Fast, Helpful, and Wrong: Critical Thinking for AI-Assisted Delivery
AI can write code, summarize an incident, generate Terraform, and sound entirely confident while being entirely wrong. For engineering teams that creates a new category of risk, and it runs opposite to intuition: the better the output sounds, the easier it is to stop thinking critically about it. The failure is not really in the model. It is in the review that the model's fluency talked out of happening.
This talk is about what critical thinking looks like once AI is part of software delivery rather than a novelty beside it — where it earns its place in platform, cloud, and DevOps work, where it fails in ways that are dangerous rather than merely wrong, and how those failure modes actually show up: hallucinated infrastructure changes, shallow root-cause analysis, bad assumptions buried inside generated code, and confident summaries that skipped the context that mattered.
What you’ll take away
- Where AI actually earns its place in platform, cloud, and DevOps work — and where it does not.
- The failure modes that matter in practice: hallucinated infrastructure, shallow root cause, assumptions hidden in generated code, overconfident summaries.
- How source verification, review loops, sandboxing, and better prompting work as habits rather than policy.
- Where policy guardrails and clear ownership have to sit once AI touches production-facing work.
- A framework for keeping speed from quietly replacing thinking.
Who it’s for
Engineers and engineering leaders working out how AI fits their delivery process without loosening their grip on production. Fits conference AI tracks, internal engineering sessions, and leadership offsites; attendees leave with a framework they can apply immediately.
Book this talk in the format that fits
This talk adapts to your slot — pick the length and shape that fits your event.
- 20-minute lightningSharp single-idea talk for short slots, openers, or AI tracks
- 40–50-minute conference sessionStandard conference / track session length
- 60-minute internal talkEngineering all-hands, internal summits, brown bags
- 90-minute practitioner briefingDeep, opinionated working session for senior engineers
- Panel or podcast guestDiscussion, debate, mainstage panel, or guest interview
Invite Brian to give this talk.
“Fast, Helpful, and Wrong: Critical Thinking for Shipping Software with AI” is available for conferences, internal engineering events, leadership offsites, panels, and podcast interviews. Tell us your audience, format, and timing.
Invite Brian to Speak ← All talk topicsWant the same lens in podcast form? Browse the SRE episodes on Ship It Weekly.