Doing the right thing

 Was reminded off these two events. 

Things being cryptic are intentional!!

Bunch of us were evaluating a bunch of firms. 5 firms to be precise. Most of the firms were represented by the stakeholders themselves. One of the firm had made an exception and sent their senior person. 

None of us were bothered that an agent was sent on behalf of the principal. 

Across the board with all the 5 firms, the discussions were casual. More open, more informal. More friendly. As a result, guard was down and people were speaking freely.

This exceptional firm who had sent a senior person, made an off handed remark, in fact offered an information unsolicited, something to the effect of "Oh we don't do X because it is risky" Basically the person made that statement to showcase they are robust, they have best practices. 

The room was full of men. One of them, responded, "Thank goodness, She (mentioned the name) is not here, if not you would have had a tough time" The aforementioned "She" was the final decision maker for our evaluation. 

This exceptional firm was really exceptional, they were the largest, best, would have ranked the top in all metrics we were evaluating. 

At the end of the day, we huddled up internally for the review to prepare the recommendations before presenting to the decision maker.

The analyst who had prepped the deck outlining pros and cons for each firm had just mentioned, "Not culturally aligned" across this exceptional firm. None, literally none said anything about how exceptional this exceptional firm was or how they were actually crushing. All of us moved on without spending another second. 

Another Incident

Bunch of us were at a dinner. It was me who had organized that dinner with internal and external stakeholders to get to know each other and could help us with one of the critical initiative at the time.

One of the external stakeholder was unnecessarily rude to one of the waiters. It was odd, yet could be considered par for the course. Not always, folks treat the support people in a nice way and if they think not being nice is okay, they behave that way. 

One of my senior stakeholder pulled me aside and said, ensure we don't engage with this person. 

Good culture and Good values and Good principles are about doing the right thing, when no one is watching and even when stakes are really high!!!

Open AI feature requests

 CustomGPTs in chatGPT is such a cool stuff. More especially for non-tech folks to build something cool and powerful without much Tech knowledge.

However for a tool that making life easier, it could be far better and make life easier, if following few notes and error handling, enhancements and instructions are available.

First and foremost, the access to the editor and precondition to access it.

1. The editor is not accessible in the app mode. It can be accessed from the browser

2. Even from the browser, customGPT is available, starting only from Go users. If you are not a paid member and if you try to access, it should have a message that one has to be a paid user and have a CTA or redirect link to become a paid member. Well, In India, all users are go users, so at least for the next year, editor will be accessible for India based users. It can be and should be fixed for users in other regions.

3. The profile pic file limit. Though Dall E image generation for the profile image is cool, if i try to upload an image, have no idea what are the file formats supported and the size limit. I tried uploading a 5.5MB file and just got a random error message than a specific message

4. The description doesn’t support any formatting. It would be good to have formatting. In addition it also has character limit and it is not mentioned upfront. 

5. In the knowledge section, there are limits to the file size and and number of files that could be uploaded. The message/info could be made available than to try and figure out later.

What Would Jobs Do??

TLDR:

I have built a custom GPT: What Would Jobs Do?  

https://chatgpt.com/g/g-68f7a59494c481918995e7eecb050617-what-would-jobs-do-wwjd

Long Story:

When I was 11 or 12, a neighborhood senior kid told me about Apple, Mac and Steve Jobs. I genuinely thought he was bullshitting me thinking that i don’t know about Bill Gates. 

Fast forward, when I wanted to buy a gift for someone very close during my first onsite trip, a friend was like, if you really want to gift properly, get them a iPod mini. It was the entry product and still it was too expensive when compared to other mp3 players. 

I was curious and went down the rabbit hole and that is how my fascination and adoration with Steve Jobs got started. It was too strong, I ended up writing a book about him in Tamil. It is been more than 10 years and now and then I still get fan mail for it.

The unintended consequence of deep diving about someone like Steve Job is, me losing my mind over when people (read tech bros, founders) channel their inner “A” and passing it as modeling the Jobsian behavior. 

