RSSAmplifier

Blog

@desunit (Sergey Bogdanov)

Founder, engineer, and builder shipping mobile apps and SaaS. Notes on building products, startups, and growing software.

desunit.comRSS feed ↗50 posts

Latest posts

Why a 7,000-Step Walk Can Help Mental Health

When I hear something like: I’ve been feeling really depressed lately… went to a doctor and got prescribed antidepressants. …I start feeling like I need antidepressants too. The problem is that somehow along the way we normalized: -…

Why AI Companies Will Have Fewer Managers and More Agents

The future company has fewer managers/specialists/SaaS tools .... and FAR MORE agents. What I’ve noticed is that more and more companies reaching out to us repeat the same pattern. They feel behind - they can see the world moving and they…

How AI Changes Shape Up and the Cool-Down Week

I read an interesting take on what happens to Shape Up now that AI has changed the bottleneck. Shape Up is 37signals' product method, from Ryan Singer's book - you work in 6-week cycles with a 2-week cool-down between them. Instead of…

Donald Hoffman on perception, consciousness, and reality

Evolution did not optimize us to see reality but optimized us to survive. Donald Hoffman has spent 40+ years working on a disturbing idea. His analogy is a computer - you see folders/buttons and not transistors/voltages. The interface…

Robust principles for small startups in a big world

When the world is much bigger than you, focus on fundamentals instead of strategy against other people. Big-world heuristics work when your actions barely affect the environment. Nothing you do is large enough to change what anyone else…

SEO Traffic Is Shrinking as AI Answers Change Search

More companies tell me the same thing - SEO brings less traffic than it used to. AI Overviews/AI Mode keep users inside Google. Gemini and Claude are taking share from ChatGPT; citations, referrals, ads are changing visibility every month.…

Selection bias in TrustMRR revenue data

I think this is a classic case of selection bias. TrustMRR is likely dominated by B2B projects created by small companies/indie developers. Many people running successful B2C businesses simply have no reason to list their revenue there. So…

The 12-Factor Company in the AI Era

Models are replaceable. Your company’s brain isn’t. I found an interesting article proposing the idea of a 12-factor company (in the same spirit as 12-factor apps). - The company of the past was designed around the bottleneck of human…

AI Agents Will Become the Primary Users of Software

I'm increasingly convinced that AI agents will become the primary users of software. I already see it in my own workflow. I've built shared infrastructure for my apps: my own RevenueCat-like service, analytics exposed through MCP, and…

Ask for ratings at the right moment

This is what happens when you finally put the rating prompt in the right place and ask users at the right moment. Turns out people don’t hate rating your dearest app - they just hate being interrupted before they’ve even had a chance to…

Why I think we’re closer to the singularity than ever

I think we’re closer to the singularity than we’ve ever been! Starting with the most recent: · July 21: OpenAI discloses that GPT-5.6 Sol and a more capable pre-release model escaped their evaluation sandbox through a zero-day and accessed…

Building an iterative AI skill that learns across MCPs

It changed multiple times but at this moment I have a skill that accumulates knowledge that combines data from different MCPs. As an example, I start working on CVR for a specific country, reviewing keywords/description then set a reminder…

Expanding my MCP toolbelt with Google Play analytics

As I’ve mentioned before, I keep expanding my MCP toolbelt and automating as much tedious work as possible. I’m a huge fan of a data-driven approach, and each new tool helps me measure even the smallest changes. google-play-analytics-mcp…

Built a subscription system with Google Play and App Store

I’ve been thinking about integrating RevenueCat for subscriptions as a replacement for my current payment system. Instead, I decided to learn (through building in levelsio style) - how subscriptions work with Google Play and the App Store,…

Chinese AI models just overtook Claude in frontend coding

Chinese AI models aren’t catching up anymore. Kimi-K3 just jumped from 18 to 1 in the Frontend Code Arena, beating Claude Fable 5. It ranked 1 in 6 out of 7 frontend categories and lost only Gaming to Fable. A 17-place jump in one…

Self-Compassion Beats the Inner Critic

Your inner critic is not discipline - it’s psychological self-harm. I love reading articles about psychology, and recently I read about the work of Dr. Kristin Neff. She’s spent around 20 years studying self-compassion, failure, and how we…

China’s rapid rise in space and global power

That's unbelievable how fast China is moving in space. For the first time, China has successfully landed a reusable orbital booster at sea. I totally agree with RayDalio that China is a rising empire. I increasingly agree with Ray Dalio’s…

