If you're new here: I'm Michael Karnjanaprakorn. I write a newsletter about life, work, and random tings. Subscribe with 5k+ others here.
I’ve been in deep build mode. I know I’ve fallen behind my monthly cadence, but I have a good excuse: I’ve been shipping. Here’s what I’ve been working on.
First up is Slash, the productivity superapp I always wanted. It combines voice dictation, a text expander, AI commands, quick capture, emoji and GIF support, and more, and sits at the center of my workflow. It's free during the beta, so join the waitlist and take it for a spin. If you hit bugs or have feature requests, reply and let me know.
Next is Little Mavericks, a podcast for kids. Think How I Built This or Founders, but for kids. Each episode is about 8 minutes, tells the story of one maverick who found their own way, and ends with a lesson.
My biggest observation from building: the MVP has flipped. When I started Skillshare in 2010, we launched V1 as a front-end only, with nothing built on the back-end, just to test demand. Software was slow and expensive to build, and we didn’t want to waste a year on something no one wanted. In 2026, it’s the opposite. Instead of talking to users first, I just build, see what works (and what doesn’t), and iterate. I’m completely OK throwing away ideas that don’t gain traction, or that I’m no longer excited about.
What does this mean? As I wrote in Dead Zone, I think the market will split in two: you either build "sci-fi" technology (raise capital, etc.) or you build software. Software is the new small business, and there will be lots of $100K–$10M+ businesses built this way. I'm choosing to invest in the former and build the latter.
The data is starting to back this up: the rise of million-dollar companies with just one employee. Per Stripe data, the number of solo operators doing $1M+ in revenue more than doubled between 2023 and 2025, and the number crossing $10M nearly tripled.
Using frontier models well—whether for coding or for working with agents—is becoming a real skill. The people who learn it now will have a real advantage. The hardest part of working with all these AI agents is the context switching. I thought I got good reps as the CEO of a tech startup. This is on another level.
Taste is all that’s left. When AI makes software cheap to produce, the scarce skill is judgment: knowing which of several plausible versions is actually right.
The enshittification of everything. Cory Doctorow’s term for how platforms decay: first they’re good to users, then they squeeze users to serve business customers, then they squeeze everyone to claw back the value for themselves. As Mark Manson points out, this is why a flight is now a seat plus upcharges for the bag, the meal, and the legroom, and why Netflix went from $8 for everything to three tiers, ads, and six other services. It’s also why BMW charges a monthly subscription for your own seat heaters.
Which is why you should take advantage of this window: a $200/month AI subscription gets you around $5K worth of LLM API calls every month. That’s a 25x “return.” Two subscriptions get you about $10K a month in compute for under $5K a year. This won’t last forever, so use it before the enshittification kicks in now that Anthropic is a public company.
Most of my SaaS subscriptions are gone, replaced by AI subscriptions. The latest to go: Runna. I replaced it with my own AI running coach on Grok Bot. It checks my sleep score, signs me up for races, and plans my workouts. It also works with my longevity coach, which tracks my biomarkers and keeps me up to date on new research. All of this for $30 a month. It’s already paid for itself. And that’s just one of the many things it does.
I've started looking at new health insurance options. My premiums went up 15% while my benefits got worse. A friend told me about CrowdHealth, so I've been digging in. It's not insurance. You pay $65 a month and chip in to crowdfund other members' medical bills, up to a monthly max. When you have a health event, you pay the first part (up to $500 per event), and other members crowdfund the rest of the eligible bill directly to you. I haven't signed up, but I'm very intrigued.
I went to see Maybe Happy Ending. A+. It’s about two retired “HelperBots” in near-future Seoul, robots built to serve people and then left behind as obsolete. It won six Tonys in 2025, including Best Musical. It’s a musical about robots, but really it’s about loneliness and connection.
I am almost finished reading American Hagwon by Min Ji Lee (who also wrote Pachinko). It’s a 650-page family saga that follows the Kohs from Seoul to Sydney to Orange County, from 1992 to 2007. At its center is the hagwon, the private after-school academy where Korean kids spend nights and weekends drilling for an edge. The 1997 Asian financial crisis upends the family, and the kids end up working at an American hagwon. Underneath it all is one question: what is education actually for?
Collectibles are booming, and I think vinyl records could be the next big thing. I just sent some of mine to AMG (Audio Media Grading), Steve Aoki's grading company, to get graded.
I've also been writing again: Dead Zone, Concentrate, Then Diversify, and Last Mile.
Most of us raise girls to be agreeable and easy to get along with. Here’s some contrarian advice for daughters: don’t be likeable.
Until the next one.


great to have you back at it!