Probably, one of the most misunderstood personality. Thanks to the way, he was portrayed by most of main stream media (For what it’s worth, folks close to Steve himself felt the authorized biography of Walter Isaacson itself didn’t do justice and they got Brent Schlender and another author for redux of Steve Jobs bio)

So for more than few years, I have been meaning to write two books in English.

One, compile all his quotes and put it as a book.

Two, based on quotes, based on pivotal moments, write about how one could learn from Jobs to make a dent in the universe rather than being a painful “A”

I started compiling all his videos, interviews, books that I missed earlier and the books published after my book. (I had to rely a lot on iCon and lot of early articles across publications like Forbes, BusinessWeek, Time, Wired and Fortune to write my book; The Walter Isaacson wasn’t published when I was done with mine)

However, I never got around to writing them. I couldn’t get around to getting the transcripts from the YouTube video complilations. 

Later, I started playing around with NotebookLM and chatGPT to write scripts to compile the transcripts from videos. Meanwhile, I also wanted to be AI literate and was reading around and playing around with tools. 

Then it occured to me, wait a minute, what if I use AI to model Steve Jobs behavior. I have these many books, these many articles, videos and I could give it Open AI’s GPT builder.  Simple. (I couldn’t be more correct as well as more wrong; Correct cause, it is still relatively easy to use the GPT builder but still it is not that simple to build the Knowledge files to serve as inputs to the GPT builder) 

One, using the books as a whole would violate, Open AI’s usage guidelines which prioritizes fair use boundaries.  

More than that, the Input Token limits and practicality. 

The GPT file upload system doesn’t “absorb” the entire content at once like training a model. Instead, it uses retrieval techniques that pull in relevant file snippets (based on the prompt) when generating a response.

• Even If you upload an entire 700-page book, or books it won’t be searched efficiently.

• You may get partial context, or the model may miss relevant insights buried deep in long documents.

• There’s also a token cap (~100k per request) which limits how much of the file it can process per response.

And Quality of Responses Decreases with Raw Bulk

Feeding a GPT unstructured, untagged, long-form books:

• Makes the assistant guess context too often.
• Reduces precision in responses.

So, I realized this was more challenging (in a way) than writing a book cause, you still have to create topic notes from books, themes, organize it and not just that, these files should explicitly call out as  the GPT wouldn’t automatically “know” what’s from Jobs and what’s from me unless I make that distinction explicit in the formatting. It needs, content in a specific format, with attributes and tags, calling out, Quote, Source, Comment, Interpretation and the files cannot be ambiguous or cannot have unclear content. GPT may not know where the quote ends and where my comment begins. It might conflate my interpretation with Jobs’ own words.

It is cause of the way, how the GPT works.

When you upload a file to a Custom GPT, it doesn’t “read” the whole thing every time. Instead, it:

1. Searches for the most relevant chunks of content related to the prompt.
2. Pulls those chunks into the response as background.
3. Uses context to decide what’s a quote, what’s commentary, and how to blend it into the answer.

The clearer your formatting, the more accurate this retrieval. So how to build the files.

Q1:.txt or .docx — Which Is Better?

Format	Pros	Cons
.txt	” Lightweight, loads fast✔ No hidden formatting✔ GPT parses text cleanly	No headings, styles, or tables Can get messy with long documents
.docx	Supports headers, bolding, bullet points, tables✔ Good for structured documents with commentary	Slightly heavier Risk of hidden metadata (tracked changes, formatting issues)

Recommendation:

• If you’re uploading clean, well-structured material with multiple sections, .docx is better.
• If you’re uploading shorter, raw text content like quotes or notes, .txt is fine.
• In both formats, use clear headers or tags ([QUOTE], [COMMENT], etc.) for GPT to parse meaningfully.

You can use .docx files with tables, and GPT can infer structure better from them than from unformatted .txt.

 One Big File vs. Multiple Small Files?