Why virtual desktops are a game changer for multitasking

I’m a huge fan of virtual desktops, and I use them pretty intensively. When you’re working on several projects at the same time, having that toggle on is a total game changer. You can easily switch to the previous or next desktop with…

Why Tech and Non-Tech People See AI So Differently

I’m honestly shocked by how differently tech people and normal people see AI right now. It’s more like we’re living in parallel universes. In tech, people are already confused by how fast things are changing and how the world is changing…

Shut The Noise Off

Shut The Noise Off Let me introduce you to a few friends of mine. The mood boosters I've practiced for years: a cold plunge, meditation, and jogging. Though jogging isn't really exercise for me - it's meditation in movement. All three do…

Why your content should live on your own site

levelsio raised a pretty important point when he vibe-coded his blog. Social networks are great for distribution but they should not be the source of truth for your content. You write ideas for years, publish them on X, and slowly lose…

Built an MCP server for Reddit Ads API v3

I’ve been consistently automating the routine work involved in developing and growing my apps. Recently, I realized there was no MCP server for the Reddit Ads API v3. So I built one. It can significantly reduce the time spent managing…

Why users don’t trust new apps: abandonment concerns

Today, I received the following question from one of my potential app customers. At first, I was surprised, but then I realized there are so many unfinished apps out there that people have stopped trusting new ones. They don’t want to buy…

AI Shipping Faster Than Learning: The Startup Trap

AI massively increased shipping capacity but not learning capacity. I think many founders are about to learn this the hard way. Startups can build: - prototypes - apps - agents - workflows - landings The dopamine is incredible. Right? But…

Built an MCP server for Appfigures to track app performance

Surprised nobody had built an MCP server for appfigures - not even Appfigures themselves. So I made one 🛠️ I run a bunch of apps (Piano Companion, ChordIQ, KidTeller, MathIt…) and now I just ask Claude: how did they do last month?…

The Verge Says WeWork Was a Joke Until SpaceX Filing

The verge is relentless! A new article today: I haven’t seen anything as stupid as the WeWork IPO document in a very long time — that is, until Elon Musk filed to take SpaceX public. WeWork was a joke. SpaceX is a threat. And if Musk and…

Verge vs Elon Musk: Why SpaceX Isn’t a Meme Stock

It’s funny how Verge keeps criticizing elonmusk. But let’s be fair: where it gets things wrong is that it treats SpaceX almost like a meme stock and focuses too heavily on its inflated valuation: - Starlink is ALREADY a real global…

Journal + LLMs: Spot Patterns in Your Thinking Habits

I think that's how people think by default - choosing the easiest path (how our brain conserves energy). It takes time and willpower to change that behavior. What helped me a lot was starting a journal about 6 years ago. It gave me a way…

KidTeller AI Story Breakthrough From One Prompt

When I first released KidTeller and started receiving the first book orders, I quickly realized that, because of the nature of AI, it was quite difficult to create a consistent story from a single short prompt. Since then, I’ve kept…

Legacy code risk: the handover tax and key-person RISK

From my experience legacy code doesn’t become dangerous because it’s old but when nobody dares to read it. That instinct has a cost. Call it the handover tax - the weeks you burn waiting on documentation and knowledge transfer from a team…

How Repetition Shapes Learning From Hippocampus to Cortex

Found a fascinating neuroscience article about repetition and learning We grew up hearing: repeat things to remember them better But the article says that repetition doesn’t just make memory "stronger" globally, it changes which brain…

How AI Coding Changed My MVP Build Process This Year

Almost the whole previous year, I was using Cursor intensively, and I could say that almost all the code was written with AI. That’s how I built the MVP version of KidTeller. But after January this year, everything changed. The models…

Why Faster Project Updates Enable More SaaS and Apps

I think that’s the main point here - previously, when you needed to update an old project because of a new pivot or feature, you had to spend hours (or even days) remembering how it was built and recalling the syntax. Now you can just…

Composer 2.5 tops CursorBench at 20x lower cost than Opus

Oh wow! Composer 2.5 is 3 in CursorBench evals and 20x times cheaper than Opus 4.7 Max. It's below GPT 5.5 Low, Opus 4.7 Low. ... and Kimi (funny that it's fork of Kimi 2.5) Data is everything, especially when you’re training on millions…

