I signed up for another SaaS

There's a particular kind of whiplash that happens when someone tweets a hot take and then quietly contradicts it 24 hours later. Ben did exactly that this week. He called SaaS dead on a Tuesday, then by Wednesday he'd maxed out a free plan and paid for MagicPath, a tool that lets multiple AI agents generate design assets, components and animations together on a shared canvas.

The backstory matters here. Ben's been building an interactive course, the kind where readers can actually manipulate a concept instead of just reading about it, and he'd burned through a pile of models and tools trying to get components that looked right and felt right. Nothing clicked until Pietro, someone he met at OpenAI's Dev Day last year, pointed him toward MagicPath. One test run produced a set of mechanical-style components that apparently nailed the vibe he was after, buttons, prompt boxes and all, and suddenly he had a real workflow instead of a pile of half-finished experiments.

That's the quiet story under the joke. Multi-agent canvases for design work are becoming a genuine category, not a novelty. And the same edition of Ben's newsletter is stacked with evidence that agentic tooling is maturing fast on the coding side too. Claude Code shipped a security plugin that flags risky patterns like unsafe command execution while the model is still writing the code. DeepSWE, a new benchmark built from 113 long-horizon tasks across 91 repos, has GPT-5.5 fixing problems that average 668 lines across seven files, hitting 70% versus Claude Opus 4.7's 54%. Ramp, meanwhile, ran 10,000 homegrown security agents that found and patched close to 100 issues in six days, with humans just reviewing the pull requests.

So when Ben says SaaS may be dead and then pays for a SaaS product days later, it's less hypocrisy and more a symptom of where things actually stand. The tools people keep signing up for aren't the old dashboards-and-seats software. They're thin wrappers around agent behavior, and they earn subscriptions the old-fashioned way: by actually saving someone real time on a real problem.