WorkBuddy interface tour: six concepts distinguished at once

A friend used it for over a month and one day asked me: "Does it not have an automatic execution feature?"

I said, have you seen that "Automation" in the interface? He froze for a moment—the thing he did manually every day could have been set to run by itself at 8 a.m.

This is not an isolated case. In the interfaces of these tools, several names all look like "features", but they actually govern completely different layers. The consequences of not distinguishing them are concrete:You did work that could be automated manually for a month.

This article will explain the six concepts at once, then give a decision mnemonic, and finally cover the three most confusing points.

First remember one sentence: these are three layers, not six side-by-side buttons

Don't memorize these six words as a row of features; they are actually three layers:

• The top layer: conversation.You hand over a task, it completes it, and it's over. One-time.
• The middle layer: workspace.A folder containing all the materials for this task and the result you want. Reusable.
• The bottom layer: automation.Without you saying a word, it starts by itself when the time comes. Resident.

Skills, connectors, experts, and memory are—components attached to these three layers—skills are "what it can do", connectors are "what it can reach", experts are "in whose capacity it works", memory is "what it remembers about you".

Once you think of it this way, it clicks:Whether you can use it depends on how you use the three layers above; whether you use it well depends on whether the four components below are configured correctly.

Six concepts, explained one by one

1. Conversation: the smallest unit of work

It is the window where you and it talk back and forth. Within one conversation, it remembers what was said before, so you can say “do this, then do that” and “revise the version from just now.”

Key judgment:One task, one conversation.Mixing “organizing reimbursements,” “writing weekly reports,” and “debugging code” in one conversation causes the context to interfere with each other—it may apply the reimbursement format to the weekly report. Start a new one when done.

2. Workspace: its activity boundary

A folder. It can read what you put in, and what it produces is also placed here.

This isa security setting, not an organizing habit. The larger the scope you give, the greater its capability: if you give it only one project folder, it can at most act on this project; if you give it the entire disk, it can theoretically touch all your files.

My usage is to separate by task:D:AI报销、D:AI资料、D:AI合同Clear it or archive it after finishing a project.

3. Skills: things it already knows how to do

A skill is a “method package” written by someone else (or officially): when encountering a certain type of task, it handles it according to a fixed process and standard. For example, making PPTs, creating spreadsheets, formatting documents, and reviewing a contract all have corresponding skills.

The benefit is direct:You don’t need to write hundreds of characters of prompt, nor teach it industry rules, and it will follow that method.

The thing beginners most easily overlook is this part—describing requirements from scratch yourself; spending ages writing still isn’t as accurate as invoking a skill.

4. Connectors: which external systems it can reach

Conversations and workspaces address the “local,” while connectors address the “outside”: code repositories, email, cloud documents, knowledge bases, and various third-party services (the MCP kind).

Authorization is required before use; the first time it will ask you to log in or grant permissions. The attitude toward this should be:Authorize only what you need to use; don't connect what you don't need.The more you connect, the larger the surface for problems, and you yourself can't remember what you've granted.

5. Automations: the kind that runs without you asking.

This layer turns “must do every day” work into “runs on its own when the time comes.” Set it once, then it starts on its own according to time (or conditions), and gives you a result when it finishes.

There is only one criterion for deciding what work should be set up as automation:The steps for this task are the same every time.If it is the same every day and must be done, it should be automated; if it requires on-the-spot judgment every time, don't set it up—if you do, it will be reworked every day.

There is another pitfall that must be mentioned upfront:Automation failures are silent.If it didn't run, ran empty, or fetched an empty page, you may not know. So every automation must add a line: “report even if there is no result”; otherwise you think it is on duty, but it stopped long ago.

6. Experts: work from a different role.

Experts can be understood as “a packaged role”: not just capabilities, but the role's perspective on problems, order of doing things, and delivery standards. Give the same question to an “operations expert” and a “legal expert,” and the conclusions will differ—this is a good thing; it shows it is thinking according to the role.

When to use?When you need “professional judgment,” not just “getting the work done.”Organizing data doesn't need an expert; reviewing the pitfalls in a contract does.

There is also: memory.

Memory solves the problem of “having to explain it repeatedly every time.” Your industry, your writing taste, the words you hate, your fixed output format—say it once, it records it, and it persists across conversations.

Its value grows with time in use. It's not obvious at first; after using it for a month you will clearly feel:The explanation cost is decreasing.

A table: which task should use which layer

Three things that are most easily confused.

First, "skills" and "experts".Skills are "how to do it"—a set of processes and standards; experts are "who does it"—the judgment of a role. Skills can be attached to experts, but the two are not the same thing. If you need processes, use skills; if you need judgment, use experts.

Second, "automation" and "reminders".A reminder tells you at the scheduled time, "you should do this now"; automation is: at the scheduled timeit does it for you.Many people think they have set up automation, but they have actually set an alarm.

Third, "connectors" and "uploading files directly".For things you use only occasionally, just drag the file into the workspace; there's no need to hook up a connector. The value of connectors lies in "fetching repeatedly and fetching in real time"—for example, checking new issues in a repository every day. Authorizing a bunch of permissions just for one use is not worth it.

Here's roughly how I use it in a day.

When I get to my desk in the morning, I first look at the few reports that automation ran last night: whether there were errors, whether there were data anomalies, and whether there is anything that needs me to make a decision.

In the morning, I handle the big items that require thinking: one chat, one workspace, throw all the materials in, have it produce a first draft, and I revise it.This is where it saves the most time—the savings are in "reading materials" and "building the framework"; revising a draft is much faster than writing from scratch.

In the afternoon, it's mostly fragmented tasks: check a piece of data, revise a piece of copy, convert a table into another format. This kind of work isn't worth setting up a workspace for; open a chat, ask, and leave.

In the evening, I basically don't look at it. If something urgent really comes up, automation will push what needs attention to me.

Once this rhythm is running smoothly, the biggest change is not "fast", it'sthat I no longer have seven or eight things hanging in my head at the same time.—leave the part that stays running to it.

Here’s a three-day plan

Day 1:Use only one conversation and one workspace to finish the smallest task at hand. The goal is not efficiency, butconfirming that it can deliver.。

Day 2:Go through the skills list, find the two or three most relevant to your daily work, and try each once with a real task. This step usually brings the surprise of “turns out I don’t need to write prompts myself.”

Day 3:Pick a routine task you do every day, set it as an automation, and add “report even if there is no result.” Look back after three days, and you’ll know whether this setup is really worth it to you.

If you haven’t installed it yet, first read the previous article "How to install WorkBuddy and get the first task running end-to-end", which covers how to create a workspace and how to write the first instruction. The whole series starts with "What WorkBuddy Is", and it is recommended to read in order.