Log inStart Free Trial
Meta Tech Provider

Change a WhatsApp automation by asking your AI

You can change a WhatsApp automation by asking your AI assistant through the easyCoexistence MCP connector: it reads the automation with list_automations, sends the change with update_automation and checks it with test_automation. Fields left out keep their value, but a trigger, steps list or settings object that is sent replaces the old one whole, so the AI must start from the current version.

Whole list

update_automation replaces the steps list whole when steps are sent; fields left out keep their value.

4 and 10

Defaults for the pause hours and AI replies per chat per hour that a partial settings object falls back to.

60 steps

Most steps per automation, branches included (30 per line, nested 4 deep); Stop only as the last step of its line.

30 days

Run history kept per automation; editing keeps it, deleting the automation removes it.

Superseded

Moving an automation to another number with connection_id ends its waiting runs at once, skipped as superseded.

What happens when you ask your AI to change an automation?

Three tool calls, plus a fourth later. The AI calls list_automations, optionally with the number's connection_id, and gets each automation's trigger, steps, settings, status and whether it is on. It builds the new version and sends it with update_automation and the automation_id. It then runs test_automation with a sample message, which runs the saved steps for real, AI and webhook included, without sending anything to anyone. After customers have written, list_automation_runs shows whether the new version answered and why a run was skipped.

1list_automationsRead the current trigger, steps and settings
2update_automationSend the changed version
3test_automationRun it on a sample message, nothing sent
4list_automation_runsCheck real runs after customers write

Which fields does update_automation replace whole?

The merge happens field by field at the top level only. A field left out of the call keeps its stored value. A field that is sent replaces the stored one entirely, with no merging inside it, so changing one keyword means sending the whole trigger and changing one step means sending every step. The dashboard editor works the same way, since it saves the full automation each time:

If left outIf sent
nameKeptReplaced
enabledKeptReplaced: true is on, false is off
triggerKeptReplaced whole: type, match, keywords, mode and hours together
stepsKeptReplaced whole: the new list is the full automation
settingsKeptReplaced whole: a missing field returns to its default
connection_idStays on its numberMoves it to that number

Why must your AI read the automation before changing it?

Because a partial list or object silently removes what it leaves out. Asked to change the text of step 2 in a three-step automation, an AI that sends only the new step 2 leaves an automation with one step. Asked to add a keyword, an AI that sends {"keywords": ["menu"]} alone sends a trigger without type and match, which is refused. Asked to set the pause to 8 hours, an AI that sends {"takeoverHours": 8} alone also resets AI replies per chat per hour to 10, whatever it was.

The safe pattern is to read, edit the copy, and send the full field back. Saying "read the automation first and send the full steps list back" in your request makes most assistants follow it, and asking for a before and after summary lets you spot a dropped step before testing.

What does a correct change look like?

Here the owner asked to change the price list text in the second step of a keyword automation that answers "price". The AI read the automation, kept step 1 and the Stop as they were, and sent all three steps back with only the text of step 2 changed. Name, enabled, trigger and settings were left out, so they keep their values.

update_automation sending the full steps list (example)
{
  "automation_id": "<from list_automations>",
  "steps": [
    { "type": "reply_text", "text": "Hi {{first_name}}! Here is our price list:" },
    { "type": "reply_text", "text": "Sourdough $7 · Croissant $3.50 · Cakes from $38. Delivery $4 within 5 km." },
    { "type": "stop" }
  ]
}

What happens to runs already under way?

A change applies to the next message that triggers the automation, and runs paused at a Wait step pick it up too. When a waiting run resumes, it continues with the automation's current steps from the step number where it stopped. If the edit inserted or removed steps before that point, the resumed run lands on a different step than planned, so edit a follow-up automation when nothing is waiting, or accept that waiting runs follow the new list.

Two changes stop waiting runs instead. Setting enabled to false makes them end as skipped with "Automation is off" when they wake. Moving the automation to another number with connection_id ends them at once as superseded.

Which requests work well?

Requests that name the automation and describe the result in your own words, in any language your assistant understands. The AI translates them into the fields, reads the current version first and sends the full field back; you check the before and after summary and the test reply. A request that touches only enabled is the one change that can be sent alone:

"Add the keywords cardápio and menú"Full trigger sent back with the new keywords added
"Close at 20:00 on Saturdays"Full trigger sent back with the new hours
"Make the AI answers shorter"Full steps sent back with a length rule in the prompt
"Wait 2 hours instead of 1"Full steps sent back, wait minutes changed to 120
"Stay quiet 8 hours after I reply"Full settings sent back with takeoverHours 8
"Turn off the away message"Only enabled false sent, everything else kept

What can your AI not change this way?

AI keys, which are entered only in the dashboard under AI keys: no tool accepts or returns one, and a change that adds a Reply with AI step for a provider without a saved key is refused with "save your ... API key on the Automations page first". The content of an approved template, which belongs to Meta's review: ask for a new template with create_template and point the step at it once approved. The number's message routing, which is set_inbound_mode, a separate tool. Other refusals come back as plain messages the AI can fix and resend, such as "Stop must be the last step", "Add at least one keyword" or an unknown timezone.

How to set it up

  1. Name the automationTell your AI which automation and which number, or ask it to list your automations first.
  2. Describe the result you wantSay what should change and ask it to read the automation first and send full fields back.
  3. Review the before and afterAsk for a short summary of the old and new trigger, steps and settings, and check nothing was dropped.
  4. Test the new versionHave it run test_automation with a message that should trigger the change, and read the reply.
  5. Check the first real runsAfter customers write, ask it to read list_automation_runs and explain any skipped or failed run.

Common mistakes

  • Sending only the changed setting. A settings object with takeoverHours alone resets AI replies per chat per hour to its default of 10.
  • Adding or removing steps while follow-ups are waiting. Waiting runs resume at their old step number in the new list and can run the wrong step.
  • Letting the AI rewrite the whole Instructions to change one fact. Ask it to change only that line, so tested wording elsewhere stays the same.
  • Editing the same automation in the dashboard and from your AI at once. Both save the full automation, so the last save wins and the other change is lost.
Doing this with EasyCoexistence

easyCoexistence lets your AI change your WhatsApp automations by name, through one MCP connector with more than 30 tools: it reads the automation, sends the full new version, tests it without sending anything and reads the runs afterwards. The number stays in the WhatsApp Business app throughout. US$ 9 down to US$ 0.75 per number per month, falling as you add numbers, with the first 3 days free.

easyCoexistence is a multichannel API and MCP server that connects AI agents such as Claude and ChatGPT to WhatsApp, Instagram, TikTok, Facebook and more. On WhatsApp it uses Coexistence: the business keeps using its number in the WhatsApp Business app while the AI works through Meta's official API.

Frequently asked questions

Can I undo a change?

Yes, if you keep the earlier version. Ask your AI to show the current automation before changing it and save that answer; to undo, it sends that version back with update_automation. There is no automatic version history.

Will ChatGPT or Claude ask before changing an automation?

ChatGPT may ask you to confirm tool calls that write, such as update_automation. In Claude you can set each tool's permission under Customize > Connectors, including Blocked for tools you never want it to call.

Does an edit fix an automation paused for a refused key?

Saving a working key under AI keys restarts it on its own. An edit that removes the paused provider's AI step also returns the automation to active.

Does editing cost anything?

No. Edits and automations are included in the per-number price. A test_automation call with a Reply with AI step spends tokens on your own AI key, billed by your provider.

Keep reading

Ready to get started?

Set up WhatsApp Coexistence in minutes, not months. The app keeps working on the phone.

Start Free Trial

Verified on