Half a year ago a woman who does content work asked me whether her team could batch-publish on Xiaohongshu. Nobody there could write code, and she wanted to know if that closed the path.
I asked what she actually does to publish one note. Open the app, pick the photos, write the copy, add topics, tap publish. Two or three minutes each, and a dozen or more a day.
I told her those actions do not need code. Say them once. She tried it that afternoon on one phone. Three days later she told me the flow was already saved as a template, and one click before she leaves work runs it through.
Some jobs cannot be done this way. The boundary comes first. How to start comes after.
What you can do, and what you cannot
Work you can hand off without writing code has its actions on the screen. The judgement comes from looking.
| Task type | Without code? | Notes |
|---|---|---|
| Repeating a fixed flow | Yes | Open an app, tap a button, type text, upload images, submit |
| Deciding the next step from the screen | Yes | Close a dialog if it appears; treat “done” on screen as success |
| Capturing evidence | Yes | Screenshot each step and keep it, then check by hand afterwards |
| Running one flow across many phones | Yes | Pick a group of devices and send the same flow |
| Pulling API data into a form | No | Needs code to call the API |
| Branching on a return value | No | Complex conditions need code |
| Integrating with CI | No | That is engineering integration |
| Writing results into a database | No | Needs data-handling logic |
Repeating the same actions every day on a few platforms sits in the yes column. Work that processes data, makes decisions, or connects to other systems still needs someone who writes code. Forcing a no-code setup onto that work only makes it harder.
Three modes on the conversation page
All three sit at the bottom of the conversation page: consult, run, and refine. They differ in whether the phone is controlled, and whether the request gets talked through first.
Consult is for asking how to write a step, how a feature works, or how to fix an error. You do not have to pick a device, and the phone stays untouched.
Run fits when the steps are already clear. Pick the phone and write one line, such as “open Settings and wait 3 seconds”. It plans immediately and operates the real phone. Simple requests take the least effort in this mode.
Refine fits when you only have a rough idea. Describe what you want in a loose way. It asks back what the screen looks like, how long to wait after a tap, and what to do if a dialog appears. After a few rounds it turns the request into steps you can read. You confirm, then switch to run.
The first time you do something, use refine. You do not know that screen well yet, so one description usually misses a detail, the run stalls on the third step, and you cannot tell whether the description or the action was the problem. Talking it through before the run raises the chance it finishes.
The description from that chat can be saved on the prompts page. The next time you do a similar task, pull it out and use it. You do not have to talk it through again.
Once it runs cleanly, save a template
Conversation is for trying a flow. Workflows are for running it over time.
When a flow has been fine for several days in a row, save it as a workflow. Later you only pick the phones and the parameters.
These items on the canvas get used:
| Feature | What it is for |
|---|---|
| Validate | Checks that step connections and parameters are filled in, so a gap shows up before the run is halfway through |
| Arrange | Tidies the layout once there are many steps, so the canvas stays readable |
| AI workflow collaboration | It helps generate or edit steps |
| Trial run | Runs the flow once on one phone, then you use it for real |
Leave the parts that change as parameters. In a note-posting flow, the fixed parts are opening the app, picking photos, and tapping publish. The parts that change are the copy and the image numbers. Leave those as parameters and one template covers the batch.
Trial-run after the setup is done. The result is written to execution history, which holds up better than watching the screen.
A button that is only an icon, with no text, or a screen that is custom-drawn, cannot be found by reading words. The AI gets stuck.
Use image templates. Capture a small image of the button, save it in the library, and write three things. Keep the name short, such as “publish entry”. Put the visual features in the description, such as “orange circular plus button at the bottom of the home screen”. Put the timing in the usage, such as “use this when the publish entry is needed after reaching the home screen”.
On the conversation page you can choose a template, or an image template, and attach a library entry to the current session. Add a template for every icon in the flow that is easy to misread. Later runs stay steadier.
How to run the first task
Follow this order and you can see a result within half an hour.
First, install and confirm the phone is online. Follow install selection, install the workstation and the phone-side program, open the devices page, scan the phone in, and confirm at least one shows as online.
Second, pick something you were already going to do today. Keep it small. Writing one line in Notes is enough. This step checks whether the whole path works.
Third, describe it in plain language. Start a new conversation, select the phone, and say what you want done. Use refine the first time.
Fourth, send it and watch whether the phone moves. If it stays still, read the error in execution history first. Most cases come down to two causes: the device is offline, or the automation environment never started.
Fifth, once it runs cleanly, save it as a workflow. If it needs to run every day, add a schedule, and later you will not have to start it by hand. A schedule has a condition: the workstation on the computer has to be on, and the target phone has to be online at that moment.
The woman from the start later turned posting, comment replies, and screenshot archiving into three templates, and saves about two hours a day. Work that needs judgement, or that processes data, stays with the colleagues who write code. Keep those two kinds of work apart and the setup stays usable. Install it and try one small task, and you will see which side your own work falls on.
Frequently asked questions
- Can I really do phone automation with no coding at all?
- For tasks built on interface actions and visual judgement, yes. You describe what to open, where to tap, how long to wait and what counts as done, and it runs on a real phone. Publishing, liking, replying, filling forms and capturing evidence all fall in that range.
- What can I not do without code?
- Anything needing precise values or complex branching. Pulling data from an API into a form, deciding which path to take from a response, integrating with CI, or writing results into a database all still need code.
- Conversation mode or refinement mode, which should I use?
- It depends on how clear the requirement is. If the steps are already decided, conversation mode sends one instruction and it runs. If you only have a rough idea, refinement mode lets it ask questions first, filling in ordering and exception handling. For a first attempt, use refinement mode.
- Do I have to describe the same task every time?
- No. Once a flow runs cleanly, save it as a workflow template, and afterwards you just pick phones and fill parameters. Values that change each time, such as titles, prices and image numbers, become parameters, so one template covers a whole batch.
- What if a button is a pure icon with no text?
- Capture an image template for it: save a small screenshot of the button and describe what it is and when to use it. Once that entry exists, the AI selects it when it needs to find that button. This is the most effective fix for icon recognition.
- A run stalls halfway. How do I find where?
- Check execution history. Every task records its source, whether it is conversation, schedule or a workflow trial, plus execution status, target device and error message. Device offline points to the device page, automation environment points to the install page, and unauthorised points to workflow status.
- Can it run on a schedule?
- Yes. Scheduled tasks come in workflow schedules and conversation schedules, and fixed-step work suits workflow schedules. The workstation on your computer has to stay on, the target phone has to be online, and USB-connected devices need the connection intact at that moment.
- Do several phones running at once interfere with each other?
- No, because each device runs its own session. That said, on a first batch run it is worth staggering the start times slightly, not because of software conflicts but because it makes it obvious which device is which when something goes wrong.
Try it without code
Describe the task once and let the phone do it
Software is completely free and runs on your own computer. No coding needed.