The Humanizer

Anti AI-slop, operationalized for prose

Anti-slop

Your brand is anti-AI-slop. This is that stance made concrete for writing — the tells that betray a machine wrote it, and how to remove each one. It's the enforcement layer under every other copy page: write the thing, then run it through here. Same move as your failure-modes catalog, pointed at prose instead of agents.

Pattern catalog adapted from the open-source "humanizer" skill, itself based on Wikipedia's "Signs of AI writing" (WikiProject AI Cleanup, CC BY-SA). These are widely-observed, factual tells. Examples rewritten on your topics.

Pass 1 — required

Remove the tells

29 patterns across content, language, style, communication, and filler. Each has a specific fix.

Pass 2 — also required

Add soul

Clean isn't enough. Voiceless prose is as obviously AI as sloppy prose. A person has to be behind it.

Content tells Language & grammar Style Communication Filler & hedging Add soul The final audit

Content tells

The machine puffs up importance, invents significance, and hedges attribution. These are the tells in what the writing claims.

Inflated significance & broader trends

Watch stands/serves as, a testament to, pivotal moment, marks a shift, evolving landscape, setting the stage for, deeply rooted

The Memory Ladder marks a pivotal moment in the evolution of agent design, reflecting a broader shift toward structured cognition.

Human

The Memory Ladder is four storage stages: files, database, RAG, graph. You climb when the data outgrows the current rung.

Undue emphasis on notability

Watch widely cited, featured in leading outlets, active social media presence, recognized by experts

His frameworks have been cited across the industry and he maintains an active presence on multiple platforms.

Human

He argued for graph-database agent memory in 2025, before most people had moved off flat files.

Superficial "-ing" analyses

Watch highlighting…, underscoring…, reflecting…, ensuring…, contributing to…, showcasing… tacked onto a sentence's end

The ladder starts with files, reflecting a preference for simplicity and underscoring the principle that complexity should be earned.

Human

The ladder starts with files because most projects never need more. Complexity is earned, not adopted.

Promotional / advertisement language

Watch boasts, vibrant, rich, profound, groundbreaking, seamless, nestled, in the heart of, must-read, stunning

This groundbreaking approach offers a seamless, intuitive way to give your agents a rich and profound memory.

Human

This approach gives an agent memory that persists between sessions. It's more setup than flat files and worth it once you outgrow them.

Vague attributions & weasel words

Watch experts argue, observers have noted, industry reports suggest, some critics say (with none named)

Experts believe agent memory plays a crucial role, and observers have noted its growing importance.

Human

In my own 100-agent system, the agents that share a memory store make far fewer repeated mistakes than the ones that don't.

Formulaic "Challenges & Future Prospects"

Watch Despite its… faces several challenges, Despite these challenges…, Future Outlook, Challenges and Legacy

Despite its promise, agent memory faces challenges. Despite these challenges, the field continues to thrive.

Human

The hard part is knowing when to climb a rung. Move too early and you've built a graph to store four rows.

Language & grammar tells

The tells in the words themselves — the AI vocabulary, the elaborate constructions, the forced symmetry.

Overused "AI vocabulary"

Watch delve, crucial, pivotal, tapestry, testament, underscore, intricate, landscape, showcase, vibrant, foster, garner, interplay

Delving into the intricate tapestry of agent design reveals a crucial interplay of components.

Human

Agent design has three layers: instructions, charter, and skills. They interact in specific ways.

Copula avoidance (dodging "is/are")

Watch serves as, stands as, represents, boasts, features, offers — where "is" or "has" would do

The charter serves as the identity layer and boasts five distinct components.

Human

The charter is the identity layer. It has five parts.

Negative parallelisms & tailing negations

Watch "Not only… but…", "It's not just X, it's Y", and clipped fragments like "no guessing" tacked on the end

It's not just a memory system; it's a philosophy. The agent recalls automatically, no manual lookups.

Human

The agent recalls what it needs without you writing the lookup by hand.

Rule-of-three overuse

Watch everything forced into groups of three to sound complete

The system is fast, flexible, and future-proof, offering clarity, control, and confidence.

Human

The system is fast, and it holds up as your data grows.

Elegant variation (synonym cycling)

Watch the same thing renamed every sentence — the agent / the system / the assistant / the tool

The agent stores context. The system recalls it. The assistant applies it. The tool improves over time.

Human

The agent stores context, recalls it when needed, and gets better as the store grows.

False ranges ("from X to Y")

Watch "from … to …" where the two ends aren't on any real scale

Agents handle everything from remembering your name to reshaping the future of work.

Human

Agents can remember prior sessions and act on that history.

Passive voice & subjectless fragments

Watch "No configuration needed", "The results are preserved automatically" — the actor is hidden

No setup required. The memory is persisted automatically.

Human

You don't set anything up. The system saves the memory for you.

Style tells

The visual fingerprints — punctuation, formatting, and casing habits the machine can't help repeating. Several of these overlap your own preferences already.

Em-dash overuse

Watch em dashes standing in for commas, periods, or parentheses — everywhere

The ladder works—but only if you climb it deliberately—not all at once—which most people miss.

Human

The ladder works, but only if you climb it deliberately. Most people try to jump straight to the top.

