Pratik's Year in Review: 2025
adventure of a lifetime
Every December, I sit down to write one of these. Some years, it feels like a chore. This year, it felt exciting.
2025 was a year of experiments. With work, with health, with relationships, with how I spend my time. Some worked out. Some didn’t. I did more interesting things across the spectrum than I had ever done before. I traveled, I built, I failed, I learned. I also slowed down enough to notice what actually makes me happy.
Here’s what I learned.
My Experiments in Vibe Coding
Master design, orchestration, planning, swarming, and code health. Become a “100x-1000x” super-engineer capable of owning an entire company’s software stack.
Let’s talk more about it. 2025 was a game-changer for me, like others. Coding happened at the speed of thought. I have been using it in an end-to-end manner across the board, from deciding what to build to how to talk about what I build. I think the Claude code team has done a fantastic job. Recently, I have fallen in love with Claude Code’s skills and subagents. I have been treating it as a natural language program, and it can run long jobs without significant failures. Skills are the hot-swappable LoRA adapters that require no parameter training.
What surprised me most was being able to run multiple tasks simultaneously. You keep evaluating outputs as they come in, nudging them forward. You make them work while you have lunch. While you go out. While you sleep. I’m not waiting for code to compile anymore. I’m not blocked on a single thread of work. I’m orchestrating. Or at least trying to.
One workflow I automated that actually works well: research paper summaries on Maxpool. I dump the link to the HTML or PDF on arXiv, and it creates the summary in my preferred format. Makes it easier to absorb key insights from any paper.
I’m getting high-quality ideas every day while building in the flow. I feel like the founder of a company called “myself”. I no longer think of myself as a developer/researcher/writer. I’m a wrapper of specific, wide, and deep skills that keep changing every day. Still figuring out what that means.
I personally made level-up progress through three things:
Mindset: AI cannot replace me.
Focus: Master one tool per category instead of trying many.
Apply: Take on a hard problem
My next year’s goal is to further increase the width and depth by 2-3x. I’m interested to see the fundamental limit of scaling our augmented intelligence. I can already feel it when I try to work on many things at once:
Reading speed
Ability to hold context (short-term memory)
Long-term memory
Alone, you go fast; together, you go far. But when AI can work together with you, there is no limit. Keep delegating and collaborating. There is no wall.
mental model for success = parallel x ultrathink x subagents x self eval x self consistency x progressive disclosure x context offloading x tool retrieval
Experiments with Maxpool
This year, I took a big bet with Maxpool: migrate the community from WhatsApp to Discord and build what I hoped would be the world’s first agent-driven AI community.
The vision was ambitious. We had nearly 900 members on WhatsApp, approaching the platform limit. Discord offers organized channels, a searchable history, and, most importantly, the ability to integrate AI assistants directly into conversations. We built Max, a custom bot designed to help members find resources, answer technical questions, and enrich discussions.
The migration was harder than I expected. Of 500 active members, only 50 voted for Discord. We spent weeks on personal outreach to bring early adopters. Some people joined, fewer stayed active. Getting critical mass on a new platform is brutal. The energy that flows naturally on WhatsApp doesn’t just transfer over. So we rolled back the transition.
I still believe Discord is better if we can get the critical mass. The structured discussions, the voice channels, the bot integration. It’s all superior. But communities aren’t built on features. They’re built on habits. And habits are hard to change.
We expanded the admin team this year. Ankur, Nico, and Shailen now help run different groups. I’m still searching for more admins. If you’re reading this and want to help shape Maxpool, reach out.
New Maxpool website: I added learning content, research paper summaries, and resources that anyone can reference. I genuinely enjoy reading and writing research papers. It’s become my personal library as much as a community resource.
We had our own little year in review → https://maxpool.dev/year-in-review/2025.html
My Best First Experiences This Year
This year was filled with many first experiences. Some are mainly due to visiting the US.
Attended concerts of artists I’ve loved for years: Coldplay and Lucky Ali
Visit to SF, NY, Chicago and LA - had my first super ride experience at Six Flags
Saw dinosaur fossils
International art in Chicago
Giraffe, penguin, polar bear and zebra
Yosemite hiking with Tesla autonomous long drive & close-up view of a rainbow
First hackathon win in the US
It was a pleasure to be invited to talk on Latent Space - the #1 AI Engineering podcast in the world.
I wish to do more of this next year. Speaking forces you to clarify your thinking in ways that writing cannot.
That last one deserves a moment. As you are aware, I started writing books in 2024. My new and fourth ebook, Mastering multi-agent systems, turned out so well that we decided to print it. And I must say, there’s something about having a physical version of your book. Aishwarya gave us a nice shout-out after reading it!
