My Top Stories of 2025
A year of AI, decisions, and building in public. Plus, your subscriber-exclusive toolkit for 2026.
Happy New Year, everyone.
2025 was wild. Personally, professionally, technologically.
My challenge by the end of this year is to make peace with a year that refused to be predictable.
And after I dream of having a full cycle of sleep and meeting 2026 with open hands instead of clenched fists.
I turned 30. I built companies (more companies). I watched AI go from “interesting tool” to “wait, this changes everything.”
I wrote more than I’ve ever written in my life (still don’t know if I’m like this, but just being honest)
Looking back at what resonated most with you, I see a pattern. You wanted frameworks, not bs. So here are the 10 pieces that defined this newsletter in 2025. The ones that got the most views, shares, sparked the best conversations, and still hold up.
But before that, I also wanted to share with you something important. For 10 years, Michael and I built Belkins’.
We never sat down on camera to talk honestly about what building this company has really cost us, so here is something special for you. We recorded this podcast to share our wins and regrets, our mistakes and our ambitions.
and now let’s revisit this year.
Let’s go
The Top 10 of 2025:
What I’ve learned that made my life better
I turned 30 last September. Celebrated in the Cotswolds with my wife and friends. Tennis. Great food. Nature. The kind of weekend you actually remember.
To mark it, I wrote down 50 things I genuinely believe after a decade of building. No generic advice. Real learnings from honest mistakes.
This one is personal. If you read one thing from this list, read this.
Sub Agents
Hey, As you probably already understand, I'm documenting my learning curve with AI week by week. I'm keeping you posted about everything I learn, see, or have success with.
The frontier of tech just shifted again, and most people haven’t noticed yet
Anthropic released sub-agents for Claude Code. The technical community barely blinked. But this is a game-changer.
Think of sub-agents as specialized AI personas. A dedicated architect for system design. A specialist debugger for fixing errors. A growth hacker for marketing. Each has its own context window and laser focus.
I shared my entire sub-agent arsenal, the YAML configs, the invocation patterns, and the chaining prompts.
Sub-agents solve AI’s biggest weakness, context pollution. Most people are still wrestling with one AI that forgets what it was doing.
You don’t have to.
The Great Restructuration
Last week, I had coffee with three friends who all got laid off from major tech companies.
Why corporate America is about to get smaller and 10x more profitable
Three friends got laid off from major tech companies. Different roles, same story: “I honestly don’t know what I accomplished in my last two years there.”
That’s when it hit me. We’re not watching random cost-cutting. We’re watching the most systematic corporate evolution since the assembly line.
This isn’t about layoffs. It’s about the death of the “professional meeting attender” as a job category.
The year superintelligence stops being science fiction
“First, it happens gradually, then suddenly.” Every technologist who has lived through a platform shift knows this feeling.
I mapped out the timeline. The compute curves. The capability jumps. Why 2027 matters.
AI Generalist
This is a paid subscriber exclusive. If you're reading this, you're one of the first 200 people to access the career opportunity that will define the 2030s.
Playbook on how to make $300K+ while everyone else fights for scraps
This was for paid subscribers only, and it remains one of the most practical career pieces I’ve written.
The specialist era is ending. The new valuable skill is orchestration, knowing enough about many domains to coordinate AI across all of them.
You don’t need to be better than AI at any single thing.
You need to be better than AI at connecting things.
Why People Fail to Make Important Choices
Today, I was inspired to share my thoughts on Why People Fail to Make Choices. Important Choices. I’m pretty sure that everyone who will read or is reading this is fighting with their own decisions and variety of choices.
Decisions and destiny, or how our choices define us
One of the older pieces on this list, but it keeps resonating.
As a CEO, I always have 10-20 important decisions on my plate. The most challenging part is not knowing whether they’re good or bad in the moment. I only understand after 3-12 months.
Not deciding is itself a decision.
Usually, the worst one.
Success with AI is Probabilistic and Not Deterministic
I watched my friend founder blow $50,000 on AI tools last month.
Why the best AI users think like poker players, not chess masters
I watched a founder blow $50,000 on AI tools. Not on API costs. On the wrong mental model.
He treated AI like a vending machine. Insert prompt, receive output. If the output was bad, the machine was broken.
He never pressed the button twice.
Good prompt + Good model = One sample from a probability distribution. Not a guaranteed output. You have to iterate.
7-by-7 upgrades for sleep, food, movement, work, and life
I asked my AI agents for the most efficient, proven ways to improve quality of life. No philosophy. Just levers.
7 for sleep. 7 for food. 7 for training. 7 for work. 7 lifehacks.
You don’t need 100 tricks. You need a small number of system changes that quietly run in the background.
When Tech Giants Become Feudal Lords
Is capitalism dead, replaced by a digital feudal age? Platforms own the digital land, the app stores, the feeds, and the cloud. The rest of us are tenants.
Your followers, your users, your data, they feel like yours. They’re not.
You’re building on rented land.
Instances
Every single interaction with AI is a dice roll. A cosmic lottery where the same prompt can get you gold or garbage, genius or gibberish.
The hidden truth about why AI gives you different answers every time
Every single interaction with AI is a roll of the dice. A cosmic lottery where the same prompt can get you gold or garbage.
I generated 847 vocal stems for three techno tracks. 23 were technically usable. 6 fit the vibe. 2 made the final mix.
That’s a 0.2% success rate. And that’s how AI actually works.
Most people generate once, maybe twice, and accept whatever they get. They’re leaving millions on the table.
2026
The newsletter will continue to do what it does: find the overlooked angle, share what I’m learning, and provide you with things that actually work.
Thanks for being here.
Thanks for reading.
Thanks for the replies, the shares, and the conversations.
Let’s make 2026 count.
Vlad