Approach	Pros	Cons
One Large File	Easy to manage✔ Avoids GPT file upload limit (“10 files max)Good for a single tightly themed topic	Retrieval becomes fuzzy if file mixes many unrelated topics❌ Hard to update incrementally GPT may struggle to find the most relevant snippet
Multiple Small Files	Better for topic-based retrieval” Easier to test and update individually” More accurate context matching	More upload effort (GPT Builder has file count limits) Slightly harder to maintain if not organized well

Best is to use multiple small files grouped by theme or purpose.

Why this works:

• GPT fetches only what’s needed based on your prompt (e.g., hiring questions won’t trigger irrelevant product quotes)
• Easier for you to add/update one theme without breaking the others

Eventually, I ended up with a guideline of max of 20 files, with each file at 100-200 pages /25-35K words per file.

• Each file must be ≤ 512 MB  .
• Text/doc files are capped at 2 million tokens per file

(More on the 20 file limit and 2 million tokens per file are at the end of the article)

So Now my challenge was with all these video transcripts, articles, and books, how do i organize the files and with the format to train the GPT!!

So when the files are uploaded properly, 

• GPT indexes each file separately
• When a user asks a question (e.g., “What would Jobs say about launching MVPs?”), it:
  
1. Searches for relevant files (e.g., Design + Product + Risk files)
2. Extracts relevant quotes/snippets
3. Uses those to inform the response

➡️ Better-separated files = better semantic matches, fewer irrelevant quotes, and faster processing.

I was wondering what could I do to create the files in a better way (read as without going through the pain/pleasure of reading each and every line of the content again, write notes and organize it for the machine to understand)

That is when I realized what If I could use NotebookLM. So uploaded all the content I had to NotebookLM. Asked it create mind maps. It helped me to come up with the themes. Again fed the mindmaps to chatGPT and Gemini and asked it to build the themes that should be covered. Then I gave the themes and asked the chatGPT to create prompts for each theme to query in notebookLM and created theme based documents with both prompt and response which became the knowledge builder files.

From there, it was easy to build the GPT builder following the instructions and prompting the chatGPT on how it could be built. 

The shocking secret anatomy of a Dan Brown Novel

For a while, was resisting the temptation to read the latest Dan Brown Novel. His last one, Inferno, was a disappointment  and me in my Locked In Mode, felt I am better off reading non-fiction, and AI related stuff.  

Last Saturday night, woke up in the middle of the night and I really needed a break to take my mind off of things and thought, what the hell, let me give it a shot.  Though it was a usual Dan Brown, it didn’t disappoint and it was unputdownable and there in lies his genius. 

More than 2 decades back, I read Da Vinci Code in a single sitting and was so amazed for few days.  In fact for a while until i read the secret of the success by the Author Sujatha. He said, the success is due to the factor, he mixes the fiction and non-fiction in the right proportions and you as a reader at the moment of reading, cannot discern which is fiction and non-fiction. After reading this it was like, lifting of the veil/disappearance of the fog. 

Later, the more I read, I could realize and appreciate, how they have written the book, how they place the red herrings, how the chapter ends, how the characterization is done and so on. [Until recently, I didn’t realize or know on how to articulate this phenomenon /habit. Apparently it is called reading for form, most of us read for content, we don’t read for form or structure, not with the intent of oh, this is what the author is trying to do, basically, you don’t dissect the anatomy] [In fact, hated the Inferno, mostly cause, second chapter and I was able to figure out the antagonist and it was such a bummer to read]

[Oh, I thought, I figured out the plot twist at two places even here but he did pull a fast one or should I say Sidney Sheldon but it was a good sleight of hand]

However, when I read this latest one, though I could see what was happening, and how it could happen, it was intriguing and kept me on the hook. Though I didn’t finish the book in one sitting (Hello Old age, and other responsibilities of an adult) the book was indeed a good one and would recommend reading

But wait where is the Anatomy??

