Video summary
Prompting is dead. Here is how you create loops
Main summary
Key takeaways
Overview
The video argues that “prompting agents” should be replaced with automated “coding loops / loop engineering.” The speaker claims most people discuss loops but don’t show how to build them, then presents a concrete, reusable loop design intended to automate most “vibe coding” work.
Core technological idea: replace manual agent prompting with autonomous loops
Before loops (traditional prompting)
- You repeatedly prompt an agent to write code.
- Then you manually test it (e.g., open localhost, click around).
- If you’re not at the computer, progress stops—so the person becomes the bottleneck.
After loops (“Fin cycle”)
- You provide lightweight human input.
- The system runs an autonomous workflow across the day:
- create specs
- build repeatedly
- test/review repeatedly
- merge when ready
Product/feature claim: the “Fin cycle”
The speaker’s main contribution is a named workflow called the “Fin” cycle, described as consisting of three skills:
-
Specification
- An interactive “spec” skill that asks questions until it fully understands the task.
- Produces tasks with:
- acceptance criteria
- scope/testing notes
- Example tooling: stores tasks in Linear as a “backlog.”
-
Development / Assembly (build cycle)
- A loop that repeatedly builds all pending specifications from the backlog.
- Uses an example command pattern like:
/loop 5m /build(runs every 5 minutes).
-
Review / Inspection (verification cycle)
- A loop that checks and tests specs marked “Ready for review.”
- Performs actions like:
- opening a browser
- clicking/testing
- generating a pull request
- instructing how to test in a sandbox/preview
- Example command pattern: a similar 5-minute loop that waits for labels/states.
Automation behavior (end-to-end workflow)
-
Morning
- The user gives a general idea.
- The spec skill interviews and generates detailed tasks/issues.
-
During the day
- The build loop implements based on those spec tasks.
- The verification loop tests results and moves tasks forward (e.g., labels as ready).
-
End of day
- The system posts a merge request/preview to a messaging channel (the speaker uses Slack).
- The user approves merge via an emoji reaction (rocket emoji).
- The system marks the merge complete.
Claimed benefits
- The speaker claims ~95% of vibecoding work is automated.
- User effort is reduced to:
- manual confirmation at the start (ideation/spec prompts)
- manual confirmation at the end (approving merges)
- Reliability improves because verification/testing is built into the loop, rather than depending on the user to keep running prompts/tests.
Tutorial/guides and setup assets mentioned
-
A free HubSpot guide referenced: “Claude Code Playbook”
- Focuses on building trustworthy AI agent systems:
- reliability
- tips for Claude “rules file”
- hooks
- dynamic workflows
- subagents
- orchestration scripts
- Focuses on building trustworthy AI agent systems:
-
A Google Doc mentioned that includes:
- prompts to configure each of the three Fin cycle skills
- a prompt to connect external tools (Slack/Discord/Telegram, Linear, etc.)
- copy-paste wiring instructions for integrating the system
-
Suggested setup acceleration:
- ask an agent to read the transcript/video and set up the loop automatically.
Notes on tools (integrations)
- Compatible with “vibe coding” tools such as Claude Code and Codeex (and “any tool” that supports custom skills/instructions).
- Uses Linear for task tracking:
- backlog / issue states / labels (e.g., “Ready for review”)
- Uses a chat/notification channel:
- Slack preferred
- Discord/Telegram also possible
- Mentions preview behavior via a sandbox/preview branch:
- the system tests changes before merge.
Main speakers / sources (from the subtitles)
- Primary speaker: creator of the video / “Finfam” channel (unnamed in the subtitles)
- Referenced third parties:
- Peter Steinberg (mentioned as advising stopping prompting agents and building cycles)
- Boris (described as “from Claude Code,” credited with inventing it; said to focus on cycles)
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.