What this is.
Opinion + Experience + Fact (45% opinion · 30% experience · 20% fact · 5% fiction)
Written in collaboration with AI — I discuss, I do not outsource.

Writers ship words online at scale. Most of them don't tell you what kind of writing you're getting — fact, experience, opinion, or imagination. So you sort silently, paragraph by paragraph, in your head 🔍. I did that as a reader for 20+ years before I started writing. As a writer, the first decision I made was to fix that.

Chapter 1. Ideas Came Easy. Articulation Took Practice.

20+ years of building 🛠️. 40+ products shipped 📦. Plenty of stories to share.

What took longer to develop was articulation. English came to me later in life. The ideas came fast 💡. Putting them down the way they lived in my head took work — passes, then more passes. That has always been the real craft.

▸ First principle. Ideas are abundant; the scarce resource is clear articulation that other people can actually use.

For most of my life I dealt with that gap by working around it — leaning on people whose craft was articulation. The next chapter is about one of them.

Chapter 2. Two Writing Desks at Home

When our kids want to write something, they know exactly where to go. For what to write — the idea, the angle, the thing worth saying — they come to me. For how to write — grammar, structure, polish — they go to their mom. She is a technical writer.

Yes, our family literally has two service desks. And yes, the queue at her desk is always longer 😊. I have always been the what parent. The how has always lived at a different desk.

▸ First principle. Separating "what" and "how" works inside a family, but if you want to ship serious work, you eventually have to own both.

Owning both used to mean hiring an editor or learning to edit yourself. Today there is a third option — and it changed how I think about writing entirely.

Chapter 3. Finding My Own "How" Partner

Today I finally have a "how" partner of my own — my AI agent 🤖. One I can talk to anytime.

The Age of Google taught me how to learn 🔍. Search, read, find sources of truth, then stitch those into whatever you are building. I started as an embedded engineer, but that approach helped me build complete products — hardware to cloud, bench to factory floor 🏭.

The Age of AI gives me a different mode. AI can answer questions too — that is a valid use. What makes AI genuinely new is discussion 🧠.

▸ First principle. Tools change, but the job is the same — turn information into working systems. Google did it with links; AI does it with dialogue.

The shift from search to dialogue sounds small. It is not. The next chapter is about why.

Chapter 4. Why Discussion Beats One-Shot Answers

Under the hood, AI predicts the next best token from the context it has. A bare question gets a generic answer. A discussion does three things at once: it feeds richer context, it lets me correct as it pushes back, and it sends the AI to research and validate against real sources. Each turn, the context sharpens. The predictions sharpen with it.

What emerges is articulation ✨ — work the two of us did together. Google gives you results. AI discusses, validates, and builds an answer with you.

Those of us who survived a wall of Stack Overflow tabs know which one we prefer 😅.

There is another reason discussion matters. A recent MIT study compared writers who used an AI assistant to generate essays upfront with writers who wrote first and brought AI in afterward. The second group — the ones who led with their own thinking — showed higher brain engagement, better memory recall, and stronger ownership of their work.

Discussion sharpens you over time 🎯. Outsourcing sharpens the AI.

▸ First principle. If the tool does the thinking, your brain goes idle. To grow, you have to carry the reasoning and let AI amplify, not replace, your work.

So discussion is better than dictation. But there is still a missing piece — and it has nothing to do with AI at all.

Chapter 5. The Idea Lives in Your Head, Not in the Prompt

The idea lives in your head. AI cannot read it.

That is the whole problem. When I ask a question — anyone's question — the words on the screen carry only a fraction of what is actually going on upstairs. The full picture, the constraints, the weird edge case I am worried about, the several products I have seen this fail on, the part I cannot quite put into words yet — all of that stays in my head.

Engineers have known this for a long time 🦆. Rubber duck debugging says: explain your bug to a rubber duck on your desk, and in the act of explaining, you solve it yourself. The duck never spoke. You did the work. What the duck did was force you to articulate.

AI works the same way, with one big upgrade. It listens, pushes back, and asks what you missed. Each question it asks is a nudge to surface more of what is already in your head.

