The Landscape is Changing

This article continues our exploration of what may be the most interesting phenomenon in the software world this century – the emergence of a variegated and rapidly growing army of AI-empowered programmers. These folks are building, sharing, and shipping apps created by AI. They run the gamut from people with zero technical understanding of software, to experienced developers who are evolving the ability to work with unprecedented speed. This market emerged out of nowhere in the past months, starting when it became clear that the frontier models are actually quite good at writing code. When Anthropic dropped MCP and then Claude Code, things really took off. The reason? Claude Code is misnamed. Sure it can write code, but so can many other products. What sets it apart is the powerful workflow engine. Paired with MCP, it becomes easy to wire anything to almost anything else. The folks at LiquidMetal were among the first to figure this out and are already running with it. THey released their own MCP along with another key ingredient: a custom CLAUDE.md file that takes full advantage of CC’s workflow capabilities. Here’s a diagram:

I’ve used this over a dozen times so far. It works amazingly well, given the key ingredients are just a few months old! A single pass generally yields beta quality results, depending on the complexity. That’s impressive, but what’s really got me excited is the future possibilities. LiquidMetal can easily refine this workflow over time, and I’m sure they will. They could even maintain different versions for specific use cases. Users can customize it to meet their needs as well. I’m presently working on a few ideas – generating documentation, test automation that runs after every build, similar to a CI/CD system.

As part of the writing process, I created a new app. Simply to keep the process flow top of mind. That’s how easy it is – you can go through these steps in the background while focusing elsewhere. I paused it last night and resumed this morning. Below are some screenshots of the progression from definition to completion. The upper right one shows the MCP call that triggers the start of coding. You’ll notice:

• It’s at the end of a long checklist, each step must pass validation before Claude continues. The workflow is rigid when you’re inside it, but tweaking it is as simple as editing a text file. This combination of internal rigidity with external fluidity seems like a big deal.

• The call itself is dead simple, just the identifiers Raindrop needs to determine what you’re working on and where you’re at in the process. The rest is stored locally, where the creation happens. Deployment happens via a dedicated tool that Claude calls. I appreciate being able to see what’s going on and the choice to control or automate, depending on the situation.

Honestly, I’m not sure what to call this anymore. It’s way more than just a PaaS – I toyed with the term ‘software factory,’ which is popular in industrial and defense verticals, but it’s not that either. A factory is static, and reconfiguring it takes considerable effort, whereas experimenting with Raindrop + MCP & Claude Code couldn’t be easier. The best description I currently have – it’s a state machine for reliable software development and deployment. But it’s also constantly evolving. Between this blog post and the last one, the Raindrop team added a feature that allows you to turn any backend project into it’s own MCP. Just one extra step, and now you can plug your new creations into other apps. I’ll be trying that out as soon as I hit the publish button.

---

What sets Claude Code apart is the powerful workflow engine. The folks at LiquidMetal were among the first to figure this out — they’re already running with it.

---

Don’t get me wrong – I’m not claiming some lone wolf is going to singlehandedly build the next Hubspot or Slack with this system anytime soon. But that’s not the point. Remember, we’re talking about the shiny hordes of vibe-coders out there, who are building apps for their customers, teammates, family & friends. Raindrop is not the platform for recipe catalogs or book club apps, lovable’s got that covered. This is the platform for business users, researchers, and other geeks. For people who need help managing massive piles of data locked up in hodgepodge collections of files and web pages. They could never afford pricey enterprise solutions, but they can easily afford this. Whether it’s the stock market, climate data, digital archives, urban planning, theoretical physics, biochemistry, warehouse logistics, the bar is suddenly much lower.

I’m not going to end with any grand pronouncements about where this is heading. Extraordinary claims require extraordinary evidence, and while this certainly seems extraordinary to me, my aim is not to convince you. I’d much rather have you try it and share your thoughts! Signup link is on the home page, docs are here, you can find me in the Discord.