Drive coding agents from your phone, then verify on a preview link

How this helps you

You can ship small UI and app fixes while away from your desk, and you only merge after you have seen the change in a real preview.

I am away from my desk. Here is a screenshot of the issue: [attach]. Change: [what to change]. Acceptance: when I open the preview, [exact behaviour I must see]. Use [Cursor cloud agents / T3 Code] with [model preference if any]. Open a PR, paste the preview link here, and list harness + model in the agents note. Do not merge until I confirm the preview.

What they did

Ben Davis stopped treating Grok Bot as a chat that writes code. In his setup the Bot stays running in the cloud with two major connections: Cursor cloud agents and T3 Code over MCP. He used to keep a couple of giant "god bots" that did all of his development work; those threads got cluttered fast, so he switched to many smaller Bots plus group chats instead of pasting walls of text into Discord. From his phone he can screenshot something he dislikes in the app, tell Grok Bot what to change, and let it spawn a Cursor or T3 run. The hard part, he says, is verification. He does not closely read the YAML a agent produces. He checks that the change shows up on a preview link, clicks through the app, and confirms the behaviour. He also keeps a note in agents.md that records which harness and model ran each task. Once CI is green he merges, sometimes without opening the PR on a laptop. Commenters immediately asked how he avoids blowing the usage cap; one reply argued orchestration only pays if each task arrives with its acceptance test written down.

Who it's for: Builders who already use Cursor or T3 Code and want Grok Bot to manage runs from a phone.

Before you start

• Grok Bot connected to Cursor cloud agents and/or T3 Code
• A project with preview deploys or a local preview the Bot can share
• Comfort reviewing PRs on mobile

How to try it

1. Connect Grok Bot to the coding harness you already pay for (Cursor cloud agents and/or T3 Code) and confirm the Bot can open your project environment.
2. Split work into small specialist Bots or threads instead of one god Bot that owns every repo task.
3. From your phone, send one clear change with a screenshot and the acceptance check (what must be true when it is done).
4. When the run finishes, open the preview link and click through the app yourself before you merge.
5. Record which harness and model ran in your agents notes so later audits are not guesswork.

What to expect

You screenshot a messy chat sidebar on your phone, ask for a tidier layout, get a PR with a sticky preview link, tap through it, then merge when CI passes.

Common mistakes

• Sending a vague "make it nicer" without an acceptance check, then spending usage on follow-ups.
• Merging from the green CI badge alone without opening the preview.
• Stuffing every repo into one god Bot until the thread is unreadable.
• Forgetting to note which harness ran, so you cannot debug a bad change later.

Related tips

• Make your Bot the manager, not the coder: one rule and locked doors: Let your Bot plan and check the work while coding agents write the code. Start with one rule, and remember every Bot shares one computer and its logins.
• Watch coding work live and get recordings back: With Cursor connected, your Bot links you to the cloud agent live and sends screenshots and screen recordings.
• Two habits that make usage last: batch requests, send words not screenshots: Every message makes a Bot reread the chat, so send related requests together, skip 'thanks' replies, and paste text or cropped images instead of full screenshots and PDFs.

Worth knowingFrom Ben Davis on YouTube (5 Nov 2026). The phone loop, Cursor/T3 split and preview verification are his; we skipped Firecrawl sponsor lines.

Learn more: Grok Bot Use Cases: Real Workflows From Real Users · All Grok Bot workflows tips