a PaaS for the AI era

Building software has never been easier, but delivering secure, reliable software to a growing user base remains challenging. I’m exploring how that could change in the near future.

Vibe coding is all the rage these days, for good reason. Building prototypes, demos, personal apps using AI is MUCH easier than doing it by hand. Millions of people are building simple apps for themselves & those around them, often with zero prior experience. Business world processes for creating software are also changing. Why mess around with wireframes or powerpoints when you can make a functioning application to communicate your idea? AI dev tools are already being used in production, but this still requires someone with the right expertise and mindset at the helm. One thing hasn’t changed — consistently delivering secure, reliable software at scale remains the hardest problem in technology.

Production Is Still Hell

Everyone knows that nifty app AI built for you in a few minutes isn’t production ready. The big question is, how to get there? Previous generations of PaaS offerings were barely keeping up with the old world, let’s be honest. There are many options to choose from, and some made significant strides in streamlining deployment & management. Each has their strengths and weaknesses, and almost all of them can work fine, provided the team managing it knows what they’re doing. That’s all way too much for the new crop of AI-enabled developers, whose demands are higher in many ways. They need complex stateful systems that are capable of interacting with multiple LLMs over the course of a single workflow. They need frictionless access to their data, whether it’s in a SQL database or a PDF. They need a new set of defaults.

The Dream of Autoscaling

Every PaaS promises "automatic scaling.” It’s possible to achieve , but a significant amount of nontrivial configuration + ongoing care and feeding is usually required. Sometimes it can feel like the only thing perfectly automated is the bill! The time is ripe for the next generation to take the stage — here’s what they need to deliver:

Native AI Infrastructure - Not bolted on, built-in from day one
Opinionated Defaults - Standard workflow to get anyone going
Escape Hatches That Don't Suck - Opinions are great until they're wrong
Cost Predictability - I should know my bill before I get it
Intelligent Infra - File & data storage with intelligence built in

Raindrop, the new platform from LiquidMetal AI, seems to be designed with exactly these requirements in mind. It’s highly opinionated by default, and honestly I love that! Remember when Ruby on Rails made web development accessible by making decisions for you? Raindrop does the same for AI applications. "Here's how you structure an AI app. Here's where your vectors go. Here's how you handle state." The platform enforces patterns that work:

Services for your business logic
Actors for stateful processes
Tasks for background jobs
Built-in SQL and vector databases
Automatic API generation

You don't get infinite flexibility, but that was always a recipe for madness anyway. Instead, you get a proven path from idea to production.

What Caught My Attention

I first learned about Raindrop at a conference in Seattle, on the day they launched. It caught my attention immediately, here’s why:

1. Claude Code + MCP: The Interface We've Been Waiting For

Using Claude as your development interface isn't just a gimmick – it's transformative. The Model Context Protocol (MCP) integration means Claude understands your entire project, not just the current context.

"Add a background job that processes uploaded documents every hour"

Claude doesn't just write the job—it provisions the task queue, sets up the schedule, configures the storage, and deploys it. One conversation, complete implementation. Is it perfect? No. Claude occasionally misunderstands requirements. But when it works (which is surprisingly often), it feels like the future.

2. SmartBuckets: Storage That Thinks

This is Raindrop's killer feature that nobody else has figured out. SmartBuckets aren't just object storage—they're intelligent document processing pipelines disguised as simple buckets. When you drop a PDF into a SmartBucket, the following happen automatically:

Automatic text extraction - No need for PyPDF or document parsers
Intelligent chunking - The system determines optimal chunk sizes based on document structure
Embedding generation - Documents are automatically vectorized using state-of-the-art models
Index creation - Everything becomes instantly searchable
Metadata extraction - Titles, authors, dates pulled out automatically
Version tracking - Updates to documents are handled gracefully

The code to use it?

const results = await smartbucket.search("What does the document say about revenue?");
// That's it. Drop file, get searchable content.

No configuration. No embedding pipelines. No database management. It just works, and the implications are huge. Every SaaS app that deals with documents (so, all of them) need this functionality. SmartBuckets give it to you by default.

3. Multiplayer by Default: Development as a Team Sport

This is subtle but game-changing. Raindrop maintains persistent project state across sessions and team members.

Your PM can define requirements in the morning:

/new-raindrop-app
# creates new project and automatically starts the PRD process

Your developer implements in the afternoon:

/reattach-raindrop-session
# All context, decisions, and progress ready to go

You deploy that evening:

/update-raindrop-app
# Push changes without losing anything

No more "which Slack thread had that decision?" No more context lost between handoffs. The entire team operates from a single source of truth.

Final Thoughts

Anyone who needs to ship with high velocity should take a look at Raindrop. It’s still quite young — less than a year old, but maturing quickly. Remember: every new platform is a bet. The question isn't whether Raindrop is risk free; it's whether the upside justifies the risk for what you're building.

For my part, I’ve already tried a few experimental builds, and been impressed. Planning to share some of those projects and some new stuff I’m cooking up. I'll be documenting what I build, what breaks, what surprises me. The quickstart guide is your entry point if you want to experiment alongside me.

Production Software Should Not Be Rocket Science