The First Time I Put a Client Brief Into AI, I Realized I’d Been Repeating Myself

At 4 p.m. on a Tuesday, in a shared office in Shanghai, I was staring at the six-page requirements document Chen Yu had sent ahead of our quote meeting the next day. What did the client actually want? Which items were mandatory? Which were just offhand ideas? I had to tag every line myself. This was always the job I dreaded most: plenty of information, but very little I could act on.

I got stuck too. The first time I asked AI to summarize the brief, it turned the client’s “wants” into “must do immediately.” I only noticed before quoting that our understanding had already diverged. Later, I understood where AI actually fit: it could handle the organization and produce a first draft; I still had to make the final calls.

What Is This, and Who’s Using It?

METR observed Anthropic’s internal work for 11 weeks and estimated that its R&D pace ran 1.3 to 3 times faster than before, with roughly 80% of the improvement linked to AI involvement. I don’t read this as “someone is about to do our jobs for us.” It feels more like a reminder: the value of a tool often isn’t negotiating with clients; it’s taking repetitive information processing off your plate.

My friend Lin Shan, an independent consultant in Hangzhou, now puts client interview notes, emails, and product explanations into a fixed AI workspace. She asks AI to extract goals, audiences, budget signals, and open questions, then checks the output herself. She doesn’t turn the whole pile into a final proposal in one click. First, she looks at “where the evidence is missing,” then goes back to the client. This setup works well for freelancers and teams of one to five people: I still own the relationships and judgment calls; AI just turns conversational notes into a checklist.

What It Costs to Replicate

Money: The free allowance on a common AI chat tool may be enough; I’d budget 0 to 100 yuan per month. Time: Building the first workflow takes about 30 to 60 minutes; after that, each project should take roughly 10 to 20 minutes. Technical barrier: No coding required. Copying text and separating public information from sensitive material is enough. First step: Open the AI tool I already use, start a new conversation, paste one anonymized set of requirements I have permission to process, and enter: “Only organize; don’t make decisions for me. List the goals, must-haves, questions, and missing information.” If I don’t have a tool yet, I can try a familiar general-purpose AI service. There’s no urgency to test it today; keeping one workflow I can understand is more useful for now.

How I’d Use It at Each Stage

If I’m just getting started, I’d practice with my own product description: ask AI to rewrite it in three ways customers can understand, then check closely whether it invented any facts.

If I already have one or two clients, I’d bring it into the requirements-confirmation step. After each meeting, I’d have AI generate a one-page “confirmed, pending confirmation, next step” checklist, then verify every item before sending it.

If I’m scaling up, I’d standardize the team template: organize everything under client, industry, goal, evidence, and next action, then spot-check for errors regularly. The larger the operation, the more I need to write review responsibility into the process instead of assuming AI is always right.