1. Robert Langdon should do swim in the first few chapters 

2. City as a character. More it is mystic and historical better. [I really do want to know, does he holidays in such places and comes up with the inspiration or does he decide, well, i want to holiday in the city so let me make it a center piece of my novel]

3. Pick up an organization that we all think we know but has more layers to it. Vatican, NSA, and CIA

4. Through out the novel, have few central themes that are focus of interest to public but are little obscure and controversial in nature. The old vs new, the new is not that new, but actually old. Fringe research, interesting anecdotes, trivia that nerds would know but make it mainstream. (For e.g did we really land in the moon, does moon exist, you get the drift]

5. Add conspiracy. Take science fiction and make it look it is plausible. 

In other words, I think the inner prompt he gives himself is, take equal proportions of Michael Crichton, Sidney Sheldon and Jeffrey Archer and dumb it down and take that would appeal/correlate to a large audience. 

After this out of curiosity did as the ChatGPT, “ What are the common themes across all dan brown novels” It is also a good read!! (I also think it is not trained on his latest book)

https://chatgpt.com/share/69076371-8838-8000-a4ad-5107e6e6fa9f

To buy the book: https://amzn.to/43NUxTw

How Mckinsey does it and what we could learn from it ;)

 "McKinsey today is itself truly a global firm, with a non-American majority controlling its shareholder committee and the real possibility of electing a non-American as managing director in the near future."

I strongly believe in two things.

One, History repeats. At the least rhymes. Hence it is important to read history and learn a lot from it which helps to predict the future. Sort of like Machine learning but in a different sense.

Two, No one knows the future. Prediction is futile. So is planning. (Yet, I also annoy some folks with my super optimized planning running different options and scenarios)

As a result, one of my favorite thing is to read archives. Old articles, old books. What better way to time travel. (or Old Movies)

To give you a sense about articles, articles that wrote the obituary of Steve Jobs, Articles that praised the likes of Elizabeth Holmes and Sam Bankman-Fried. HBR articles that talk about the might of Yahoo kind of ones are my favorites.

Well, one could argue, "Hindsight is 20/20" Journalists were writing what they say, it was the zeitgeist of the times. No one knew any better (well, thank you, exactly my point)

Neverthless, they are fun and helluva good reads and good intellectual exercises. 

Hence it is rare to read articles, that are balanced, nuanced and in general stand the test of time. (say except for few things here and there, things spoken in the article would be relevant even today)

https://fortune.com/article/how-mckinsey-does-it/ A fortune article on mckinsey written in 1993 is one such article which calls Russia, Eastern Europe, China and India as fledgling markets to Mckinsey. 

The opening lines of the post are from this article. If you haven't guessed it, in the election that happened in the next spring after this article was published, Rajat Gupta was elected as the MD. And yes, again, no where his name his mentioned in the article though it talks about the upcoming MD election and the front runners and favorites. 

One could also argue, the takeaway is not more about the good article but more about Mckinsey

Hello Instapaper, bye raindrop

Someone who knows me very well often tells me, "Hey you are like a machine! No feelings, nothing" At the same time the person also tells, "you are sentimental" I think, probably, more influenced by the "it is the best of times, it is the worst times" quote of Dickens.

Well, the reason I bring this up is, though I don't care about products getting killed and services getting shutdown, I am still reeling with sadness over the demise of Google Reader and Pocket.

(Also the algo change and as a result the user behavior on the platform. Twitter used to be a gold mine for me to discover and come across a lot of interesting articles)

I blame their demise to my digital addiction to Twitter, Insta and Youtube reels.

If only they had been around, I would have been spending more time reading rather than wasting time. 

Anyways, sometime back, I found raindrop.io as the alternate to Pocket. Even exported the pocket articles to Raindrop.

Somehow, never got to read the articles or use them. Nope, it is not that Twitter or Insta was the reason. 

I just felt the UI/UX was off. Just couldn't put a finger on it.

Thankfully, recently came across Instapaper. Interestingly, it is easy to add the articles. Easy on the eye and have been slowly back to reading. 

Oh and also they give 3 month free trial to premium if you export articles from Pocket.

