Article

I hate agents - How to create one with Grok Bot

Ruben Hassid, How to AI · 2026-08-12.August 12, 2026

Summary

This practical guide shows how to build a simple AI agent with Grok Bot, using a daily AI-news digest as the worked example. The setup is deliberately lightweight: open Grok Bot, create a new bot, give it one clear goal, answer its follow-up questions about frequency and scope, connect the apps it needs through Plugins, and then refine the instructions based on the results it produces. The useful lesson for knowledge workers is that a first agent does not need complex engineering; it needs one bounded recurring task, clear instructions, the right tools, and feedback over time.

Article source handoff

Ruben Hassid, How to AI · 2026-08-12

This page is a short editorial source record. The summary and notes help you decide whether to continue, while the full article remains with the original author or publisher.

PublicationHow to AI, Substack
Author / creatorRuben Hassid
Source dateAugust 12, 2026
Read full article at source

Key Takeaways

  • Start in Grok Bot with one specific job. Click + to create a bot and describe the goal in plain language; Hassid’s example is a daily digest of AI news relevant to his work.
  • Let Grok Bot turn the goal into a routine. It asks follow-up questions such as how often the agent should run, after which you can make the task more precise by specifying exactly what sources, topics, and outputs you want.
  • Connect the tools the agent needs through Plugins. Grok Bot can connect to external apps; Hassid highlights integrations such as Gmail and Apify, allowing an agent to gather information and deliver its result through another service.
  • Write operational instructions, not broad ambitions. A prompt such as “grow my business” is too vague; a useful agent defines the recurring task, what information to collect, what to ignore, how the output should look, and when it should run.
  • Keep the first version constrained and refine it through feedback. Hassid recommends drafts only at first, clear stop conditions, and telling the bot what you dislike so subsequent runs improve rather than giving the agent unrestricted authority immediately.