When You Don't Know What to Build with AI: A 4-Step Weekly Self-Observation Routine for Automation Ideas

Many creators and developers want to build practical tools or automate repetitive workflows with artificial intelligence, but getting stuck on deciding "what to build" often stalls projects before they even start. AI prompt and workflow creator Prompt_what (@Promptwhat) shared an actionable four-step routine that flips traditional brainstorming upside down: letting an AI assistant observe your daily workflow for one week to uncover genuine automation opportunities grounded in your real-life friction.

When starting an AI project, many people attempt to invent a grand software product or hypothetical consumer app from scratch. Such top-down ideas often lose momentum because the creator is not their own active daily user. The routine proposed by @Promptwhat takes the opposite approach by using periodic conversational check-ins to capture raw, authentic pain points as they happen throughout the workday.

1. Starting from Real Daily Friction Instead of Forced Ideation

Abstract brainstorming frequently results in abandoned side projects because the underlying problem lacks personal urgency. In contrast, automating tasks that you personally perform by hand every day delivers immediate, tangible relief and provides clear criteria for iterative improvement.

As @Promptwhat noted: "Ideas last longer when they emerge from your own discomfort rather than when you squeeze them out of your head. Thanks to this routine, most of the tasks I used to check manually every week are now handed over to AI." Becoming the primary user of your own automation is the most reliable way to maintain focus and build tools that genuinely save time.

2. The 4-Step Weekly AI Observation Routine

The core principle is positioning your LLM not merely as a passive question-answering tool, but as an observational partner that systematically monitors your daily task flow.

• Step 1: Set bi-hourly scheduled check-ins in ChatGPT or Claude: Configure a scheduled prompt or recurring reminder from morning until bedtime that asks you every two hours: "What are you working on right now? What was tedious?"
• Step 2: Reply with brief, low-friction answers for one week: Whenever the notification arrives, record a short, unfiltered response describing your current task and any annoying friction points. Keeping responses concise—without drafting full paragraphs or polished sentences—is essential to sustaining the habit across all seven days.
• Step 3: Analyze the full weekly log with a synthesis prompt: At the end of the week, collect the entire multi-day conversation log into a single prompt and ask the AI: "Identify the 3 recurring tasks I found tedious, and suggest tools or workflows to reduce them."
• Step 4: Pick one item and take action: Select one candidate from the three prioritized pain points and solve it—either by building a lightweight custom script or agent using modern vibe-coding tools, or by integrating a proven existing off-the-shelf application.

3. Practical Considerations for Sustainable Execution

To get the most out of this observation framework, keep the following considerations in mind:

• Minimize logging friction: If answering the bi-hourly prompt feels like extra administrative work, the habit will quickly break down. Brief fragments such as "manually reformatting CSV columns" or "renaming thirty receipt PDFs one by one" provide more than enough context for the weekly synthesis.
• Keep the initial build scope tight: Rather than attempting to build a comprehensive productivity platform, target single-purpose scripts or custom instructions that turn a 15-minute daily headache into a one-minute automated step.
• Shift the role of your AI assistant: Moving from one-off transactional queries to longitudinal logging allows models like Claude and ChatGPT to serve as analytical mirrors, revealing repetitive behavioral bottlenecks that you might otherwise overlook in the heat of daily execution.

Original source

• Prompt_what (@Promptwhat) Original Post: X (Twitter)