I used to do this too: writing a quote while editing a client page

At 9 p.m. last Tuesday, I was rushing a client quote in a Hangzhou café while an AI tool edited a website in another window. I bounced between all three until the quote went out late and the client page was only half finished. I got stuck here too: the problem wasn’t that AI couldn’t do the work. It was that I handed it the entire project at once, then didn’t know which baton to take next.

What this approach is

Gemini 4 Hydrogen uses a main task to break down and coordinate the work, then passes subtasks such as writing, research, and code changes to separate execution steps. Xiaolin, who runs an independent store in Shenzhen, handles product copy, customer replies, and page updates every day. She turns each job into a checklist, lets the orchestration layer decide the order, then executes one task at a time. Her rework has dropped noticeably. It works more like a project assistant than a one-click solution that handles everything for you.

The practical costs today

Money: Start with Google’s Gemini access within the included allowance. The free tier is usually enough for an initial trial, so there’s no need to buy a plan first. Time: About one hour to organize the workflow. Technical barrier: No server setup or coding required, but I do need to learn how to frame tasks with a goal, inputs, and a desired output. First step: Open the Gemini 4 Hydrogen page in Google Antigravity, create an orchestration project, and add the most urgent small task on today’s list. Don’t move the whole company into it.

How I’d use it at each stage

If I’m just starting out, I’d test it with one real task, such as turning client requirements into a quote outline. It’s fine not to try it right away; doing the work manually may be steadier. With one or two clients already, I’d turn recurring replies, quotes, and research cleanup into fixed workflows, then let AI handle them in sequence. While scaling, I’d add checklists and human approval points to every workflow so errors don’t get replicated in bulk. Not everyone needs this tool. I’d first see whether it reduces my context-switching cost, then decide whether it’s worth investing further.