Already your rule: no em dashes for parentheticals. Semicolon if both halves stand alone, comma if one leans.

Boldface overuse

Watch mechanically bolding phrases mid-sentence for emphasis that isn't earned

It combines files, databases, and knowledge graphs into one unified approach.

Human

It combines files, databases, and knowledge graphs into one approach.

Inline-header vertical lists

Watch bullet lists where each item is a bold header, a colon, then a sentence restating it

• Speed: It's faster. • Memory: It remembers. • Scale: It scales.

Human

It's faster, it remembers between sessions, and it holds up as the data grows.

Title Case in headings

Watch Every Main Word Capitalized In A Heading

## Building Robust Agent Memory Systems

Human

## Building robust agent memory systems

Already your rule: sentence case everywhere except the wordmark.

Emojis as decoration

Watch 🚀 headings and ✅ bullets sprinkled for energy

🚀 Launch: shipping soon. 💡 Insight: memory matters. ✅ Next: try it.

Human

Shipping soon. The insight is that memory is the bottleneck. Try it and see.

Curly quotation marks

Watch “curly” quotes where straight ones belong (a ChatGPT fingerprint)

He called it “earned complexity” and the phrase stuck.

Human

He called it "earned complexity" and the phrase stuck.

Communication tells

The chatbot leaking through — the assistant's helpful voice pasted where content should be.

Collaborative chatbot artifacts

Watch Of course!, Certainly!, I hope this helps, Would you like…, let me know, here's a…

Here's an overview of agent memory. I hope this helps! Let me know if you'd like me to expand.

Human

Agent memory is where most of the reliability problems actually live.

Knowledge-cutoff disclaimers

Watch as of my last update, while details are scarce, based on available information

While specific details are limited, it appears agent memory became a focus sometime around 2024.

Human

I was arguing for graph-based agent memory in 2025, and most people dismissed it.

Sycophantic / servile tone

Watch Great question!, You're absolutely right, That's an excellent point

Great question! You're absolutely right that memory is complex. Excellent point about persistence.

Human

Persistence is the part people underestimate.

Filler & hedging tells

The padding — words that take up space, soften claims into mush, and announce instead of doing.

Filler phrases

Watch in order to, due to the fact that, at this point in time, has the ability to, it is important to note that

It is important to note that the system has the ability to process data in order to improve recall.

Human

The system processes the data to improve recall.

Excessive hedging

Watch could potentially possibly, might arguably, it may be the case that

It could potentially possibly be argued that structure might have some effect on reliability.

Human

Structure improves reliability. I've watched it happen across dozens of agents.

Generic positive conclusions

Watch the future looks bright, exciting times lie ahead, a step in the right direction

The future of agent memory looks bright. Exciting times lie ahead on the journey to excellence.

Human

Next I'm testing whether RAG-first retrieval can decide what to query in the graph.

Hyphenated word-pair overuse

Watch third-party, data-driven, real-time, long-term, end-to-end, decision-making — all hyphenated with perfect consistency

A data-driven, real-time, end-to-end approach to decision-making.

Human

A real time approach driven by the data, from input to decision.

Persuasive-authority tropes

Watch the real question is, at its core, in reality, what really matters, the deeper issue, fundamentally

The real question is whether agents can remember. At its core, what really matters is architecture.

Human

Whether an agent remembers comes down to how you built its memory, not which model you picked.

Signposting & announcements

Watch let's dive in, let's explore, let's break this down, here's what you need to know, without further ado

Let's dive into how agent memory works. Here's what you need to know.

Human

Agent memory has four stages, and most projects should stop at the second.

Fragmented headers

Watch a heading, then a one-line paragraph that just restates the heading before the real content

## Memory
Memory matters.
When an agent forgets, it repeats work.

Human

## Memory
When an agent forgets, it repeats work it already did.

Pass 2 — the half everyone skips

Add soul

Removing the tells gets you clean. Clean isn't the goal — voiceless writing is as obviously machine-made as sloppy writing. It reads like a press release nobody signed. The second pass puts a person back in.

Have opinions

React to facts, don't just report them. "I still don't know how I feel about this" is more human than a balanced list.

Vary the rhythm

Short, punchy sentences. Then a longer one that takes its time getting where it's going. Mix it.

Acknowledge the mess

Real people have mixed feelings. "Impressive, and a little unsettling" beats "impressive."

Be specific about feeling

Not "this is concerning" but the concrete thing: "agents churning away at 3am while nobody's watching."

Clean but soulless

The experiment produced interesting results. The agents generated large amounts of code overnight. Some were impressed, others skeptical. The implications are unclear.

Has a pulse

I genuinely don't know how to feel about this one. A pile of code, generated while everyone slept. Half the room is thrilled, half is explaining why it doesn't count. The truth's probably somewhere boring in the middle — but I keep thinking about those agents working through the night.

The final audit — the two-prompt pass

1Write your draft, then run passes 1 and 2 above.

2Ask: "What makes this so obviously AI-generated?"

3Answer honestly — name the remaining tells in a line or two.

4Then: "Now make it not obviously AI-generated." Revise against your own answer.

✓The self-audit catches what the checklist misses. It's the step that separates clean from human.

Created for you, with ❤️ by Robert Evans · The AI Design Architect