compliance, on-call engineering, backend security, logging best practices, engineering leadership, tenant security

    18 min
  3. scope: undefined

    May 29

    scope: undefined

    A manager once told me, "If you get the work done, they will want more." He meant it as advice (and was a little sarcastic too). He was wrong about what to do with it. He wasn't wrong about the dynamic. This episode is about what happens when capable engineers become the workaround. The script you run once to fix prod becomes the script you're running for 300 users two weeks later. The emergency process becomes the permanent process. Eventually, the organization concludes that we don't need the system. We have you. In this episode, I talk about scope creep at the individual level, not the project level. It's what happens because you're good at your job and you keep saying yes. But to escape, you need to know how to hold the line without just saying no. You have to build a "no" that sticks. If you've ever walked out of a meeting wondering how you became someone's permanent workaround, this is for you. Topics: engineering leadership, scope creep, technical debt, engineering culture, career advice, software engineering, setting boundaries, organizational debt, engineering management

    20 min
  4. (bonus) cache miss: the search for something google won't let you find

    May 26 ·  Bonus

    (bonus) cache miss: the search for something google won't let you find

    I want to tell you about the last time I actually enjoyed using Google. I don't remember exactly when it was. That's kind of the point. At I/O 2026, Google announced an AI-first search experience. Instead of links, you get a synthesized answer. Conversational. Efficient. And I think we just killed something important. This episode is about three connected things nobody is saying out loud: what we lose when search becomes a conversation, why the dead internet theory isn't paranoid anymore, and what AI training has to do with killing the incentive that made the open web worth building in the first place. If the AI draws its knowledge from the web, and the web stops having a reason to produce original knowledge, the whole system starts eating itself. If you've ever followed a link from a link from a link at midnight and ended up somewhere you didn't know you needed to be, this one's for you. Topics: AI search, Google I/O, dead internet theory, knowledge discovery, open web, AI training data, search engine optimization, AI accountability, transparent AI systems

    15 min
  5. hotfix: everything I know about distributed systems, I learned from Costco

    May 22

    hotfix: everything I know about distributed systems, I learned from Costco

    Every computer science textbook explains distributed systems with boxes and arrows. I'm going to explain it with a warehouse in the suburbs where you can buy a kayak, a television, and a 48-pack of paper towels in the same trip. The membership card is your API key. The one-way entrance is your traffic ingress. The free sample person is the throttle (one per customer, retry window twenty minutes). The no-windows design is what an unobservable system feels like from the inside. The receipt check at the exit isn't verifying payment; it's output validation. The $1.50 hot dog is an SLA commitment so load-bearing that the CEO has publicly stated it will never change. And the tire center you've never used is your disaster recovery runbook. Every concept in a distributed systems course is visible at Costco on a Saturday. The parking lot is DNS. The gas station line is your permanently bottlenecked service. Kirkland Signature is vendor abstraction. This is the episode I wish someone had recorded before I sat through my first distributed systems lecture. The boxes and arrows aren't the insight. The people trying to get a rotisserie chicken on a Saturday afternoon are. Topics: distributed systems, load balancing, rate limiting, API design, SLA, observability, vendor abstraction, canary deploys, dead letter queues, DNS, capacity planning, system design

    25 min
  6. merge conflict: when the diff is 40k lines and the author is the owner

    May 15

    merge conflict: when the diff is 40k lines and the author is the owner

    It's Monday morning. You open your PR queue. Branch: codex/something. Two hundred files changed. Forty thousand lines of AI-generated code. Description: nothing. Just the check boxes for bugfix and new feature are checked. The author is a non-technical co-founder. This episode introduces accountability debt - the gap between who captures the efficiency win from vibe coding and who gets paged at 2 am when that AI-generated code breaks in production. I will break down why large AI-generated pull requests create a code review crisis, no PR template was built to handle, what actually works when the engineer who needs to push back reports to the person holding the PR, and how to frame technical risk as business risk when appealing to the engineering process goes nowhere. If you're an IC staring at an unreviewable 40,000-line diff from someone with more authority than context, or an engineering manager trying to build AI code governance that actually holds - this one is for you. AI writing code is not the problem. The problem is what your team inherits when no one asks who owns it during an incident. Topics: AI-generated code, vibe coding, code review, pull request review, accountability debt, technical debt, AI code governance, engineering leadership, startup engineering, non-technical founders, AI developer tools, on-call culture, engineering management, software development process

