Notes and Thoughts - June 27
From my reading list, bookmarks and some notes for my own self.
I consume a lot of content, bookmark tweets and I drop the good ones into my knowledge agent but I realized something interesting today.
The realization is that as much as I want to read and discover, I might not be learning (registering) everything - because I don’t write as much.
As the knowledge base gets smarter with semantic search and surfacing the best insights for me, I now have all the information whenever I need it but I have been forgetting a lot of things.
I was speaking to Vikas recently and he mentioned how I used to always carry my notebook everywhere and used to write down everything we discussed, debated or argued about.
Today, I sent another reminder to Abhijit Khasnis to restart publishing and give us a 2026 version of his 2016 short essay - The Perils of Personalisation.
It was originally published on 8 July 2016, 10 years later, the planet is completely consumed in short videos, personalised feeds and AI generated content.
On the other side, I noticed a lot of people talk about never being able to come back to their bookmarks on X/Twitter and some folks have tried agents to scrape through their bookmarks and summarise/analyse or whatever.
I identify as a heavy note taker and my pronouns are he/him
Today, I am trying this new thing inspired by a section in Ben’s Bites by Ben Tossell.
Curating my Twitter bookmarks, discoveries and thoughts into a weekly note.
This way, I’m hoping I can spend some time going back to everything I found interesting and write about it every Saturday so I can remember better.
Personal AI
I have been spamming all my socials talking about Personal AI and how I believe proactive intelligence should work for consumers. My original pitch deck and UI for Seek in 2021 was shockingly similar to what ChatGPT Pulse looked like.
ChatGPT Pulse is being sunset now.
This tweet from Olivia Moore was an amazing validation to my “thesis” in personal AI.
I believe Personal AI will become a default for every consumer, like there was a time we used to ask our friends - BBM pin? Are you on WhatsApp?
Messengers and Chat platforms existed before WA, but messaging (aka SMS) evolved into its most powerful capabilities when WhatsApp brought everyone into the ecosystem with a single unique identifier they already shared - the phone number.
On the work side, we got nifty tools like Calendly for faster scheduling.
In a meta view, you could say everyone has an email address, phone number (also called WhatsApp number) and a significant percent of business folks would have a calendar appointment system or link.
I am sure we will be asking this very soon - Do you have an agent?
Let’s hand this off to our agents, to find time, schedule meetings, order food, and what not.
I think personal agents will need their own unique identifier like UPI/Email/Phone so I can
The culture shift is visible in circles where people text before they call.
We will soon have our agent talking to their agent, before texting or calling.
Here, Sajith Pai talks about an agentic interaction between Founders and VCs.
Interesting insights on Claude User Behavior
Sovereign AI + Frontier News
China has swiftly moving from being the cheapest fastest manufacturer of things to being the cheapest fastest model provider. Their frontier contenders are getting really good. But, looks like the copying, stealing, reverse engineering is not limited to small players - this level of accusation on a player like Alibaba is HUGE.
Frontier Agents
I don’t have a better name to differentiate between Consumer AI tools and deeper frameworks designed for the savvy users like OpenClaw or Hermes.
Calling them Frontier Agents for now.
I have been playing a lot with personal agents from the lightest consumer tools to the most powerful harnesses like Hermes/OpenClaw.
Hermes launched Mixture of Agents. These concepts are similar to LLM council by Andrej Karpathy but the possibilities are multiplying.
This is appearing to be the strongest direction for independent agents, avoiding dependency on frontier labs, using multiple models for the same task - broken down for each model’s strength.
I call my own setup “Supernova”, it gives me a Kanban view, ability to assign tasks to models and also handles the jugaad for multiple accounts, rate limits and utilising them well.
What I’m Exploring
Vercel introduced eve, a Nextjs like framework for building agents. I tried it, loving it so far - built a small replica of Seek (the personal agent harness) with eve but I am still experimenting on whether the final infra should be on Cloudflare or Vercel since Cloudflare recently granted us $10K in credits.
In May/early June, I built a small Agent for my brother, it analyses business CIBIL reports and identifies actionables, issues to fix, proactive credit management tips, etc.
Spoke to Vikas Malpani about it on the same call and he suggested this is a painkiller, you should ship it.
Found a few companies doing it, didn’t prioritize it for Seek yet but it is amazing to see CRED and Oolka doing this for personal credit reports.
It’s amazing how AI agents have given us capabilities to build small personal tools over a weekend that can really be turned into a business if you go deeper.
Other Stuff
I believe the biggest mistake I have made in the last few years is not sending regular investor updates - everyone around me says you should definitely send one, even if there’s nothing meaningful to share according to you.
Someone who gave you money, deserves to know what you’re doing; at least until you haven’t called it quits.
I hope I’m able to drive an outcome like TJ Parker that forgives the radio silence.
June has been a good month for talking about Seek publicly and I hope to restart investor updates to existing and prospect investors soon.
Kunal Bahl wrote a very insightful piece that served a good reminder for how I should be building Seek this time.
I’ve learned this the hard way and the popular term now is “wedge” - build one thing and do it right, then expand into your larger vision.
Sairee Chahal told me something similar when I first spoke to her in 2021. She felt Seek Rewards was trying to solve too much at launch and suggested I should focus on just one thing, one metric and prove it before expanding.
This week, I also spoke to Aryan Mulchandani, CFA after he published his notes on rewards and loyalty space in India. Great insights, we will hopefully discuss and write more on this soon.
Most interesting discovery of the week, must have DM’d this to 20 people I know. Thank you Wouter.
Saturday Project
Hired an AI Content Creator at Seek, would you listen to her?
I don’t have closing thoughts on the note but a few things I am still thinking about as I end this:
Will people someday trust AI Content Creators the same as a human marketer?
I am fighting hard against using a Chat UX in Seek’s Personal AI and some friends have been disagreeing - are you okay with all AI experiences being chat first?
Last week, I started writing about my story and work so far and the origins of Seek but haven’t been able to complete it - Why is it difficult to write about a single thing in depth compared to writing about 10 things in short blurbs?
Thank you for reading, skimming, scrolling - see you next week (hopefully).