How I wish, Mozilla had open sourced pocket which could be run locally.

Still I feel there are no discovery or recommendation services like the way Pocket did. 

AI Feature Ask

Though I very much like how the various LLM AI apps are outdoing each other with various new feaure launches, I wish they would make the following happen.

1. Ability to pin few of the following conversations.

I might have multiple conversations but few of them are the ones that I might go to often and would use it in a regular manner. It would be helpful to have this feature.

Teams has it, Whatsapp has it. 

Reason calling out that other apps have it is to drive home the point that it should not be that difficult to build it

2. Ability to respond to specific message

When I am having a long back and forth conversations in an app. There would be a need to go and refer one conversation or follow it up from there specifically. Though most of the apps remember context and pick up the cues instinctively, it would be good to have the ability to respond to one particular message and take it from there

3. Table of contents

There are times, a conversation is pretty long. Scrolling up and down and finding one particular thing becomes such a difficult thing. If there are long conversations, it would be good to create an automated table of contents sorta thing which helps in easier navigation.

Blast from the past: Tales of clearance

  With the today's twitter war on accusations by an importer, was reminded off the two tales I heard quite sometime ago...

I was staying in a bachelor mansion kinda place... (Kinda cause it wasn't a mansion mansion per se. Stories about this place is for another day.)

One person who was staying there with us was a customs clearing agent. One day, he had left a huge wad of substantial cash lying around unattended and was sleeping. (To be fair, amongst us, he always carried a huge amount of cash) (I always thought he was the highest earner as well, only later I realized he carried a lot of cash but not earned as much)

One of our other mansion mate chided him for leaving the money that way. The agent man responded, no big deal even if i lose, no one cares. it is all money that anyways, "he gives away" Intrigued me, ended up asking lot of naive questions about what he does how he does.

In simple words, he said, his job was to get the imported goods out of port at the earliest. Money was the best way to get it out. He also explained no one bothered as all the businessmen were focused was to get the goods out of the port at the earliest. More the inventory lied in the port, more the loss.(Then, little did i knew about the cash conversion cycle, inventory turn around and all that but that day, i learnt, faster they got the goods out, it is good for business) He also told me the money spent to get the goods out was literally nothing in a big scheme of things and the money was marginal. He also said the money "spent" to clear the goods was always reasonable as the people never wanted to kill the golden goose.

After hearing about all of it, The naive innocent me asked him a  follow up question: "How did people react to the Indian movie?" (Not the Indian 2, you can have a fair estimate of my age but you would be wrong) Was any one scared, worried? did anyone change?

He laughed and went on to tell me another story. 

One clerk got rich enough to buy a car in a super short time frame. He asked me to guess the quantum of the "scheme" It was one of the toughest case study question, I have come across. Later he revealed, the person was actually selling a form that was supposed to be given free for a single digit value. No one was aware that form was actually free or didn't give a damn as the amount was miniscule. Yet the numbers were so huge for the person to buy a car!!!

Amongst us was also a guy who worked for a large industrial house, who was supposedly known for their squeaky clean image. The industrial house guy was in finance and had to regularly deal with external government offices, clearing lot of approvals, papers, taxes and what not...

 The cynical me (Yeah am naive as well as cynical; go figure) turned to him and asked him, what about this firm? are they really clean or is it all made up stories, and I have never seen you carrying cash as this fellow?

He smiled and said, he doesn't carry cash cause they don't pay and the stories are true. I was like how do you get the work done? Is it because you are too big and powerful? do you deal with different level. 

He said, we flat out refuse. we make it clear, we will follow rules, if the officer wants to delay they could delay.

As a result, people knew they will follow rules and papers move. every now and then there will be a new officer who will try something funny with them. Either they will be patient or the new officer will be schooled by other colleagues that the tactics doesn't work for them.

Humanizing your AI created content!

 In the world of AI, you know what is worse than AI Snake Oil Peddling,

Clickbait articles with the above misleading headline which goes on to offer solutions like the following to humanize AI created content

1) Telling AI to write like Human

2) Using a AI Humanizer. Well apparently there are AI tools which will make your AI generated content more human like...