This is how AI helps you think out loud. That is why the same AI tool gives different quality of answers to different people. The output depends on the quality of the conversation.

▸ First principle. The real bottleneck is not the model; it is your ability to externalise your own context so the model has something meaningful to work with.

Externalising your context changes the writing. It also changes how you read everyone else's. Once you start sorting your own thoughts into categories, you start sorting other people's the same way.

Chapter 6. Four Lenses I Read Through

Two decades of reading others' writing taught me to ask one question first: what kind of writing is this? 📖

Is it fact — something I can verify? Is it experience — something this person lived through? Is it opinion — a position, even an informed one? Is it fiction — a hypothetical, a scenario, a thought experiment?

The answer changes how much weight I give it. An engineer's lived experience building a sensor driver is worth more to me than an essay theorising about sensor drivers. A cited benchmark is worth more than a comparison opinion. A hypothetical has its own role — it is how we imagine futures.

All four have their place. They are just different 🎯. A post built on pure opinion is still useful, if I know it is opinion. Writing that tells you what it is earns the reader's trust.

When you are shipping products, you learn to sort quickly. What can I rely on? What is someone's view worth considering? What is storytelling making a point?

I still read that way. I just did it silently, for years, in my own head.

▸ First principle. Every reader silently sorts what they read into fact, experience, opinion, or fiction. Labeling that sort upfront speeds trust on both sides.

As a reader, I could afford that silent sort. As a writer, my role flips — and so does the responsibility.

Chapter 7. The Responsibility Flip: Reader to Writer

Today I am writing 🎤. That changes my role.

As a reader, I could quietly judge what kind of writing each post was. The writer never knew. As a writer, the responsibility flips. If I am asking readers to spend a few minutes with my words, I owe them the same clarity I always demanded as a reader.

Saying "this is opinion" out loud is a small act. It lets the reader weigh it accordingly. Saying "this is experience from 40+ products" tells them: this is what I lived — one builder's view. Saying "this is a hypothetical" tells them I am imagining, not reporting.

So here is the commitment 🤝 — every post I write from today forward will tell you what it is. The labeling will be visible. The formula will be honest. If I get the mix wrong, please tell me. Accuracy matters to me more than appearing perfect.

Writing publicly is a privilege. The least I can do is be honest about what I put in front of you.

▸ First principle. Long-term trust comes from showing your confidence and perspective directly — letting readers see what kind of writing this is, not making them guess.

So that is the why. Here is the how — four steps, simple enough to apply to any post.

The Four-Category Stamp

Every post I write from now on carries one. Four steps you can apply to your own writing 📐:

1. Tag each paragraph by its dominant substance — Fact, Experience, Opinion, Fiction (or Hypothesis, if you are making a serious unproven prediction rather than a playful one).

2. Weight by word count. Round to the nearest 5%. Pretending to be 62.7% opinion is its own kind of dishonesty 😊.

3. Surface the breakdown at the top (a short top declaration) and bottom (full stamp with sources). Both should match exactly.

4. Cite anything labeled fact with a source the reader can follow. If you cannot link it, it is probably opinion.

When I committed to this approach, I checked whether others do something similar. They do. Scott Alexander (Astral Codex Ten) puts a one-line "epistemic status" at the top of his posts. NPR clearly labels opinion content and distinguishes it from news, analysis, and reviews. Each of us found a version of this idea independently. Mine is four categories with percentages, built for a builder writing on LinkedIn in the age of AI.

The Stamp at the Bottom Is the Commitment Kept

If you take one thing from this: writing online is more useful when readers know what they are reading 🎯. Twenty++ years of ideas. The labels are what make them worth your time now.

Next: a pattern I have watched repeat on project after project — and what finally stops it.

Labeled: Opinion + Experience + Fact
(45% opinion · 30% experience · 20% fact · 5% fiction)

Sources:

  • Your Brain on ChatGPT — MIT Media Lab study (2025)
  • Scott Alexander / Astral Codex Ten — epistemic status tags
  • NPR Opinion labeling policy and transparency standards
  • Advanced Chess / Centaur Model — Kasparov, 1998

(Written in collaboration with AI — I discuss, I do not outsource.)

— Ritesh | ritzylab.com