The dream of software that does exactly what you need has been theoretical for most of computing history. You got the features some product team decided to build, in the interface they designed. Want something else? Learn to code - or accept the compromise.
That calculus is shifting. A wave of AI-assisted coding tools is creating what's being called "personal software" - applications you build for yourself rather than buying off a shelf. The Verge recently made the case that the gap between "an app exists that sort of fits my needs" and "I built the exact app I need" is narrowing fast.
The mechanics are simpler than they sound. Tools like Bolt and Claudee Code](/tools/claude-code/) let you describe what you want in plain English - "a spreadsheet that tracks my client invoices and flags ones overdue by more than 30 days" - and generate working code without you writing a line. This is what people mean by "vibe coding": you define what the software should do, the AI figures out how to build it.
What This Actually Solves
Generic software is designed for the average user of a category. Generic CRMs are built around a hypothetical sales process. Generic project management tools are built for a hypothetical team. Most businesses aren't average, and the workarounds required to bend a tool to a specific workflow add up fast.
A marketing consultant who needs a dashboard pulling from exactly three data sources, formatted the way she reads data, used to have two options: pay a developer ($5,000 minimum for something simple) or spend two hours every Monday reformatting exports manually. There's now a third option: describe what she wants, iterate with an AI coding tool, and have something functional in an afternoon.
Small business owners are the clearest beneficiaries here. They have specific, bespoke needs that enterprise software doesn't serve well, but they've rarely had budgets for custom development. AI-assisted coding doesn't fully close that gap, but it narrows it substantially.
Where This Still Falls Down
Personal software built through vibe coding has a maintenance problem. The app that works today may break when an API changes or a browser updates. If you can't understand the underlying code well enough to debug it, you're starting over rather than fixing a known issue.
The tools are also better at greenfield projects - building something new from scratch - than at connecting to complex existing systems. Getting a simple tool to read from a Google Sheet is doable. Integrating with a Salesforce instance full of custom fields and a decade of legacy data is a different project entirely.
There's also a skill floor that doesn't disappear. Getting useful software out of tools like Cursor or Bolt still requires knowing precisely what you want - which turns out to be harder than it sounds. Specifying requirements clearly enough for an AI to implement correctly is its own discipline. The mental model of what software can and can't do still matters.
None of that undercuts the fundamental shift. The question is no longer whether you can afford custom software. It's whether you can articulate clearly enough what you need.