◇ Could this help me?
Yes for the argument, with two parts to leave behind. The central claim is correct and useful: tools have genuinely different strengths, a design tool feeding a coding agent works better than asking one product to do both, and his correction about customer support is the sharpest thing in the video, since an assistant that only exists when you open it cannot answer mail overnight no matter how well its FAQs are organized. Do not copy two specifics: granting an agent full control of the machine it runs on widens the blast radius of every mistake it makes, and connecting an advertising account so an agent can run campaigns hands real spending authority to something with no confirmation step, which is a risk that belongs to a person. Also note the harness he recommends is third-party software whose publisher and permissions are worth checking before it runs anything. General practice.
I want to stop defaulting every task to one AI vendor and instead pick tools by what each is actually best at. Given the work I describe below, break it into jobs, for example interface design, code generation, long-running agent execution, scheduled reporting, and customer communication, and for each one tell me what capability the job genuinely requires, such as whether it needs something running continuously rather than only when I open an app. Then recommend options per job, name the trade-offs, and be explicit about which jobs should keep a human approval step, particularly anything that spends money or controls my machine.