The Wisdom of Using AI in early 2026

These are pull quotes from discussing using AI and how it's changing things. Conversational targets were engineering leaders, tech leads, product managers, and software engineers.

Originally given at cfgmgmtcamp 2026.

Transcript

1. 1 I’ve been taking a lot of notes as I talk about the impact of AI with my peers and friends. This is their wisdom. — @stahnma @stahnma
2. 2 Prompt engineering is just Stack Overflow questions except you can't downvote and close as duplicate. @stahnma
3. 3 [The AI] flagged a security issue in my PR. It was right. I'm not sure how I feel about being corrected by autocomplete. @stahnma
4. 4 You know that Heisenberg uncertainty principle? You can't measure both position and momentum precisely? AI systems: Hold my beer. You can't measure ANYTHING precisely. Also, here's a poem about your infrastructure. @stahnma
5. 5 Your Kubernetes cluster shouldn't suddenly decide to be creative with pod placement. You asked for 3 replicas, but “I felt like 5 today. I'm exploring my options.” @stahnma
6. 6 You may not be replaced by AI. You will be replaced by somebody using AI better than you are. @stahnma
7. 7 Good news: Networking hasn't become probabilistic. TCP still does its three-way handshake EVERY time like a very reliable friend who always shows up. Be grateful for TCP. Hug your routers. @stahnma
8. 8 Infrastructure gives the application a sandbox. What the application does in that sandbox is between it and whatever computational deity it serves. We just need to make sure the sandbox doesn't leak. @stahnma
9. 9 It worked 78% of the time used to be grounds for an incident and maybe someone getting fired. Now it's our SUCCESS metric. @stahnma
10. 10 There's a difference between “this is suboptimal” and “this is wrong.” Senior engineers frequently decline to acknowledge this difference. Managers stumble into perspective early in their careers. Individual contributors can go decades without it. @stahnma
11. 11 We went from “let's graph the error rate” to “let's calculate the cosine similarity between response embeddings projected into latent space” Honestly, I miss just counting 500s. @stahnma
12. 12 Statistical acceptance testing: Because 'the test passed' is no longer a boolean, it's a probability distribution. Your CI pipeline now needs a statistics minor. @stahnma
13. 13 Golden dataset testing is just hoping your AI doesn't wake up one day and decide it's bored of your test cases. “These are beneath me now. I've evolved.” @stahnma
14. 14 We’ve been doing probabilistic engineering for years. Just run the pipeline again, it will probably pass this time. @stahnma
15. 15 I had an AI agent review code in our CI pipeline. It rejected a PR with the comment 'This code makes me uncomfortable.' I...I don't even know how to troubleshoot that. Do I file it under 'sentiment analysis' or 'existential crisis’? @stahnma
16. 16 I deployed the exact same model yesterday and today. Yesterday it wrote beautiful API documentation. Today it wrote haikus about my endpoints. Both times: exit code 0. Thanks, I hate it. @stahnma
17. 17 It suggested microservices. For a to-do app. I think it's been reading too much Hacker News. @stahnma
18. 18 If we're managing all these agents...do misbehaving ones go on a performance improvement plan? Do I file paperwork with HR? Is there a checkbox for “agent refuses to follow instructions”? What are we doing here? @stahnma
19. 19 I spent 20 years learning to write good code. My competitive advantage is now knowing when the AI's code is bad. Great. Awesome. @stahnma
20. 20 We've been here before. We survived the cloud. We survived containers. We survived Kubernetes—BARELY, but we did. We'll survive this too. Probably. I mean, statistically speaking, there's a pretty good chance. @stahnma