Coding Is Over. Get over It
Writing in a Hacker News post, a software engineer at JPMorgan Chase with more than 15 years of software-building experience says he used to love the work but now sees AI as pervasive company-wide. He notes these are his own observations, not JPMC's position, and that his vantage point at a very large company may not generalize. AI adoption and ways of working with it vary widely as everyone learns, and while the bank has sane guardrails, he is fortunate to be somewhere that lets him experiment a lot. Costs are high, but the firm can afford them; letting people learn the tools first and worrying about costs later may be a good investment, though it is hard to know now whether it will pay dividends.
He argues that high costs create pressure to prove worth, yet he does not know how ROI could even be measured. It is hard to say whether engineering teams generate more value than they spend on AI; the same work probably would have been cheaper without it, maybe slower, and it is unclear whether a cheaper model would have done just as well—to his knowledge nobody has run that experiment. Measuring an individual engineer's performance has always been difficult, and AI has only added a new variable. He dismisses confident claims like 'we built this with AI 50% more efficiently' as BS unless backed by rigorous evidence.
His current view is that AI is a major enabler: it has let him do things he genuinely could not have done without it. Although he has extensive app-building experience, infrastructure—especially Terraform—was a new field for him, and working with AI agents gave him more confidence in their output and in their assessment of plan diffs than he would have had alone with limited expertise. Whether that was worth the cost, he still does not know and may never know.
A trend he says is worth tracking is cheap, small models, which are getting genuinely good. GPT 5.6 Luna with max reasoning was a big surprise to him, and he predicts inference cost will soon become close to irrelevant for many use cases—a rounding error rather than a cost item. If that happens, he thinks the bigger revolution will not be in coding at all but everywhere else: all that can be automated simply by asking your computer.
He cautions that anyone who tells you with confidence what the future looks like is guessing, including himself; nobody actually knows. He also stresses that the paradigm shift is hard to overstate: model capabilities evolve so fast that what was true last month may not be true today, making it hard to stay up to date even when working with AI every day. The excerpt ends mid-sentence referring to a recent statement from 'the head of Cl...'