This is Fine! A podcast about resilience engineering and software · Colette Alexander and Clint Byrum

The Messy 9 and Coding with AI - A Panel Discussion

·1 hr 43 min·4 clips

Renowned safety researcher explains why humans must interpret new technology through existing frameworks.

Clint and Colette start with weekend logistics and weather. It is Monday, and the mood is already part tired, part amused. Snow days, a cold snap, a sick kid, skiing, and hockey all come up before the technical talk begins. That opening works because it sounds like people talking first, and panelists second. The conversation turns when a speaker compares older programming advice to the newer habit of reaching for a large language model first. The old regular expressions joke gets a fresh AI spin. Now the issue is not just that the tool was used. It is that whatever comes out still has to be supervised. The panel keeps coming back to that idea. One person points out that good engineers already make judgment calls about when not to use an LLM. That judgment is part of the craft. Another draws a line to rubber ducking, where explaining a problem out loud can expose what is actually wrong. The group treats that as both familiar and newly relevant. An LLM can act like a more responsive duck, but it can also tempt people to skip the thinking that matters. That tension sits at the center of the discussion. They do not rush toward a neat conclusion. They pause, qualify, and reframe. Someone notes that the hard part is often not getting an answer, but knowing whether the answer deserves trust. That is the real burden of AI-assisted coding. The conversation keeps circling the difference between externalizing a problem and solving it. Those are not the same thing, and the panel is fine leaving that distinction visible. There is also a social texture to the way they talk about work. People describe the craft as something done with other people in mind, even when the task is solitary. The humor stays dry and close to the surface. The technical language stays plain enough to follow, but specific enough to matter. Terms like supervision, reflection, and elicitation end up doing real work in the argument about how engineers think. The episode never pretends AI coding is simple. It treats it as a changing habit with tradeoffs that depend on the task. Some uses are clearly worth it. Others create more trouble than they remove. That practical split is the point of the whole conversation. The panel is trying to describe a workflow that is still being negotiated in real time. The closing stretch returns to gratitude and future plans. It feels like a group that knows the discussion was useful even if it stayed messy. That messiness is not a flaw so much as the shape of the subject. AI in software engineering is not a settled story. It is still moving, and the episode stays close to that fact. The useful idea here is that AI does not remove the need for judgment. It makes judgment more visible. Best for software engineers, technical leads, and listeners trying to decide where AI belongs in day-to-day coding. It also fits people who like technical conversations that wander a bit before they land. Not for listeners who want a polished demo, a single clean answer, or a short intro to AI coding.

As heard by us

A loose panel that treats AI coding as a supervision challenge, not a magic trick.

A panel spends most of its time on the practical shape of coding with AI: where LLMs help, where they turn into a supervision problem, and when the better move is to leave them out.

Read the full review in PlayNext →

Why you'd press play

Play this if you want a candid talk about coding with LLMs and the supervision problem they create.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from This is Fine! A podcast about resilience engineering and software