Mark

TO ALL NON-TECHNICAL FOUNDERS We’re opening up 5 spots to help start an AI software startup in 4 weeks. No idea or technical cofounder needed. DM me if you’re interested.

AI hasn't replaced most jobs yet. But it has raised the floor for acceptable productivity significantly. Teams that haven't adopted AI workflows are competing against teams that effectively have more hours in the day. That gap only grows.

Software as a category is being redefined. It's no longer what your engineers built. It's what problem you're solving and how intelligently your product adapts to solve it.

1-person businesses are hitting $1M ARR faster than ever. AI collapsed the cost of design, development, marketing, and operations to near zero for someone who knows how to use it. The new question for investors isn't 'how big is your team.' It's 'how fast are you growing.'

The models shipping in Q1 2026 are extraordinary by any historical standard. They reason better, write better, code better, and hallucinate less than anything from 12 months ago. Founders building on top of these tools are operating with capabilities that outperform what most enterprise software teams had in 2023. The infrastructure is no longer the constraint. The constraint is knowing what to build and for whom.

Most people overestimate how much they need to know to start a business. The knowledge required to launch is much smaller than the knowledge you build by actually launching. Most businesses fail from inaction, not from insufficient preparation. The cost of waiting is always higher than it feels when you're sitting on the idea.

The irony of AI writing 41% of global code is that the people who understand code best are now more valuable, not less. You need judgment to review what the model generates. Domain expertise and taste are back as core technical skills.

The businesses that will look obvious in 5 years are the ones that seem mildly interesting and slightly early today. Most people won't touch them because the market isn't proven yet. That's usually the signal.

GPT-5.4 Thinking is benchmarking at what OpenAI internally calls GPT-6-level reasoning in a faster, smaller architecture. Every 3 months the capability curve does something that rewrites the 12-month product roadmap for AI-native businesses. Adapt fast or get outpaced.

No-code was a category. Vibe coding was a movement. AI software generation is becoming infrastructure. The difference is that the last one doesn't require you to know anything about the tools. Just the problem you're solving.

The best startup ideas in the AI era are highly specific vertical plays. Apply AI to a domain problem where the buyer has obvious pain and no good existing solution. Horizontal AI tools are hard to differentiate. Vertical AI tools with domain expertise are hard to replicate.

Headcount is no longer the leading indicator of a team's output. Output per person is the new benchmark.

Anthropic's CEO said AI may handle nearly all software engineering tasks within a year. Whether the timeline is exactly right or not, the structural change is already underway. A single experienced engineer working alongside AI now outputs what a 3-person team produced two years ago. The orgs planning for this will be better positioned than the ones still surprised by it. This is not a future risk. It's a current reality.

Stop waiting to be ready. That feeling doesn't come before you start. It comes after.

The first generation of AI-built apps had a tell: they worked in demos and broke in production. The second generation is starting to ship things that last. The tools caught up and so did the operators who know how to use them.

We are entering the judgment economy. Anyone can generate content, code, designs, and marketing with AI. What becomes rare and valuable is good judgment: who to serve, what to build, what to ignore. Judgment doesn't scale with compute. It's the one resource AI can't commoditize.

The gap between 'I have an idea' and 'I have a business' has never been smaller. The gap between 'I have a business' and 'something scalable' is where the real work starts.

Distribution still beats features. Build something people can find, share, and talk about.

The hardest part about starting something isn't the technical build anymore. It's picking the right problem, staying focused when things are slow, and not pivoting too early because it gets uncomfortable. AI collapsed the cost of execution. It didn't replace the judgment required to know what to execute on. That's still a human skill and it still separates good founders from great ones.

If you told someone in 2020 that a solo founder with no engineering background could ship a production-grade SaaS app in 48 hours for a few hundred dollars, they would have called it impossible. That's a normal Tuesday in 2026.

Build for a real customer with a real problem. Not for the demo. Not for the launch tweet. The product that survives is the one someone would actually miss.