We confront the uncomfortable truth that the systems we built to optimize efficiency are now teaching AI to inherit and amplify the web's darkest deceptive patterns. From Arin Bhowmick's warning that prompting for conversions increases manipulation to Eleanor Hecks' critique of session timeouts that punish human pacing, we explore how our past shortcuts are becoming our future liabilities. We also untangle the community's growing anxiety over homogenized design and leadership traps where AI is used to bypass human judgment rather than assist it. Ultimately, we ask whether we can untrain these models before they make our work feel entirely disposable.
