Last Monday, I Was Chased by Client Requests Again

At 3 p.m. Monday, I was replying to client emails while copying requests into our project tool and checking whether the designer and developer had seen them. I used to think this meant we were busy. Later, I realized the workflow simply had not grown any “hands.” If I am asking in a chat every day, “Who is handling this?” or “Where are we now?”, we may be dealing with the same problem.

Linear’s AI Agents Do More Than Write a Few Paragraphs

Linear started as a project tool for managing team tasks and bugs. It now lets AI agents participate in the work: agents can create tasks, organize information, and help move workflows forward. Co-founder Karri Saarinen had worked at companies including Airbnb and Uber; Linear later spread organically among startups such as Vercel and Ramp. What interests me most is not “AI can write code,” but agents taking on some actual work. Cristina Cordova said that a year ago agents created about 3% of the work; now that figure is over half.

But the team still draws one line: AI does not handle customer support requests on its own. That warning matters. It is good for saving us time spent on internal organization and assignment, but it is not meant to carry our promises.

Trying It Today Is Not Cheap

I got this wrong before: I saw “more than half” and assumed I should buy a premium account first, then connect a pile of APIs. Instead, I found a workspace in Linear with AI enabled, checked whether the current plan supported agents, and chose one internal task for a trial run. The first step is to open the AI or agent settings; if the option is missing from the plan or interface, I would not force it.

For the budget, I would try spending RMB 0–100 per month for one month before buying anything more expensive. For time, I would reserve one to two hours each week for the first two weeks to see whether it reduces repetitive work. The technical barrier is understanding task flow—who owns what, and which status counts as done. Programming skills are unnecessary, and there is no need to configure API keys. When actually replicating this, the core is not connecting more tools, but defining the boundaries clearly: what an agent can handle, and what still requires human confirmation.

Choose Based on the Stage You Are At

Just starting out: I would run the process manually for a week and note the most time-consuming repetitive actions. Once the pattern is stable, I would let an agent only organize tasks, without letting it send messages automatically.

Already serving one or two clients: I would add a rule: anything involving delivery dates, refunds, or client commitments goes to a human for review. Once a month, I would review how much time it actually saved.

Scaling up: I would break common requests into fixed workflows, such as “collect feedback—create task—assign—follow up,” then grant permissions gradually while keeping human spot checks.

Not every team needs this tool; if I only have a few recurring jobs, a spreadsheet can work just as well. There is no urgency to try it now. Defining the workflow clearly often matters more than bringing AI into it.