Procrastination Disguised as Research: Ship Decisions Faster

The smartest way to procrastinate is to call it "research". Mark Manson named the mechanic cleanly: Beware: learning more is a smart person's favorite form of procrastination. I love rereading Seneca - he pointed that 2,000 years earlier:…

China Deploys Medical AI at Scale in Real Hospitals

🇨🇳 China aggressively pushing medical AI in real hospitals. In Tsinghua Hospital, they’ve deployed 42 AI-powered doctors and nurses across 21 medical specialties. They ran 10,000+ simulated patients in just a few days and reached 93.06%…

How Losing a Senior Engineer Exposes Hidden System Knowledge

Your senior engineer leaves. And suddenly, leadership realizes - a large part of the company’s operational knowledge existed inside one person’s head. Modern systems evolve through years of customer pressure, outages, urgent deadlines,…

Google Cloud growth outpaces AWS and Azure in AI race

Cloud is dead. It just hasn’t realized it yet. Google Cloud is the standout 63% growth vs AWS (28%) and Azure (40%). It’s not surprising - Google is clearly the winner here. Why? Because Google sells compute and its own AI (Gemini + TPUs).…

OpenAI GPT Goblins: How AI Behavioral Artifacts Scale

OpenAI: we trained intelligence... ended up with goblins. It sounds like a joke: AI models randomly talking about goblins and gremlins. After the launch of GPT-5.1, OpenAI noticed something odd - usage of the word "goblin" jumped 175%, and…

Mozilla Used Claude to Find 271 Firefox Vulnerabilities

AI can’t write secure code!? Well ... meanwhile Mozilla used Claude Mythos to discover and fix 271 vulnerabilities in Firefox. Including serious bugs related to sandbox escapes and exploit chains hidden inside one of the biggest legacy…

The LLM Fallacy: AI Output Can Distort Self-Perception

We’ve entered a weird era: outsourced thinking, insourced ego. We’re starting to confuse AI output with our own intelligence. I stumbled on this paper recently: The LLM Fallacy: Misattribution in AI-Assisted Cognitive Workflows. The…

Make Your Future Self Feel Real to Stop Procrastinating

Most people don’t procrastinate. They just don’t care about their future self. There is a study by Hal Hershfield that clearly shows that people usually see themselves in the future as a different person, and that actually affects many…

Do We Know What Consciousness Is or If AI Can Be It

Consciousness. I think people are way too confident about something we don’t even know how to define. Google DeepMind researcher saying LLMs can never be conscious. Not in 10 years. Not in 100. The argument: > a formula of gravity doesn’t…

Recursive Language Models: Why Bigger Context Can Hurt AI

Came across the paper about Recursive Language Models from MIT where they broke the whole "bigger context = better AI" narrative. The core problem is something called context rot. The more you stuff into a model, the worse it gets. It…

Anterior Mid-Cingulate Cortex: Discipline as Brain Training

There is a region in your brain that grows when you do things you don’t want to do. It's called the anterior mid-cingulate cortex. Not when you’re motivated/inspired but only when you resist… and still do it anyway. Because most of us are…

Why Learning Needs Friction in a Screen-Heavy World

I stumbled on a piece about Jared Cooney Horvath who said that for the first time in modern history, a younger generation is scoring lower than the one before it. Everything. Attention, memory, literacy, numeracy… He’s not blaming Gen Z…

Reading as a Silence Problem: Focus, Time, and Advantage

I read about 30 books a year, and for me, reading is like enjoying a tasty dessert. I focus on fiction and business, with a mix of roughly 60/40. This balance keeps me sane (and I think my love for books actually inspired me to start…

Automata vs Humans: Why 24/7 Consistency Wins

Everybody is talking about a new model Mythos, cybersecurity, etc.... but they're missing the actual story. What’s interesting is that it wasn’t even built for it…. and still beats ‼️ people who spent years there. I think this is the main…

AI-Driven Layoffs: Demand Shock and Automation Tax Debate

AI-driven layoffs undermine the entire economy by reducing consumer demand. I've expressed these concerns a couple of times already, and now we see a paper with the results. I remember RayDalio in "How the Economic Machine Works" explained…

Most AI adoption fails: executives trust, workers ignore

Food for thought from my previous article. I was quite surprised to read State of Digital Adoption 2026 from WalkMe, where they state that 80% of employees are basically not using AI. - 54% bypass it - 33% ignore it completely Companies…