Favorites of the Year
Movies
Avatar: Fire and Ash 10/10
Predator: Badlands 10/10
F1: The Movie 9/10
Materialists: 8/10
TV Series
Billions 10/10
Books
Fiction: Death’s End (Three Body Problem Part 2) ( great plot but not so great writing )
Startups: The Nvidia Way
Evolutionary biology: A Brief History of Intelligence
Growth: 5 Types of Wealth
Spiritual: Osho’s Courage, Awareness
Technical: Building an LLM from Scratch
Business: Reshuffle
Health & Workout
Runs, walks, injuries. The cycle continues. But I’ve learned that consistency matters more than intensity. Show up, even if it’s just a 20-minute walk. That way I managed to cram in a lot more than I would have otherwise. Everything gets easier if you can put a number on it, and having a tracking watch helps tremendously.
I started walking consistently and optimized for volume rather than intensity. I do walking as meditation at times, but I also enjoy talking with my parents when we’re going together. All in all, I did mild or medium effort exercises on about one-third of my days, but I had large gaps due to minor injuries in my knee and feet. Otherwise, I maintained a consistent streak.
I tried more experiments with vegan protein this year but nothing suited me. Different powders, different sources, different combinations. My body just didn’t respond well. I’m falling back on eggs and paneer for now, in limits. Sometimes the answer is simpler than you want it to be.
Experiments with Sleep
Sleep is the single most crucial thing worth optimizing for mental and physical health. I was sleeping 7 hours every day and still felt tired. I knew something was wrong with my sleep quality.
To improve my sleep quality, I started taking Melatonin once or twice a week. It has been great so far and I wake up much fresher. Melatonin is not a sleeping pill in the technical sense. It’s non-addictive, non-resistive, with no side effects. A good sleep now costs only ₹4/night for a 3mg tablet.
This year I woke up at 5 AM on more days than ever. I consistently woke up at 7 AM most days. I want to have more 5 AM days in 2026.
Big Ideas from This Year
There is no wall.
It’s better to die of poverty than boredom.
If AI is gonna make money meaningless, what should we accumulate?
It is hard enough to have a normal life.
Life is what happens to you while you’re busy making other plans.
What is the maximum I can spend and get wealthier?
We make decisions and then rationalize them.
How to develop a taste for engineering?
Do not rely on big things to make you happy. Because it only gets harder. And that just makes happiness difficult.
Build rejection resilience: Nothing wonderful happens without enduring rejection repeatedly.
Find a role model.
Accept discomfort and rejection as the cost of growth and upside.
Our interactions with the world and the problems we solve are becoming increasingly complex. This requires 1 hour a day, 1 day a week, and 1 month a year of a complete distraction-free environment to reflect and brew ideas. If money can’t buy us this time, it will kill us.
Advice I Would Give to Young Myself
Seeing what others cannot do comes with a mix of heavy immersion and intentionality.
Some engineers struggle to secure interviews, while others juggle multiple job offers. AI is widening the gap between average and exceptional talent.
Create in public. Share what you are learning, building, or thinking about. Even small, unfinished ideas can spark a connection.
Talk to people outside your immediate circle. Many of my best introductions came from “weak ties”: friends of friends, or connections rekindled from sharing content online.
Say yes to low-cost, high-surface-area experiments. Join a panel. Write a short post. Help a founder friend test their product. You never know which one will lead to a door opening.
Create a mindset shift from living in scarcity to living in abundance. Get into jobs/roles that can help you accomplish more with AI. Then your only bottleneck is your creativity and ability to connect dots that AI cannot.
Look at progress in these terms:
New experiences
New skills
New ways of thinking
More of what I already have
Transform every 3 years
Learnings from the Agent Leaderboard
Evals are messy. It’s like no matter what you try, you’re gonna miss something. Making something as realistic as real requires real-world interactions, which are a bottleneck and costly.
This year, I published another benchmark on evaluation - Agent Leaderboard. As I worked with agents, I learned a great deal about their limitations and how to address them. Agents are impressive until you try to make them reliable. Then you realize how much work remains. The start is just the tip of the iceberg.
Benchmarks I enjoyed
Efficient learning: Solving novel problems efficiently without prior training | ARC-AGI
Software engineering: Resolving real-world coding issues in large codebases | SWE-Bench
Long context understanding: Deep reasoning across documents ranging 8k to 2M words | LongBench v2
Computer use: Operating GUIs, terminals, and multi-app workflows in real OS environments | OSWorld
Instruction following: Generalizing to unseen constraints without overfitting | IFBench
Coordination: Collaborating with others while considering their beliefs and intentions | LLM-Coordination, MultiAgentBench
Persuasion: Convincing others and resisting manipulation across multi-turn dialogues | Persuade Me If You Can
Research synthesis: Multi-step web exploration and analyst-grade report generation | DeepResearch Bench
Running a business: VendingBench tests whether agents can manage inventory, pricing, and customer interactions in simulated business scenarios. It’s the closest we have to measuring entrepreneurial reasoning.
