Raymond: Removing AI Fluff from Writing with Agent Workflows

AI Summary · A Serial Entrepreneur’s Perspective (The following content is distilled by AI; the views belong to the original author. You may skip the source article after reading.)

Raymond shared his H1 review and open-sourced a writing tool that strips out “AI flavor.” Key figures: the tool flags 35+ Chinese AI-writing tics (A·tested); he traded a pause on a bootcamp for 60+ days of white space (A·self-reported). For anyone building a business: it offers a concrete way to dial down the “AI feel” and raise professional polish, plus a mental model that prioritizes judgment over execution. Action item: try the open-source Skill to tighten your copy and borrow his AI Agent role-splitting pattern.

• Download the open-source Skill to batch-detect and rewrite AI-sounding copy
• Set up a workflow where strong models plan and weak models execute
• Study his move to pause time-heavy projects in favor of strategic white space
• Evaluate the cost-reduction potential of AI Agents in deployment and management

In the Field with the Open-Source “De-AI” Skill: Spotting 35+ Tics and Redesigning Human–Machine Roles

Content creators are losing audience trust because their work reads too much like AI. Raymond’s open-source Skill scans and rewrites copy in bulk, turning the vague goal of “removing AI flavor” into a repeatable standard process. The core move is a two-layer architecture—strong models judge, weak models execute—which lowers the barrier to professional-grade content and cuts through homogenized competition.

How the Tool Works and What the Numbers Say

Built from Wikipedia’s entries on the characteristics of AI-generated text and tuned for Chinese, the tool goes well beyond synonym swapping. It identifies and tags more than 35 typical Chinese AI writing tics, then offers concrete rewrites. In testing, when fed an article written in 2024 without AI assistance, it correctly flagged it as “no AI traces,” confirming that false positives stay manageable. For creators, that means automating the copy-edit pass and keeping human effort focused on creativity and strategy.

The Core Insight: Judgment Beats Execution

When reviewing his half-year workflow, Raymond stressed that once AI takes over most execution work, the human edge returns to “what is worth doing.” His workflow has a clear split: strong models like Fable 5 and Opus draft plans and frameworks, then Codex or other models carry out the tasks. Keeping expensive, smart judgment apart from high-volume execution stops him from clogging powerful models with trivial instructions and frees their reasoning capacity. He also paused his time-heavy bootcamp to buy himself 60-plus days of blank calendar space for writing and prep work on an Agent course—a deliberate “strategic white space” that has proven essential to sustaining long-term output.

Repeatable Steps to Implement

• Get the tool: Download the open-source “Speak Human” Skill (/speak-human-tw) and plug it into your existing content workflow.
• Batch-scan: Feed long drafts or flagship articles into the tool and pull back a report that marks 35+ tics with rewrite suggestions.
• Split model roles: In any task, force a separation between planning and execution. During planning, use a top-tier reasoning model and avoid scripting detailed steps in the prompt—let the model find the best path itself. During execution, hand off coding or content fill to lower-cost models.
• Audit your time quarterly: Review the return on each project and kill or pause high-maintenance, low-marginal-benefit routines. Redirect the recovered hours into core assets—courses, books, or long-form content.

Original post · Raymond三十｜Helping busy modern professionals work smarter and live better: a productivity guide: Read the original →

Related tool recommendation (promo): TinyFish gives AI web-content search capability