I think you're very wrong about this timeline.
1. I've never sourced hardware, so I can't comment directly on manufacturers. I have done plenty of business in Asia, and, I have purchased from alibaba. One side of those alibaba transactions is definitely using AI translation already. I don't think manufacturers in Asia will care at all "who" is sourcing product from them. For speccing a car - sure, you're right that a relationship is needed (and probably will involve doing some drinking near the factory). For speccing injection mold / printable goods? I'm skeptical that this is cut and dried in favor of humans with IRL relationships.
2. Trivially done by an AI; "Mr. Bossman do you want to validate this sample, or have me hire someone to do it?"
3. This is demonstrably untrue; OpenAI disagrees with you about this, for instance. ChatGPT pre-instructions include multiple paragraphs about this topic already. (here's a link to info about it if you're curious: https://x.com/krishnanrohit/status/1755122786014724125/photo...)
Generative AI that can spell correctly is here now, although not widely commercially deployed, but is in no way needed to AI CEO things.
My experience as the 'gluer' is that the most recent visual/multimodel models are good enough to, say, fix a CSS bug in HTML they wrote by looking at a screenshot of the output. So, I'd say the opposite is true in my experience -- with multimodal feedback, they are able to self correct.
Anyway, to me this doesn't add up to 2030. Maybe for the snazzy mac app version. But, I think we are VERY close right now, and it is a major engineering effort at OpenAI and Microsoft to get there quickly.