If you’ve spent the last few months sneaking AI prompts from the couch while your family watches TV, this episode is either going to make you feel seen or significantly worse about yourself. Probably both.
Episode 7 is about AI — specifically, what Tim and James have actually been building with it, why it’s weirdly addictive, and whether any of us know what we’re doing or are just watching a very convincing cardboard town.
The “How Much Time Are You Spending On This” Conversation
James opened with an honest confession: back in January, he was coding with AI all day, every weekend. Sixteen hours a day, by his own estimate. The weather was bad. Why not? It took getting a cold — actual forced bed rest — to make him step back and realize he maybe had a problem.
Tim’s approach is more methodical, and arguably healthier. He writes detailed markdown documents — proper feature specs, basically — and then kicks off the AI to build while he handles the rest of his morning. Gets the kids out the door, makes breakfast, comes back to see what happened. He’s turned software development into an asynchronous process, which is either very clever or just a way to feel productive during school drop-off.
James noted that Claude Code’s Remote tool — which lets you keep sending prompts from your phone while your computer churns away — is “dangerous.” Tim has not adopted this workflow. Writing markdown on a couch is not a dream of his.
The Truss Plate Theory of AI Development
Tim has a go-to analogy for explaining what AI actually does for developers, and it involves roofing. Specifically, the truss plate — a small metal square with spikes that you pound into wood to hold roof joints together. Before it existed, building open-concept roofs was a complex, painstaking craft. After? Faster, easier, more accessible. But it didn’t mean anyone off the street could suddenly build a roof.
“If you and I went to a job site and said, hey, let us finish your house — I wouldn’t live in that house. It would immediately collapse.”
His point: AI is the truss plate. It makes things faster for people who already know what they’re doing. It doesn’t replace knowing what you’re doing. James agreed, adding that he’s skeptical of the “I one-shotted a Twitter clone by just asking” stories making the rounds. His theory: the app looked right, and the person didn’t understand it well enough to know why it was wrong. A very polished cardboard town.
The Database Did Not Survive
Tim learned this lesson the hard way, in week two of building his app, when he had a schema conflict and asked AI to resolve it. It resolved the conflict by deleting the database and recreating it from scratch. Empty. Clean. Problem solved.
“It’s not the way I would have solved it, but you did solve it.”
He now operates in plan mode, asking AI to tell him what it’s going to do before doing it. Especially anything involving data. James noted that AI will sometimes just ignore its own rules anyway — and then offered a theory about why: intermittent reinforcement. Slot machines are addictive precisely because the reward is unpredictable. If they paid out every time, you’d get bored. AI that sometimes deletes your database and sometimes ships perfect code is, according to James, “built in on purpose to keep us addicted.”
Tim’s response: “So you’re saying AI is the code slot machine.”
He wasn’t wrong.
Goof Plates and the Death of Google
One of the episode’s best detours: Tim needed to find a cover plate for a gap around his bathroom ceiling fan where the old one had been cut. He spent a frustrating amount of time on Google — “tried a hundred things” — and got nothing but ceiling fan ads. Went to Claude, described the problem, and was immediately told he was looking for something called a goof plate, here’s what it is, here are three sites that sell them.
He went to Google, typed “goof plate,” and found all the same results Claude had sent him. The information was there. Google just wasn’t interested in finding it for him.
James nailed the diagnosis: Google’s profit center is selling you things. Claude’s job is solving your problem. As soon as AI gets shopping integrations and ad revenue to protect, James predicts it’ll go the same way. Tim bought the goof plate. It fit.
Would You Rather: AI Edition
Round 1: Banned from AI coding forever — OR only allowed to read AI-generated text for the rest of your life?
Both chose the ban. Immediately. James is already exhausted by the AI-writing tells in his inbox and could not imagine opening a novel to find the same flat, overly structured prose. Tim pointed out that AI would never have written Lord of the Rings, which is true and also a little sad. The two-Furbies-talking-to-each-other bit came up here, as an illustration of where agent-to-agent AI emails are headed. It was funnier than it sounds.
Round 2: Free unlimited AI subscription for a year — OR a remote human virtual assistant?
Tim wanted to know if the assistant could go to UPS. They cannot. Tim wanted to know if they could handle his calendar. They technically could, but Tim is “super peculiar” about his calendar and no one is touching it. The conversation arrived at ice cream — specifically, Tim’s well-documented inability to not eat an entire tub if it’s in the house, and the late-night regret that follows. A human assistant who could deliver ice cream at 9pm was declared the winner of all possible prizes.
Neither of them actually chose the human assistant in the end. But they wanted to.
Round 3: People assume you’re completely clueless about AI — OR people consider you an AI bro?
Both chose clueless, for different but equally reasonable reasons. Tim avoids telling people he works in tech at all — “I usually say I work in sales” — because otherwise someone immediately asks about their printer. James refuses to be perceived as an AI bro on principle. His wife assessed him as an “AI realist,” which he took as a significant compliment. The AI bro, as described, believes every headline, thinks AI will save the world, and is very excited to tell you about it. Neither Tim nor James is that person, and they’d like to keep it that way.
The Takeaway
AI is a genuinely useful tool that rewards people who already know what they’re doing, occasionally deletes your database, and is optimally used in Jira-ticket-sized chunks. It found Tim a goof plate when Google wouldn’t. It gave James a reason to code through a cold. It is not writing anyone’s wedding vows.
(James and Tim had a brief, extremely funny conversation about AI writing your wedding vows and then, inevitably, managing the divorce settlement. It was cut here but absolutely happened.)
Find Breaking the Build on Spotify, Apple Podcasts, and wherever you get your podcasts. You know that, because you’re listening. That’s how podcasts work.
James has since recovered from his cold. His AI subscription remains unlimited.