Skip to content
Plymouth AI Lab
Five Agent Tricks logo: clay number 5 with a robot agentFive Agent Tricks · Newsletter

· Issue #4 · 5 tricks

Issue #4: 5 AI-agent tricks for surveys, subscriptions, product visuals, follow-ups and routing

5 new agent tricks worth trying this week. For business owners, operators and anyone running AI agents. This week: less time on survey analysis, cancelling unwanted subscriptions, product visuals, customer follow-ups and routing requests. Five minutes to read, each trick checked against its source.


1. Turn a pile of Google Forms responses into next steps with Gemini

What it does: Google is rolling out "actionable insights" in Google Forms (announced 5 Nov). From the Responses tab, Gemini analyzes every answer by writing and running Python code, so counts, percentages and comparisons come from your real data. It looks for patterns and links between questions, then summarizes the main takeaways and recommends next steps. Google's examples are event and product feedback, research surveys, and registrations or applications.

Who it's for: anyone who collects customer feedback, post-event surveys or sign-ups in Google Forms and never finds time to read them properly. Plans: Workspace Business Standard or Plus, Enterprise Standard or Plus, Education Plus, and consumer Google AI Pro or Ultra. Rapid Release domains started getting it on 2 Nov. Scheduled Release domains start on 19 Nov, and it can take up to 15 days to show up.

How:

  1. Check the form qualifies: at least 12 responses and 3 questions, quiz settings off, and access to any linked Sheet. On Workspace, your admin needs to have smart features and personalization turned on.
  2. Open the form, click Responses, find the Actionable Insights section and click Generate insights.
  3. Click Analysis steps to see how Gemini got its numbers. The output has an executive synthesis, correlations and recommended next steps. Treat it as a draft: compare one or two figures with the raw responses before you act on it, and click Retry for a fresh take.

Source: Google Workspace Updates, "Get actionable insights with Gemini in Google Forms" (5 Nov) https://workspaceupdates.googleblog.com/2026/11/get-actionable-insights-with-gemini-in-Google-Forms.html · Google Docs Editors Help https://support.google.com/docs/answer/16231981

2. Have Grok Bot find the subscriptions you don't use, and phone to cancel them

What it does: on 4 Nov, xAI announced that Grok Bot can make phone calls. In the launch demo (as described by TeslaNorth), the Bot spots an internet contract renewing in three days and offers to call and cancel. A "Phone call" card with Cancel and Call buttons appears before anything is dialled. During the call you see a live transcript with Take over and End call buttons. In the demo it turned down a 10% retention offer and posted a recap in the chat. Since v0.78.0 (27 Oct), the Bot also has subscription templates: "Find unused subscriptions" cancels the plans you pick, "Watch subscription prices" checks renewal prices, and "Try it, don't pay for it" starts a free trial and cancels it, or reminds you, before it bills. Per the changelog, the Bot confirms with you before changing anything.

Who it's for: owners who suspect they're paying for software seats, trials and contracts nobody uses, especially the ones that only cancel by phone. xAI's post doesn't say which plans, countries or phone numbers can use calls, or whether calls cost extra, so check in your own account.

How:

  1. Ask the Bot for a template that fits, e.g. "find subscriptions we pay for but don't use". Bots can find a matching Bot template and link it in the chat (since 11 Oct). Pick which plans to cancel from the list it gives you.
  2. For anything that only cancels by phone, ask the Bot to call. Give it your limits in plain words, for example "cancel; don't accept a discount" or "keep it only if they cut the price by at least 30%". Check the number on the Phone call card before you tap Call.
  3. Watch the live transcript and tap Take over if the call goes off script. Afterwards, read the recap, and add "Watch subscription prices" for the tools you keep so a renewal price rise doesn't slip through.

Source: Grok Bot on X (4 Nov) https://x.com/bot/status/2118043939592942052 · TeslaNorth's write-up of the demo https://teslanorth.com/2026/11/04/grok-bot-phone-calls-live/ · Grok Bot changelog (11 Oct and 27 Oct, v0.78.0) https://x.ai/changelog/bot

3. Save your product photo once in ChatGPT and reuse it in every image and GIF

What it does: Images 2.6 (4 Nov) makes ChatGPT image generation 50% faster and better at keeping characters, products and brand details consistent across edits, according to OpenAI. Two features turn that into a small marketing workflow. Saved reference images let you upload a photo, save it under a name and call it later with @name. GIFs (rolling out) let you ask for a looping animation or animate a still image you upload. OpenAI's own examples are animating a brand mascot and turning a product shot into a loop.

Who it's for: small businesses that post on social media and keep re-describing the same product, logo or mascot to the AI. It's available in ChatGPT on all subscription tiers, on web, iOS and Android.

How:

  1. Upload a clean photo of your product (or logo, or mascot) and ask ChatGPT to save it as a reference, or accept when it offers. Give it a short name, like "bottle".
  2. In any later chat, type @bottle in your prompt, for example "@bottle on a beach table at sunset, square format for Instagram". Ask for changes in plain words to refine it.
  3. For motion, ask for "a looping GIF of @bottle spinning slowly", or upload a still and ask ChatGPT to animate it. Open the GIF to change its speed, then save or share it. ChatGPT can't edit GIFs you upload, and one tester reports that an image ChatGPT generated has to be downloaded and re-uploaded before you can save it as a reference.

