Andrew Ng says AI is the new electricity.

It will quietly power everything around us.

I agree.

And in my own work, manuscript writing is already one of the places where the shift is obvious.

The original system: Dump → Lump → Jump → Bump

I’ve been teaching manuscript writing for years.

And for a long time, I relied on one simple framework:

Dump → Lump → Jump → Bump

It was built for a world where you wrote every sentence.

Here’s how it used to work:

1️⃣ Dump

You pour everything you know into the document.

Snippets from papers.

Notes under rough headings.

Ideas you’ve been carrying around.

No structure.

No grammar.

Just material.

2️⃣ Lump

Now you start grouping:

• similar ideas
• related claims
• repeated concepts

You begin to see the skeleton appear.

3️⃣ Jump

You fix the flow:

• move paragraphs up or down
• reorder sentences
• ask, “Does this idea lead naturally to the next?”
• identify where the reader gets lost

You keep adjusting until the argument moves cleanly.

4️⃣ Bump

Finally, you polish:

• tighten the language
• refine topic sentences
• align with journal style
• verify every claim
• check your tone

For years, this system worked beautifully.

Then the world changed.

The moment I realized it didn’t feel right anymore

About a year ago, I taught my usual manuscript session.

Same slides.

Same examples.

Same four-step framework.

But something felt off.

Everyone in the room was already using AI — whether they said it out loud or not.

Suddenly, I was teaching a pre-AI workflow in a post-AI world.

I was still assuming that the hardest part of writing was getting words on the page.

That just isn’t true anymore.

Now the bottleneck is:

• clarifying your message
• controlling your argument
• protecting scientific interpretation
• making sure the paper reflects your thinking, not AI autocomplete

So I rebuilt the system.

→ Similar steps.

→ Completely different way of using them.

The role of AI in writing: grunt work, not thinking

Before we go into the updated system, let me be clear:

AI should handle the tedious parts. You should handle the thinking — all of it.

AI can draft, rephrase, summarize, and help with structure.

But:

• AI should not interpret your findings
• AI should not define your core message
• AI should not decide clinical implications
• AI should not carry the argument

You stay in control of the science.

And with that philosophy, here is the updated 5 step system:

Step 0: Give AI proper context (most people skip this)

A model is only as good as what you feed it.

Vague instructions create generic output.

So before using AI, prepare 5 elements of a manuscript:

1. Two to three key findings

In simple language. These should naturally lead to your core message.

2. One core message

What’s the single idea tying your findings together?

3. A clear objective

One sentence describing what your study set out to do.

(Usually rewritten after you understand your findings.)

4. The research problem

What gap does your study address?

How does it tie into the long-term problem your research program tackles?

5. Clinical implications

This part must come from you.

Only you know what your results mean for patients or clinicians.

These 5 elements act as the main context for the new AI-powered system.

STEP 1: AI-assisted Dump → starting from abundance

Dump used to be 100% manual.

Now it’s human + AI.

Your job:

• bring your five elements
• gather key references
• decide what section to start with

AI’s job:

• run a targeted search around your research problem
• identify missing angles
• surface relevant concepts
• generate a messy AI Dump Draft

This is raw material — not a draft.

You’re shaping something instead of starting from nothing.

I call this “starting from abundance”.

STEP 2: Human Pass → prune, group, own the logic

This step must stay human.

You:

• keep sentences worth keeping
• delete vague or incorrect lines
• group related points
• spot repetition
• correct misunderstandings
• add nuance only you can see

By the end, you have:

• clusters of clean ideas
• early paragraphs forming
• a clearer sense of the argument
• text that reflects your voice

Now you can shape the structure with precision.

Pro Tip: Do not take too much time in this editing —> this is just a quick editing pass. As you will see, we have a few more steps which will help shape this better.

STEP 3: Reverse outline

Most people try to outline first.

But if you could outline perfectly from scratch, you wouldn’t need a system.

Reverse outlining works really well with AI in the mix.

How it works:

1. Take your organized Lump text.
2. Distill each paragraph into one sentence.
3. This becomes your reverse outline.

Now you can see the actual structure of your argument — not the imaginary one.

With this outline, you:

• move pieces
• merge duplicates
• split overloaded sections
• correct gaps
• refine the story

STEP 4: Jump → Generate the AI First Draft

Once the reverse outline is final, bring AI back in.

Provide:

• your five elements
• your refined reverse outline
• any preserved paragraphs

Then ask for an AI First Draft.

This draft:

• follows your logic
• aligns with your data
• respects your argument
• avoids generic phrasing

You design the structure.

AI rewrites to match that structure.

STEP 5: Bump → polish with intention

Now you have a real draft — grounded in your thinking.

Bump is where you refine:

• tighter topic sentences
• correct claims
• remove hedging
• align tone with journal
• fix any model-inserted clichés

AI can assist, but only with surgical prompts.

You remain responsible for:

• accuracy
• causality language
• interpretation
• limitations
• the final voice of the paper

That part is non-negotiable.

The complete modern AI manuscript system

Here’s your workflow end-to-end:

1. Prepare your 5 manuscript elements
2. AI-assisted Dump → Start with abundance
3. Human Lump → prune, group, own the logic
4. Reverse outline → Jump → AI First Draft
5. Bump (polish)

This is what “AI as electricity” looks like in manuscript writing.

AI handles the hidden wiring — the heavy, repetitive work.

But you decide the architecture, the flow, the voice, the meaning.

How I operationalize this workflow in real life

This isn’t just a teaching framework.

It’s the exact workflow I use every week to keep all the AI-assisted steps and human judgment working together seamlessly.

And it’s the workflow I’ve been testing, breaking, rebuilding, and refining for over a year.

That process eventually became Research Boost.

A product built from hundreds of experiments, thousands of drafts, and real-world iteration with researchers across levels.

We’re still refining it.

Every week.

Sometimes every day.

Because the tools change.

AI models change.

But the core workflow: human thinking paired with structured AI collaboration remains the backbone.

(Start writing your manuscript with Research Boost FREE here: researchboost.com. It can help you with all the tedious parts, including your literature review —> you will get an in-depth literature review on all your key findings, which alone can easily take weeks.)

Want to test this on your next paper?

Try it on one section:

1. Write your 5 elements.
2. Ask for an AI Dump Draft.
3. Do a human Pass.
4. Build your reverse outline.
5. Generate an AI First Draft.
6. Polish.

Then compare it to your old workflow.

You’ll feel the difference immediately.

AI can write the words.

You decide what the words should mean.

Because the part that gives your paper weight has never come from the model. It comes from the clarity of your thinking, the discipline of your reasoning, and the choices only you can make.

So here’s the real question:

Are you using AI to expand your thinking, or to avoid it?

PROMPT OF THE WEEK

If you are not ready to try Research Boost, I have created a custom GPT for your literature review that you can access below for FREE. Just add your key findings one at a time to get key papers that agree and disagree with your finding.

P.S. Check out Research Boost for a much more in-depth automated review (deep literature review) with zero hallucinated references (guaranteed).