Who taught you to plan a feature?
\nI don’t mean writing one. I mean the boring part, deciding what done looks like before you write a line and how you’ll know it works.
\nI’ve been writing code since ’96 or ‘97 and I can’t name anyone who showed me. I learned it like most self-taught people learn anything, by shipping something half broken and then working backward from the mess.
\nIt left a gap, and I didn’t notice it for twenty years.
\nTrades used to be learned another way. You stood next to someone who already knew the work and watched. In 1989, Collins, Brown and Newman wrote about bringing that into schools and called it cognitive apprenticeship. Their point was simple. You can see the finished work, but you can’t see the decisions behind it. The expert has to work out loud, watch you try, and slowly do less of the helping.
\nSelf-taught developers got docs, tutorials and a search box. When those ran out, we asked strangers on Stack Overflow, and that went about how you’d expect. I’ve had questions closed as duplicates of questions that weren’t the same. I’ve been told I shouldn’t do it that way, by people who never said what I should do instead.
\nA lot of people stopped asking. By early 2025 the site was getting fewer new questions a month than at any point since 2009. The slide started around 2020, and ChatGPT sped it up. Amy Edmondson studied 51 work teams and found that people learn more when it feels safe to take interpersonal risks, like admitting a mistake or saying you’re lost. Ask yourself whether a comment thread ever felt like that.
\nI use AI almost every day now (mostly an agent working in my terminal). The code it writes matters less to me than the habits that rubbed off. Before it touches a file, it tells me what it’s about to do.
\nI asked for a short-link service for my domain and got a plan first, with how we’d know it worked, what it was allowed to spend, and what could go wrong with the DNS. I’d heard “acceptance criteria” for years. I’d never watched anyone use them on my own project.
\nAnother time, a review caught a bug on my own site. A page title containing a closing </script> tag could break out of the structured-data block in the page header and inject a redirect. The fix was one line. Before it called the bug fixed, it wrote a test that publishes a note with that exact title and checks that no redirect shows up. I’d written tests for years. I’d never thought of one as a record of the exact attack it keeps out.
Same with idempotency. I knew the word, but I didn’t respect it until a review found that my content scripts would quietly re-hide posts I’d republished, if anyone ever re-ran them. Teams with senior people do this by default. I never had one, and at 2 a.m. there was nobody to ask why a fifth time without getting the eye roll.
\nBut it screws up, and it sounds just as sure when it does.
\nMy site has a chart drawn like a hypnodensity. That’s the graph neural-network sleep scoring produces, showing the odds of each sleep stage at every point in the night. The AI wrote a caption calling it “how a sleep study is scored now.” That’s wrong. Clinical scoring still gives each 30-second epoch one stage and draws a plain hypnogram. I’ve run more than 15,000 sleep studies and I read right past it, because it sounded right and I was reading like a student. A review of the site’s content caught it, not me.
\nThere’s a name for that. A 2010 review of the research on automation bias found that experts fall for it as well as beginners, and that practice alone doesn’t cure it. Aviation and medicine train people against it. In twenty-some years of software, I’ve never heard it mentioned.
\nIf you want the mentor and not just the answers, ask different questions. Ask for the plan before the code. Then make it tell you what would sink the plan. When it fixes something, ask how you’d know, and get the test. If it uses a pattern you don’t recognize, ask what it’s called, then go read about it on your own.
\nWherever you’re the expert, check its work the way you’d check a new tech’s first overnight study. Line by line.
\nI spent years guessing, because asking was expensive. It isn’t anymore. I’m learning faster now than I did in my twenties. What would you go learn if asking didn’t cost you anything?
\n