Upgrading My Setup
My setup was already working fine, so these were more nice-to-haves than necessities. Got a monitor arm for my screen. Honestly, not much utility, but it looks cleaner and I like it. Small wins.
I don’t like bringing my laptop onto the table for meetings since I keep it off to the side. So I finally got an Insta360 Link 2C 4K webcam. I’d been delaying this for years, hoping something better would come along.
Switched to iPhone 17 Air from my 13 Pro Max. The weight difference is real. Weight, like inference speed, is a great differentiator. People worried about the battery and camera, but I have no complaints so far.
I will get Hue lights someday. Soon, a better mic. I have lost interest in earphones.
My current setup:
MacBook Pro M4
Samsung’s Odyssey G9
Apple keyboard / Lofree Flow84 / Keychron K6
Logitech MX Master 3S
Marshal Emberton
ErgoYou sitstand desk
Jin Office White Monitor Stand Heavy Duty for 13”-49” Inch
Cooking
Eating outside in India is risky. Too much oil, questionable hygiene, you never really know. And I always thought it was too late to learn cooking.
This year, I decided to try anyway. Made deliberate efforts to prepare things I actually like to eat. Pasta, macaroni, veg grill, egg items, garlic rice, paneer curries, noodles, and manchurian. Nothing fancy, just the basics done at home.
It’s still early. I enjoy having someone to cook with; it makes the experience more enjoyable. But here’s what surprised me: cooking doesn’t feel terrible. It’s actually a nice escape from screens and technology. You can’t multitask while chopping onions. Your hands are busy. Your mind gets a break.
Still figuring out the rhythm of it. But I’m glad I started. I know this skill is gonna age well.
Why Life Gets Bitter?
This year, I’ve been developing a mental framework for evaluating the year: forget goals, metrics, or achievements. Just count the days you felt genuinely great, whether from a planned milestone or a random, goalless afternoon where everything just clicked.
A good year = having enough great days, minus the truly bad ones.
The elegance is in its honesty. It measures how we actually felt, not what we accomplished on paper.
Tally from the year:
Great days: 30-40
Terrible days: 5-10
The verdict: positive
Some Refinements I’m Considering
As I’ve thought more about this framework, I’ve realized that days aren’t equally weighted. One devastating day can overshadow ten good ones; one transcendent day can redeem months of ordinary ones. Simple subtraction might not capture this asymmetry.
There’s also the matter of the ordinary days. If 15 great days and 10 bad days account for less than a month of my year, what about the remaining 340 “regular” days? The texture of these ordinary days, neither great nor terrible, might define the year more than the extremes.
And of course, memory is selective. We tend to remember peaks and endings, not the sum of all moments, so my count is already a curated highlight reel rather than an objective truth.
So instead of tallying days, I’ve been asking myself a different question: Would I choose to live this year again, knowing everything? This captures growth, quiet contentment, and things that don’t register as “great days” but still make a year feel worthwhile.
I started thinking about this after a conversation with friends about how life gets bitter as we age, how regrets accumulate and happiness becomes harder to maintain. The traditional metrics we use don’t actually measure what matters: did we enjoy being alive this year?
What struck me is that many of my genuinely great days had nothing to do with achieving goals. They were unexpected: a perfect morning walk, an afternoon where work just flowed, a sunset walk where everything felt right. These weren’t planned. They couldn’t be optimized for. They just happened.
Research in behavioral economics reveals that we quickly adapt to positive changes (hedonic adaptation) but tend to remember negative experiences vividly (negativity bias). One bad day can outweigh multiple good ones by a factor of about 2.25, according to loss aversion studies.
Why not optimize for peak experiences, those great days that create lasting memories?
I started asking myself each evening: “What would have made today great?” Sometimes the answer was small: I should have taken a walk, called a friend, or stopped work earlier. Sometimes it was bigger.
What This Means for Next Year
If I take this framework seriously, it suggests a different approach to planning. Instead of setting achievement goals, I should be engineering conditions for great days:
Build in more slack time (great days often emerge from unplanned moments)
Say yes to more random invitations (unexpected experiences)
Create regular rituals that have high “great day” potential
Eliminate recurring sources of bad days where possible
I need to control the inputs and the score will take care of itself.
An Invitation
I’m curious about your own year through this lens. If you counted your great days and subtracted your terrible ones, what would the tally be? More interestingly, would you choose to live this year again?
Perhaps the real measure of a life well-lived isn’t our accumulated achievements but our accumulated great days. And perhaps the way to avoid the bitterness of aging isn’t to minimize regrets but to maximize the days that make those regrets worthwhile.