Goodbye AI, Hello Authenticity

Her (2013), discussed in "How Her Makes Sci-Fi Smart Again" (BBC Culture)

I have been writing blog posts on this website for years now. I originally did this as a way to populate my personal website and impressive recruiters. But over the years, I really started enjoying writing and expressing my thoughts on tech, business, philosophy, etc. These blogs became a way to express my thoughts to myself and the world to an extent. They also served as a tool that helped me understand elements of the uncertain times we are currently living in.

But recently, something has been bothering me. So during the last 1-2 years, I noticed that I really started using AI for everything. For coding, writing, planning, math, etc. Developers in general have shifted from writing and reviewing code to instead architecting AI agents and reviewing the final results. But the same, to an extent, has been applied to writing from what I’ve see. Most people are having AI write for them, rather it’s emails or papers. Now for coding, I am fine with that shift. The industry is changing for both startups and large tech companies. I believe that no one will write code but instead orchestrate agents, for better or for worse. But what bothers me is how often it seems that AI is writing, and in turn expressing, for the human instead of the human doing it themselves.

I have never been a “great writer”. In fact, all throughout my time in school, writing was my worst subject. I always sucked at spelling and grammar, and still do today. But, I’ve always been creative and I’ve always had ideas. I just struggled with expressing those ideas as text that also engages the reader. To the point where I spent a lot of time iterating to get my writing to a “good enough” point; it was painful. But that all changed when ChatGPT came out in late 2022. AI had finally gotten good at writing, to the point where I could just vomit ideas as random texts into the chat bot and it could convert it to a readable and professional text format. In a way, it allowed me to scale my writing output and express my ideas faster and “better” then ever before. So over time, I started using AI tools more and more often for researching and writing. But in 2025, something changed.

In 2025, AI agents became a big deal with the introduction of Anthropic’s Claude Code CLI. With these agents, over time, people started not even interacting/iterating with the model like you would with a chat bot. Instead, they started just giving the model tasks/contexts and then letting the agent work non-stop until it’s tasks was completed. It made everything more hands free. Which was nice for building stuff fast, but for writing it did something different. Something worse.

I believe for any form of writing that is used to express yourself, it should be fully or majority written by you, the human. I think AI is useful for researching and experimenting but using it to express your thoughts and findings in your writing is not great. It takes away something from the piece by makes it feel more artificial. And I started to feel like this. I started to question if my writing was my thoughts or ideas the AI injected into my writing. Not only that, but I noticed people reading my writing would spend less and less time reading it when they could detect any chance of AI being used. Which makes sense, if I used AI to quickly do the hard parts of writing, why should the human put in the extra effort to read that piece? It’s energy in and energy out, and using AI to write for you is low energy.

So with that in mind, I decided to do an experiment. I decided to measure how much AI I used in my writings on my personal website. To do this, I grab the content from all of my blog posts (main & archived) and I use a service called Pangram to measure how much AI was used in each piece. Pangram is a service that uses AI, specifically a classifier neural network model, that estimates the percentage odds of a given text being fully written by AI, assisted by AI, and/or fully written by a human. It was a service I first discovered on Twitter/X from a Tweet made by Paul Graham, and after a lot of testing, I can say with confidence that Pangram is really good at detecting AI generated/modified text. So with this tool, I feed all of my writings from my website to Pangram to process and got the following:

Published Post AI AI-assisted Human
2026-08-14 The Modern Moat 100% 0% 0%
2026-04-18 What is Power? 33% 14% 53%
2026-04-11 Three Years of Trading 97% 0% 3%
2026-03-22 My Perfect Tetris 90% 0% 10%
2025-12-02 Death Of Cha 0% 0% 100%
2025-10-28 Automated Vercel Cleanup 73% 0% 27%
2025-10-22 Open-sourcing Notify Cyber 63% 0% 37%
2025-09-14 From Cha to Ch 45% 0% 55%
2025-08-19 The Barbell of Software Value 100% 0% 0%
2025-08-12 My Robotics Research Chapter 64% 0% 36%
2025-07-11 Ch CLI 97% 0% 3%
2025-06-08 8090’s Top Coder Challenge 35% 0% 65%
2025-05-24 Vibe Coding 63% 0% 37%
2025-04-30 Guardian Grid SF 19% 0% 81%
2025-04-20 2019 AgBot Challenge 7% 0% 93%
2025-04-02 Unified Formatter with FM 0% 0% 100%
2025-01-27 DocuInsight Simplifies Signing 25% 8% 68%
2024-10-23 Truthworthy AI with PAX 62% 0% 38%
2024-08-23 AI Pentesting with TARS 0% 0% 100%
2024-07-08 Claude in the Shell 0% 0% 100%
2024-04-11 GamerCraft 16% 31% 53%
2024-03-31 2014 Passion Project 0% 0% 100%
2024-03-12 Cha Terminal Chat 0% 18% 82%
2024-03-06 MongoDB to Postgres 58% 6% 37%
2023-12-15 Benchmark Local LLM Models 0% 0% 100%
2023-10-19 ABM Marketing with InsightRed 0% 0% 100%
2023-07-20 AI Art with DALL-E 2 0% 0% 100%
2023-07-07 My Portfolio Website 0% 0% 100%
2023-05-29 Why We Built Notify Cyber 0% 0% 100%
2023-05-25 Sequelize vs Prisma 0% 0% 100%
2023-01-22 Algorithm Challenges 12% 0% 88%
2022-12-18 Building Labeler NearBy 0% 0% 100%
2022-05-25 THN Discord Bot 0% 0% 100%
2022-05-10 The Diminished King 0% 0% 100%
2021-03-28 2D L-System Generator 0% 0% 100%
2021-03-01 PID Control with TurtleSim 4% 0% 96%
2021-01-09 Simple Lane Detection 0% 0% 100%
2020-12-31 Moving Pose Algorithm 8% 0% 92%
2020-12-14 Computer Vision Projects 0% 0% 100%
2020-08-01 CS Field Session 0% 0% 100%
2020-05-31 Predicting Human Actions 0% 0% 100%
2019-12-28 Clue Game in Java 69% 0% 31%
2019-11-27 RaspPi Musical Keyboard 0% 0% 100%
2019-06-26 IBM Pingmaster Converter 0% 0% 100%
2017-07-31 NOAA 2017 Internship 0% 0% 100%