Source: OpenAI, "Introducing Images 2.6" (4 Nov) https://openai.com/index/introducing-images-2-6 · OpenAI Help, "Images in ChatGPT" https://help.openai.com/en/articles/11084440-images-in-chatgpt · ChatGPT release notes https://help.openai.com/en/articles/6825453-chatgpt-release-notes

4. Let an n8n agent follow up with customers on WhatsApp after the 24-hour window

What it does: WhatsApp only lets a business send free-form messages within 24 hours of the customer's last message. After that, only templates Meta has approved get through. Before n8n 2.47.0 (3 Nov), an n8n agent on the WhatsApp channel simply errored outside that window, so scheduled follow-ups never arrived. Now the agent has two new tools: list_templates reads your approved templates, and send_template sends one with the blanks filled in. If a normal message fails because the window is closed, the agent switches to a template that fits.

Who it's for: businesses whose customers live on WhatsApp, such as clinics, shops and service firms, and who want order confirmations, appointment reminders or "your quote is ready" messages to actually arrive. 2.47.x is on n8n's beta channel as of 9 Nov (stable is 2.46.x). The WhatsApp agent channel arrived in 2.44.0 and isn't in n8n's agent docs yet.

How:

  1. In WhatsApp Manager, create and get approval for Utility templates, such as an appointment reminder with blanks for the name and time. n8n only reads templates. For now the agent sees only Utility templates that have blanks in the body, not ones with a media header, a header blank or a button that takes a value.
  2. Update to 2.47.0 or later, then publish an n8n agent with the WhatsApp channel connected. Turn on approval for the channel: send_template is a sensitive action, so it's preselected for approval and you sign off before anything goes out.
  3. Test it with a Manual Trigger → Message an Agent workflow. Send "List the WhatsApp templates you can use", then "Tell <a test number> on WhatsApp that their order A-1001 has been confirmed". In the output, toolCalls should show send_dm failing, then list_templates, then send_template succeeding. Once that works, put the follow-up on a schedule.

Source: n8n 2.47.0 release notes https://github.com/n8n-io/n8n/releases/tag/n8n%402.47.0 · pull request #41133 (what it does and how to test it) https://github.com/n8n-io/n8n/pull/41133

5. Describe a process to Mistral and get a running workflow that routes Slack requests

What it does: on 5 Nov, Mistral, the French lab behind the Mistral open-weight models, expanded Studio, its platform for running AI agents and workflows. You describe a process in Mistral Vibe, and it writes, tests and deploys a workflow, with code you can inspect and edit and a visual graph of the steps. New event triggers start a workflow from Slack, GitHub, Notion, Linear, Microsoft Graph, Lucca or Ashby, so a new Slack message or GitHub issue can kick it off.

Who it's for: teams whose requests come in through Slack, GitHub or Notion and still get sorted and passed on by hand. The workflow builder is in public preview, and running workflows on Mistral's servers needs a paid plan (Pay-as-you-go or Enterprise). Some setup is through the API, so it suits someone comfortable with a little technical work.

How:

  1. In Mistral Vibe, describe the process, e.g. "When a message lands in #requests, decide whether it's a bug, a feature request or a billing question, and post it to the right team's channel." Review the code and graph it produces, then deploy.
  2. In AI Studio, create a webhook for Slack on the Webhooks page, then a trigger on the Triggers page (or use POST /v1/webhooks and POST /v1/triggers). Point the trigger at the workflow by its name, not its ID. You can add a filter so only certain event types start it.
  3. Paste the webhook URL Mistral gives you into your Slack app as the request URL. Post a test message, then check the trigger's events (GET /v1/triggers/{id}/events) for TRIGGERED or TRIGGER_FAILED. Failed deliveries are retried up to 4 times, and each delivery starts one run.

Why bother: Mistral says its own product operations team uses a Studio workflow that sorts ideas and customer feedback into bugs and feature requests and routes them to the right team, saving about 35 hours a week.

Source: Mistral, "Build with Studio" (5 Nov) https://mistral.ai/news/build-with-studio/ · Mistral docs, triggers and webhooks https://docs.mistral.ai/studio/workflows/building-workflows/triggers-and-webhooks · release notes https://docs.mistral.ai/resources/release-notes


⚠️ One thing to watch

Attackers are now using AI agents to hunt for API keys in public code. On 5 Nov, Microsoft reported that threat actors are using an agentic attack framework (ARTEX) to search public GitHub repositories for exposed secrets, including API keys for AI services, and then use them. In one case, an agent went from searching GitHub to stealing data from an Azure Key Vault in five minutes. If you or a freelancer ever pushed a key into a repo, deleting the file doesn't help: Microsoft says exposed keys stay usable until you revoke or rotate them. Scan your repositories, rotate anything you find today, and turn on GitHub secret scanning and push protection. Their write-up: https://www.microsoft.com/en-us/security/blog/2026/11/05/threat-actors-use-agentic-ai-tooling-to-find-and-abuse-exposed-secrets/


Sources credited above. All summaries are ours. Got a trick that worked for you? Reply and tell us.

Get next week's five, free · All issues

Latest Daily Frontier Digest: 21 Nov 2026: Inkling 2 Mini: a 26B open model that thinks by looping its middle layers