The Era of Personal Software

as foundation models get smarter and the scaling laws keep holding, it feels almost inevitable that AI models become better than the average human at a widening range of programming tasks. the "densing law" suggests that model capability density doubles roughly every 3.5 months, meaning equivalent performance keeps arriving in smaller, cheaper packages. more compute goes into pre-training, and the models just keep getting better.

andrej karpathy coined the term "vibe coding" in early 2025 to describe this shift: you describe what you want, and the AI writes it. by 2026 it was Collins Dictionary's Word of the Year, and roughly 85% of developers were using AI tools regularly. kevin roose at the NYT called his experiments with it "software for one", personal tools built for exactly one person's needs. that phrase stuck with me.

i read somewhere that 2026 is really the year for builders. look at Peter Steinberger's OpenClaw, a weekend project that hit 200,000 GitHub stars and landed him an acqui-hire from OpenAI in under 90 days. no venture funding, no team, just one person building something useful and shipping it.

but it's not all upside. a Stanford study found that employment among software developers aged 22 to 25 fell nearly 20% between 2022 and 2025. average PR sizes went up 150%, and bug counts rose 9% alongside them. we're shipping faster, but we're also shipping defects faster. 75% of developers still manually review every AI-generated snippet before merging. the AI is a powerful tool, not a replacement for judgment.

this era is really the next era of change. it's build or be left behind. adopt, learn AI, or stay behind. the barrier to creating software has never been lower, and the people who lean into that are going to build things we can't even imagine yet.