3) Don't use AI (I am not kidding, the author did recommend this, if you don't want your content to sound like AI, don't use AI)

Good Bye Pocket! Hello Raindrop!

If you have been someone like me, who is a voracious reader, information junkie and hard pressed for time, Pocket was a god send. Whenever you come across an interesting article, you can just save it to pocket, read it at leisure, refer it at any point of time. More than that, the recommendations based on the reading was spot on. If I have been sharing interesting articles, now you know where they came from. Beyond all of that, it also saved articles from sites that were behind paywalls. 

It was one of the app, that I used a lot and has helped a lot. So naturally I did really feel bad and upset when I read the news Pocket would be shut down. (The last time I felt this bad was Google shut down the reader.) I also felt bad, I didn't see so many people feeling bad about the shut down and writing ode to Pocket. I hardly saw only one post. (Well, that sort of explains the shut down)

Well, we are humans, we move on, life moves on. We start wondering, what next? what is the alternative?

To be honest, I was too tempted to ask one of the LLMs to help me build a chrome plugin or an app that would let me save the articles directly to Google Notebook LM, you know to be more AI powered and so on. After dabbling a bit on that, better sense prevailed and asked the LLMs for a replacement and alternatives.

After looking at Raindrop.io and Notion Clipper, I settled for Raindrop. 

If you just want to save links and read later, Raindrop is the one to go with. If you are someone, who want to save the articles, highlight and want the content, and have already been using Notion, then Notion Clipper is the way to go.

Oh by the way, If you have been a power user like me in Pocket, and have saved a lot of articles, then Pocket lets you to export the links. Not the entire content of the links. 

Once you export the links and download them, you could import in Raindrop and get to all those articles you are yet to read.

By the way, I had 10,000 unread articles and 9000 read articles in archive

N.B

Only humans were involved in writing of the post :) :)

One thing, even AI can't do....

Somewhere in the Ministry of the Indian Government, an AI service provider's salesperson was passionately selling the power of AI.

“Saar, it can do this, it can do that,” he said with conviction.

To the politician and the bureaucrat, it all sounded eerily familiar—like the lofty promises heard during election campaigns.

But this time, when they tested and questioned it, the AI wasn't just smoke and mirrors. It wasn’t a pre-recorded demo, nor was it cleverly disguised vaporware. Heck, it wasn’t even a staged live demo—it was real-time stuff.

Impressed, the politician and bureaucrat decided to throw the ultimate challenge.

“Can it fix all the issues in our EPFO site?” they asked.

And just like that, the salesperson vanished—much like a genie who disappears the moment you ask for the one wish that’s just a bit too difficult.

What I want from AI: Personalized Recap, So far

AI has been everywhere lately, claiming it can do all sorts of things.

Well, if there’s one thing I really want (okay, I actually want a lot of things), it’s this:

You’ve been watching a series on Netflix—or any other streaming platform. Maybe even a movie. You stopped midway, or after a few episodes, or perhaps after a season or two.

Now you’re coming back to it. All you want is a recap of what’s happened so far.

Think about it: a customized, personalized recap based on your viewing progress.

How hard can it really be for built-in AI to do this?

Come on, streaming apps—rather than focusing so much on what to show me next, how about reminding me what I’ve already seen?

Paradox of AI

“A good fiction will blend a lot of reality and facts into its narration and storyline. The blend would be so fine, it would be difficult for one to differentiate fact from fiction.”

A famous Tamil writer made this remark while commenting on the popularity of Dan Brown’s The Da Vinci Code.

If one were to paraphrase, the current state of AI is just like that. It’s extremely good at certain things. It’s also extremely not so good at certain others. The challenge lies in figuring out when it’s good—and when it’s not.

Some time back, my dad wasn’t well, and I had to take him to the doctor at night. An ECG was recommended. It was taking some time to meet the doctor, so I uploaded the ECG. Voila! I had the results and a detailed reading almost instantly.

