TL;DR
Many business processes run on Excel — training management, material tracking, resource planning. Custom software used to be too expensive. With coding agents like Claude Code or Cursor, implementation costs approach zero. What remains: time for definition and testing. But the project becomes 80% cheaper.
Contents
- What’s the problem with Excel?
- Why was custom software too expensive before?
- What has changed with AI?
- Which processes are suited for mini-apps?
- Where are the limits?
- FAQ
What’s the problem with Excel?
In short: Excel is flexible but not built for collaboration and complexity. Eventually, it becomes a risk.
It’s an open secret: Many business processes run on Excel and brainpower. Lists are maintained, data pulled together, formulas linked. It works — until it doesn’t.
The typical problems:
- Multiple people work on it, versions get confused
- Complex formulas break, nobody knows why
- Data must be manually copied from other systems
- The one employee who understands the Excel goes on vacation
Examples from practice:
- Training management: Which employee needs which certification renewed when?
- Material tracking in R&D: Where were which components installed? What were the results?
- Resource planning: Who has which skills? Who is available when?
- Financial planning: Run scenarios, but please collaboratively
Why was custom software too expensive before?
In short: Two cost drivers — definition and implementation. Both were effort-intensive, both were expensive.
The classic solutions:
- Excel — Flexible but fragile. No real solution for collaboration.
- Extend existing software — Customize SAP, ERP. Big projects, high costs.
- Buy SaaS — Ready-made tools, but I have to adapt to their process. (When SaaS makes sense and when it doesn’t)
- Custom software — Perfect fit, but unaffordable.
Why custom software was so expensive:
The definition: Before anyone can code, you need to define: What exactly do we need? How should the process work? Edge cases? The business department does this — and they have no time.
The safeguarding: Because changes later are expensive, you put 30% of project effort into planning upfront. You try to anticipate the future.
The implementation: Software engineering, UX design, testing. That needs a team, that takes weeks or months.
The result: Projects for €50,000–200,000. For an internal process? Not economical.
What has changed with AI?
In short: Implementation costs approach zero. What remains is the thinking work — but it distributes differently.
With coding agents like Claude Code or Cursor, the equation changes:
- I write a specification — not a hundred pages, a few paragraphs are enough
- The agent codes — and builds in hours what used to take weeks
- I test — in the business department, with real users
- I iterate — not right? Agent does it differently
The crucial difference:
Previously, the specification had to be perfect. Every error became expensive later. So: months of requirements analysis before anyone writes a line of code.
Today, I can start with 85% correctness. I see the result, say “I meant something different,” and the agent adjusts. The time for definition remains — but it distributes across the project instead of blocking at the start.
The result: Projects for €5,000–15,000. Suddenly economical.
Which processes are suited for mini-apps?
In short: Anywhere Excel stands today and really needs a proper solution.
Good candidates:
| Process | The Excel Problem | The Mini-App Solution |
|---|---|---|
| Training management | Track expiration dates across hundreds of employees | Dashboard with reminders, filter function |
| Material tracking R&D | Which component installed where? Results? | Database with links and search |
| Resource planning | Bring together skills + availability + orders | Matching tool with visualization |
| Competitive analysis | Gather data from various sources | Automated research + overview |
| Financial scenarios | Complex formulas, multiple editors | Collaborative tool with versioning |
The pattern: Data from multiple sources, multiple users, no perfect SaaS solution on the market.
Where are the limits?
In short: For internal tools with manageable complexity, it works. For Salesforce-like products, not (yet).
What works:
- Internal process tools
- Manageable number of users
- Clear requirements (even if not perfectly defined)
- No extreme scaling needed
What doesn’t work (yet):
- Products for thousands of external users
- Highly complex systems with many development teams
- Security-critical applications without extensive testing
The open question: Is AI-written code as robust as hand-written? We don’t know for sure yet. But: For internal tools with limited complexity, the risk is manageable.
What changes: Previously, clean code was important so other developers could understand it. With AI-assisted development, code becomes more “disposable” — I can regenerate or adjust it anytime. The quality question shifts from “Is the code maintainable?” to “Does the app work?”
FAQ
Do I need programming skills?
Not to use the app. For development with coding agents, technical understanding helps, but you don’t need to be a software engineer. The AI explains itself.
How long does such a project take?
The first working version: days, not weeks. Then iteration with users. A complete project: 2-4 weeks instead of 6-12 months.
What does it cost?
A fraction of traditional software development. Instead of €50,000+ more like €5,000-15,000. The biggest cost factor: the business department’s time for definition and testing.
Can I do this myself?
If you’re tech-savvy and have time: yes. More realistically: A consultant sets up the system, you iterate yourself after. Similar to the AI-first website approach.
What happens when the developer leaves?
With traditional code: Risk that nobody understands the code. With AI-generated code: Less risk because the next person can use AI to make changes.
Does this replace our existing systems?
No. The mini-app fills a gap — where SAP is too big, SaaS too ill-fitting, and Excel too fragile.
Conclusion
The cost of custom software is falling radically. What was previously a luxury for corporations — process tools that fit exactly — becomes affordable for mid-sized companies.
The moment is now: The specification is in the Excel files. The business department knows the process. What was missing was affordable implementation.
Now it exists.
This article is based on a conversation between Manuel Zorzi and Michael Kirchberger about developing business applications with AI support. Watch the full podcast episode on YouTube →