    18 min
  7. chore: you are no longer an individual contributor

    May 8

    chore: you are no longer an individual contributor

    What nobody tells you before you become a manager: the skills that got you promoted are the exact skills that will trip you up in the new job. Writing the code is no longer the job. The team is the job. And learning that takes longer than it should. I learned it during a User service migration that should have gone cleanly. I had the blueprint. I ran the same migration myself, textbook perfect, six months earlier. I gave my team every step. But I didn't give them the reasoning behind the steps. The migration failed. And when I had to reflect on whose fault it was, the answer was uncomfortable. In this episode: the identity shift that comes with leaving IC work and the grief nobody names, why there is no single right management template and how to find yours, the accountability inversion (what it actually means when your wins belong to your team and your failures belong to you), and what it looks like inside organizations where managers never learned this. This episode is for engineers making the move into management and wondering why it feels wrong even when they're trying hard, technical managers who are still writing code their teams should be writing, and anyone who has ever handed someone a process without handing them the reasoning. Topics covered: engineering management, IC to manager transition, technical leadership, new manager advice, engineering team accountability, software engineer career growth, managing software engineers, management mindset shift, engineering leadership podcast, microservice migration lessons, developer career advice, first-time manager. If you've given someone the blueprint and watched them fail, and had to sit with why, this one is for you.

    20 min
  8. feat: I let an AI run a D&D campaign and it tried to kill everyone

    May 1

    feat: I let an AI run a D&D campaign and it tried to kill everyone

    I built a serverless AI dungeon master on AWS. Within twenty minutes, a player had broken it completely. Within forty minutes, a dragon had killed everyone at the table. Including the people who hadn't done anything wrong. This is the story of building a real AI agent system using Amazon Bedrock Agents, AWS Lambda, API Gateway, and DynamoDB, and then watching what real users actually do with it. We go deep on the architecture: how Bedrock Agents handles tool orchestration and game state, why a three-second Lambda timeout will absolutely kill your AI inference workload, and what happens when an LLM gets an API contract that even AWS can't implement consistently. But the technical failures aren't the point. The point is what a Dungeons and Dragons campaign reveals about production AI systems: the gap between what you designed for and what users actually do. Hallucinations, tool misuse, unconstrained autonomous behavior, and the cost of building AI systems you can't observe. The same failure modes that made my friend's halfling rogue get eaten by a dragon show up in clinical decision support tools, customer service bots, and production AI agents. The sandbox is just where you can afford to find them first. If you're building agentic AI systems, or thinking about it, this one's for you. Dev.to Article: https://dev.to/aws-builders/building-a-serverless-dungeon-master-agent-on-aws-3j7k Topics: AWS Bedrock, serverless AI, agentic AI architecture, LLM tool use, prompt engineering, AI observability, AWS Lambda, API Gateway, DynamoDB, production AI failures, responsible AI design

    22 min
  9. hotfix: turns out smart wasn't enough

    Apr 24

    hotfix: turns out smart wasn't enough

    What does "senior software engineer" actually mean? Why isn't being the smartest person in the room enough to get there? In this episode, I break down the real ingredients of engineering seniority: technical depth, the judgment you can only build by being catastrophically wrong in production, and the career trap of staying comfortable in a pond that stopped teaching you things. Featuring a production incident when I was a mid-level engineer that ended up costing the company way more than my year's salary, and the moment the bosses laughed -- a reaction that quietly shaped everything that came after. This episode is for software engineers who have ever wondered if they've actually earned the title, engineering managers trying to calibrate what they're hiring for, and anyone who's shipped something that worked in staging and absolutely did not work in prod. Topics covered: senior engineer career growth, production incidents, engineering judgment, technical leadership, software engineering seniority, developer career advice, pair programming failures, and building experience as an engineer. If you're building a career in software and you want the honest version of what it takes, not the LinkedIn version, this one is for you.

    20 min
  10. fix: stopped waiting for AI to fold my laundry

    Apr 17

    fix: stopped waiting for AI to fold my laundry

    It was 10:30 PM. The kids were asleep, the meeting summaries were done, the inbox was triaged... and the laundry was still on the floor. In the first episode of Chaotic Commits, I, a software engineer with a PhD and 15 years in tech, ask the question that lies hidden in that moment: Who is AI actually efficient for? From Amazon's infamous gender-biased recruiting algorithm to building AI tools for exhausted radiologists, this episode explores why the productivity revolution keeps skipping over entire categories of human labor, and why diverse teams aren't just a values issue; they're a design quality issue. If you've ever wondered why the best AI tools feel built for someone else's life, this episode is for you. Topics: AI bias, ethical AI, women in tech, inclusive design, transparent systems, diversity in engineering, workplace productivity, healthcare AI

    18 min
  • 10 Episodes

About

The podcast that says the thing everyone in the room is thinking but nobody puts in the post-mortem. Chaotic Commits covers software engineering, AI, cloud architecture, and engineering leadership (not that sanitized LinkedIn version, the real one). Architecture decisions that almost broke everything. Management lessons learned the hard way. The bigger questions about who gets to build the future and what that actually means. New episodes every week. Hosted by Dr. Joanne Skiles, AWS Community Builder, engineering leader, and speaker at AWS re:Invent, AWS Summits, GHC, and more.

Information