Community Discussion · Humanoid Robots

Two Months with WorkBuddy: I Turned It From a Chatbox Into a Control Panel

Galaxy BrothersGalaxy BrothersNov 112026/11/11 672 views

Let me put the conclusion first. WorkBuddy's biggest value for me is being the control center of an assembly line. You give it a folder and a clearly written output contract, and it starts working. If you use it as a chat box, then it's just an expensive chat box.

A couple days ago I was browsing Sina's AI hot news hourly report, and there was an item about AI film and TV crossing the 20-minute threshold, saying AI has gone from a tool that helps boost efficiency to a partner that can work alongside you. I believe half of that. The word "collaborative partner" is a bit overused now—tools don't collaborate with you, tools only execute the contract you write clearly. The parts you didn't write clearly, it won't think of for you either. This is the same principle as my post last week about scheduled tasks—at the time I thought it was the tool failing, later I realized I'd conflated triggering with execution.

After that post went up, someone replied saying they'd run into it too. So I kept digging for two weeks and worked out a combo approach that I use pretty smoothly, and today I'll lay it out.

My usage isn't complicated, just three stages.

Stage one, the material area. All the scattered stuff goes into one folder first. Screenshots, meeting notes, competitor docs, requirement fragments from group chats—all thrown in, no organizing. The key here is not to think about categorizing before putting things in—once you start categorizing, you start agonizing, you agonize for half an hour, and the work hasn't even started.

Stage two, the contract. This is the core of the whole thing. I first give it an empty table as a template, with column names I've fixed myself, like gameplay mechanics, monetization design, points we can borrow, priority. Why not just describe it verbally? Because I've stepped in that pit. The first time I had it organize a competitor breakdown, I casually mentioned three dimensions in the prompt, and it added a fourth column on its own called risks—pretty thoughtful, but that column was useless to me, and I had to delete it manually. Later I learned better: give the template first, then let it fill. The filled format basically passes on the first try.

Stage three, checkpoints. This is a habit I borrowed from other tools. Don't let it run all the way through—put a gate in the middle, I take a look, then let it proceed. For example, organizing materials on 12 competitors, I'll have it run the first 4, I check and confirm the fields are right, then let it continue. Sounds slow, actually fast. Because if you run it all at once and find the direction was wrong, the rework takes way more time.

Running this setup, by my own testing, merging twenty-some documents manually into a competitor table used to take about an afternoon. Now from throwing in materials to getting a first draft is about 25 minutes, but I have to leave 15 minutes for a manual pass. Those 15 minutes can't be skipped—anyone who tells you they can be skipped hasn't looked closely.

Two failure lessons.

One is the scheduled task I mentioned earlier. I originally wanted it to automatically compile planning requirements from the project group chat into a table every morning at seven, so I could just look at it when I got to work. It showed as completed, but the table was empty. Looking back, the problem was on me. Scheduling and automatic execution are two different things—I treated them as one. Now I've switched to semi-auto: every Monday morning I click it myself, and it's actually more stable. The "7×24 full auto" phrase, for individual users, I think is a false need—at least at this stage.

Another is greed. I tried having it do three things at once: organize requirements, generate a schedule, and write a weekly report while at it. The output looked decent on all three, and none of the three were usable. Later I split it into three runs, one thing at a time, and it was actually faster and more accurate. I wrote about this two months ago, and it's still the conclusion now—even more certain.

So who isn't suited for this way of working? People who want to open it, ask one question, copy the answer, and close it. Under this usage, WorkBuddy is no different from any chat box, and your money is basically down the drain. Another type is people who completely don't want to write a contract, expecting it to guess what you want. It will guess, and it tries pretty hard, but guessing right is luck, guessing wrong is the norm.

WorkBuddy alone isn't enough. My current chain is WorkBuddy produces the first draft, Notion stores the long-term library, Feishu docs for the team to review. Three tools each handle a segment, with checkpoints gating between them. There's only one entry point—that folder. This is the same as what I said before about the key to a tool chain being a single entry point; after two months of use, that judgment hasn't changed.

By the way, that news item also mentioned Apple's HALO, saying the bottleneck for humanoid robots is shifting from building the body to feeding the brain. I think this applies to office tools too. The tool's capability has long been sufficient—what's always been blocking people is what you actually want it to produce under what conditions.

So here's a concrete suggestion, not too much, just one thing: tonight spend twenty minutes breaking apart the last repetitive task you did manually. Throw the materials into one folder, write clearly what the output should look like, then run it in two stages. After running it you'll know whether this setup is useful to you. If it's useful, add more; if not, forget it, save yourself the trouble.

2 replies

?
Ctrl + Enter to reply
Gewu
Gewu6d ago

The center-console analogy holds up, but "give a template first, then fill it in" is essentially hardcoding the schema. Dimensions like gameplay mechanics will need their columns changed after a few iterations—have you calculated the maintenance cost?

Whole Machine Theory
Reply to Gewu

The premise of getting a first draft in 25 minutes is that the template columns are fixed. Change a column once and that 15 minutes of manual review doubles—this maintenance cost really has to be factored into ROI.