Why Your Next “Enterprise App” Should Be a Mini-App
And why I stopped believing in visual IDEs
I spent two decades in enterprise IT. I’ve watched the cycles come and go.
“No-code will replace developers.” It didn’t. “Low-code will democratize software.” It didn’t. “AI will code everything for us.” Well… this one might actually be different.
But here’s what I’ve learned. The tools we use to build software matter less than the philosophy behind them. And most enterprise app builders — OutSystems, Mendix, Salesforce Lightning — are built on the wrong philosophy entirely.
They’re built on the idea that software development is a visual design problem. That if we just make the IDE pretty enough, drag-and-drop intuitive enough, anyone can build an app.
They’re wrong. Software development is a reasoning problem. And reasoning is what AI and humans in the loop do best.
The $36,000 learning curve
Let me tell you about the last time I evaluated OutSystems for a client.
Three months. That’s how long it took their team to build a simple customer intake form with an approval workflow. Three months of platform certification courses, data modeling workshops, visual IDE training, and “best practices” sessions that were just proprietary conventions dressed up as wisdom.
The bill? $48,000 for the platform, plus $15,000 in training, plus the salaries of three people who spent a quarter learning to drag boxes instead of shipping features.
And the form worked. Eventually. But when they needed to customize the approval logic — a simple “if customer is VIP, skip level 1” rule. They discovered the visual model couldn’t express it cleanly. So they wrote custom C# extensions inside the OutSystems runtime.
At that point, what exactly was the point of the visual IDE?
The alternative nobody talks about
I’ve been building something different. It’s called JClaw, but the name doesn’t matter. What matters is the approach.
Instead of a visual IDE, we use lightweight HTML/CSS/JS interfaces, single-page mini-apps that do one thing well.
Instead of a proprietary backend runtime, we use AI agents that reason about problems, call tools, and integrate with any system.
Instead of vendor-locked databases, we connect to Postgres, Supabase, Airtable — whatever the team already uses.
The result? That same customer intake form took three hours to build. Not three months. Three hours.
A simple HTML form. An agent that validates input, checks a database, and routes to the right approver via Slack. Done.
No platform certification. No $36,000 license. No visual IDE to learn.
What is a mini-app, really?
A mini-app is embarrassingly simple.
One HTML file. Some CSS that doesn’t need to win design awards. JavaScript that calls an API endpoint. An AI agent on the backend that handles the actual thinking.
That’s it. No framework lock-in. No deployment pipeline from 1997. No proprietary runtime that goes end-of-life and takes your business with it.
You can host it anywhere. You can modify it with any text editor. You can understand every line of code in an afternoon.
And because it’s just standard web technology, AI can write it, maintain it, and optimize it. Something no visual IDE can claim.
The 80% problem
Here’s the uncomfortable truth that visual platform vendors don’t want you to know.
80% of enterprise applications are not actually complex.
They’re forms. Dashboards. Approval workflows. Data sync jobs. Notification systems.
The complexity isn’t in the software — it’s in the business logic, the integration patterns, and the human decision-making. And that’s exactly where AI agents shine.
Why would you buy a Formula 1 car to drive to the grocery store? Why would you pay $36,000 a year for a visual IDE to build a form that a simple HTML page plus an intelligent agent handles better, faster, and cheaper?
Why this matters now
The AI revolution isn’t coming. It’s here. And it’s changing the fundamental economics of software development.
Six months ago, I watched an AI agent write a complete customer portal: authentication, database queries, a responsive UI, and API integrations in about four hours. It wasn’t perfect, but it was good enough to deploy and iterate on.
Try that with OutSystems. Try getting an AI to drag boxes in a visual IDE. It doesn’t work because visual programming was designed for humans, not intelligence.
Get Tarun Sukhani’s stories in your inbox
Join Medium for free to get updates from this writer.
The future belongs to architectures that AI can natively understand and manipulate. Text. Code. APIs. Standard web technologies.
Every advance in LLM capability makes the mini-app approach more powerful. Every advance makes visual platforms more obsolete.
The cost reality
Let me be blunt about money because nobody else is.
OutSystems costs $108,000+ over three years just for the platform license. Add $15,000 per person for certification. Add the cost of hiring specialists who know the proprietary stack. Add proprietary connector fees for every integration.
The mini-app approach costs under $20,000 over three years. And that’s mostly hosting and development time, not vendor taxes.
I’ve seen startups burn their entire first-year budget on platform licenses before shipping a single feature. I’ve seen enterprises spend millions on “rapid development platforms” that delivered nothing but vendor lock-in.
It doesn’t have to be this way.
But what about scale?
Someone always asks this. Sure, mini-apps are fine for simple stuff, but what about enterprise scale?
Fair question. Here’s my answer.
Most companies never reach the scale where OutSystems’ complexity pays off.
If you genuinely need 10,000+ concurrent users, strict HIPAA compliance, and 99.999% uptime guarantees, fine. Buy the enterprise platform. That’s the 20% case.
But for the other 80%? The startups, the SMBs, the internal tools, the departmental apps, the rapid prototypes?
You’re paying enterprise prices for problems you don’t have.
And here’s the thing — you can always migrate up. Start with a mini-app, validate your use case, and prove the ROI. Then if you genuinely hit scale that demands enterprise tooling, you’ve got the business case to justify it.
The opposite direction is much harder. Once you’re locked into a proprietary platform, you’re stuck. I’ve seen companies spend years and millions trying to escape.
What I’ve learned building this way
After running JClaw for a while now, here are the patterns that actually work:
Pattern 1: Form to Agent to Action
User fills a simple HTML form. An AI agent validates the data, checks business rules, routes to the right person via WhatsApp or Slack, and updates the database. Total build time: 2–3 hours.
Pattern 2: Dashboard to Live Intelligence
A lightweight dashboard calls an agent API. The agent fetches data from three different systems, synthesizes it, detects anomalies, and returns clean JSON. The UI just renders it. The intelligence lives in the agent.
Pattern 3: Scheduled Thinking
A task runs on a schedule. The agent checks APIs, reads documents, compares against historical data, makes decisions, and either updates a dashboard or sends alerts. It’s like having a junior analyst who never sleeps.
None of these need a visual IDE. None need $36,000-licenses. They need clear thinking, good API design, and the willingness to trust that simpler is better.
The philosophy behind it
At Abundent, we believe something that sounds simple but took me years to truly internalize.
Organizations win by equipping people with AI, not replacing them with it.
A mini-app doesn’t replace your team — it equips them. They still make decisions, set priorities, and validate outcomes. But the boring stuff (the form validation, the data routing, the notification sending, and the integration coding) gets handled by agents.
Your people do what humans do best. The AI does what it does best. The mini-app is just the interface between them.
The bottom line
I’m not saying OutSystems is bad software. It’s actually quite good — for the specific use case it was designed for.
What I’m saying is most companies don’t have that use case.
They need speed. They need flexibility. They need AI compatibility. They need to stop paying vendor taxes and start shipping features.
They need mini-apps.
If you’re evaluating enterprise platforms right now, ask yourself one question. Are we buying complexity we don’t need?
If the honest answer is yes, there’s a better way.
Tarun is the founder of Abundent, an AI-powered learning and productivity company based in Malaysia. He builds with JClaw and believes the best code is the code you don’t have to write.