Recently, my uncle had to visit the doctor alone and came back with an X-ray. Since none of us had accompanied him, we wanted better insight into the diagnosis. I uploaded the X-ray, provided some context about him, and again—boom. It gave all the details and even asked, “Would you like me to explain this in simple terms for family members?”

Immediately, my aunt asked: “What will happen in the future? Will any of us even have jobs?”

Now, let’s switch to a different example.

I often give the New York Times Connections Puzzle to the leading AI models. It has 16 words, and the task is to group them into four correct clusters of four related words. In a sense, it’s a very simple puzzle. (Well, not that simple—at least to me—since I only have a 40% win rate.)

Nowadays, I regularly feed these puzzles to AI, and to my surprise, the win rate is often worse than mine. To be fair, the way humans solve it is quite different from how AI tackles it. But still, ideally, you’d expect AI to crack it easily.

So, it can decode an ECG, analyze an X-ray… but fails consistently at a Connections puzzle?

Herein lies the rub.

If you give AI enough sample puzzles along with the correct solutions and then ask it to generate new puzzles, it can churn out hundreds of them in no time. But even then, there’s no guarantee it would be able to solve new ones correctly 100% of the time.

So, where and how should one use AI?

Let me walk through another example.

I was once tasked with writing a report on the evolution of a domain/industry: its origins, development, current state, challenges, and future trends. (I’m very familiar with the domain and could speak about it without much prep.)

My traditional workflow would have looked like this:

1. Understand the problem.
2. Do a deep Google search.
3. Find and read at least 10–15 relevant reports or sources.
4. Highlight important points, copy-paste excerpts, consolidate notes.
5. Build a point of view.
6. Draft the narrative and create a document.

This would usually take 2–3 days and might even need two people, depending on complexity. The final report would be 5–20 pages long.

Now, with LLMs:

1. I think more deeply about the problem statement.
2. I write a good prompt, provide relevant context, and use a few prompt hacks.
3. Within 2–3 minutes, I get a 10-page draft.

I usually run the same prompt across 2–3 LLMs and get multiple versions. I then feed them all back into an LLM and ask it to consolidate.

Now, in less than 30 minutes, I’ve reached the “consolidated draft” stage.

All that’s left to do is read through it, check for hallucinations, customize the tone and structure as needed, and send it off. If I want to do a really good job, it takes just 2–3 hours. At most, 4–6 hours.

That’s the productivity gain.

Now, if you reflect on this, here are some key takeaways

1. You need to write a good prompt.
   It’s garbage in, garbage out. Writing a good prompt is easier said than done. Beyond prompt hacks, you need to really understand the problem, the workflow, and the expected output.
   You can’t automate something you don’t understand. You can’t use AI effectively if you don’t know what you want—or if you can’t instruct the AI clearly.
2. It can boost productivity like crazy.
   But again, only if you know what you’re doing and what you want.
3. Watch out for hallucinations.
   Unless you know the content cold, you might not realize where the AI has gone off the rails. Sniffing out those hallucinations requires real subject matter expertise.
4. Long story short:
   AI is a great mimicking engine.
   It is not a substitute for original thinking.
   (One could argue it helps in idea generation or acts as a sparring partner for brainstorming—and yes, it does. But even then, you won’t get the best out of it unless you know how to evaluate and refine the options. It aids thinking; it doesn’t think on its own—yet.)

As one leader put it: “It can generate 90% of an investment prospectus.”

But the crux—and the criticality—lies in the remaining 10%.

And that’s where the human element remains irreplaceable.

So, to answer my aunt:

Yes, humans will still have their jobs—at least for now.

P.S.

These days, after I write a post like this, I feed it into AI, ask it to fix grammar, smoothen the flow, and polish it while keeping the tone intact.

One of my editor friends—the kind who fixes your grammatical errors on WhatsApp—recently told me:

“Your writing has become more polished lately.”

I don’t think the writing would be this clean without AI’s help.