I then graphed those results to get a better picture of how much my usage of AI for my writing shifted over the years. To do this, I used OpenCode CLI and OpenAI’s GPT-6.1 Sol model on high reasoning to read this data and generate a graph showing this shift over time. After verifying everything, here is the final graph:

So the graph shows some interesting insights. It shows that my average AI usage for writing/editing my blog posts has increased from ~24.8% in the first half of 2025 to ~63.1% by the second half of 2025. With likely the turning point taking place around May 24, 2025. During which I, ironically, published my blog Vibe Coding. But from 2017 to 2023, pretty much I did not use any AI in any of my writings.

The exception of this was in 2019. I did not use AI to help me write in 2019, but I did clean up a couple of my pieces from 2019 because I thought they were bad. In doing so, when I did this analysis, 2019 was the outlier. But overall, it started picking up in 2024 and skyrocketed from the second half of 2025 to all of 2026.

I think the reason why the usage increased so much after 2024 was because the models just got better and harnesses/tools like Claude Code CLI came out, making it easier than ever to give more and more control to AI to help with writing. Way more control then during the chat bot era.

This is concerning, I did not know how quickly I gave up my control of my writing to AI until I dug into this and did this analysis. Now I do think the blogs I wrote from 2025 onwards are not worthless and do express my ideas/thoughts. My latest blog post, The Modern Moat, has been essential to me. It truly embodies my view of startups during the era of AI and has been an essential philosophy/principle for my startup venture(s) moving forward. But the issue is that this blog was pretty much ghost written by AI, despite the idea of the modern moat being very important to me. Humans usually pick the path of least residence, usually subconsciously. And for this case, I believe I fell into that trap.

I am proud of The Modern Moat and I independently verified every source and accept that I am fully responsible for the final text. But still, I lost my voice to an extent. I pretty much sent my voice through a filter that stripped all my flaws and characteristics, making a more polished text but one with no personality, and in term, something that people do not really want to fully read because of that.

Knowing this, I feel like this is NOT the path my writing should continue following. There has to be a pivot. I saw the numbers and I am disappointed. So moving forward, I will NOT use any AI to write my blogs. At most, I will use AI to just:

  • Help me research
  • Run experiments (though code usually)
  • Build assets like graphs or diagrams

For spelling and grammar correction, I will just use classic methods found in tools like Microsoft Word, Google Docs, Visual Studio Code’s Code Spell Checker extension, etc. For word suggestions I will mainly use sites like Thesaurus. I’ll also avoid AI-generated images unless it’s for touching up existing images or making simple images like logos and thumbnails. I know this may be controversial but that is my current position with AI-generated images.

But to be fair, these changes only apply to my writings. I still use AI tools heavily for coding, researching, building stuff, etc. I even have a pipeline that uses AI models to translate my blogs to various other languages. Translations especially has been very valuable because it has allowed me to make my writings accessible to a larger global audience. Which would never have been possible as a solo writer, making no money from my blogs, without AI. So, I am still very bullish on AI but I don’t want to give it control of my voice at the end of the day.

To conclude all of this, I am still very proud of all my blogs even those where AI helped with the process. Those ideas are mine and I review all my works. But the issue is that for the last ~2 years I have given away my voice to AI more and more. In doing so, I am giving up my speech, and in turn, making my writing worse for me and my readers.

In a time where we are all moving fast and breaking things, there are moments where we need to slow down and reflect. And I think writing is one of those cases where moving fast and scaling is NOT the best way to advance. Instead, moving slowly and thoughtfully is the best play when it comes to writing. So moving forward, AI will no longer represent my voice in my blogs.

I always want to be honest with anyone who works with me and reads my writings. So I made this blog post to make it clear how I used AI with my writing and what I plan to do moving forward. This blog post was made out of principle to my readers and myself. My ideas are my own, all these works express my ideas but moving forward, AI will not be used to help me write or clean up my writing. That experiment has ended. My voice is my own and no one (man or machine) has the right to express it for me.

With all of that in mind, I a proud to say that this blog post was 100% written by me, a human, with no AI assists at all! To back this up, here is the final results from Pangram in regards to this blog post:

Please NOTE that some code segments were trimmed from the raw blog post to fit Pangram's daily credit limit

Thank you all for reading and I hope to see you all soon with a new post.