AI at ApPello – Not Just a Tool, but a Shift in Mindset
AI at Appello – Not Just a Tool, but a Shift in Mindset
When AI broke into the development world over three years ago, we jumped in early. We tried GitHub Copilot, then IntelliJ IDEA AI Assistant, genuinely believing it would transform our speed and quality overnight.
It didn't. At least not right away.
That first phase was really about finding where it fit. Generating unit tests, drafting short documentation — it performed well in those areas. But the breakthrough we expected didn't come, and the question divided the team: real tool, or just hype?
The shift happened when we stopped asking "what can AI do?" and started asking "what do we want to use it for, intentionally?"
A New Dimension
As newer models emerged — larger context windows, agent-based workflows — everything changed. Tools like OpenAI Codex CLI and Claude Code are no longer chat assistants. They can take a structured task description and produce working code, conduct reviews, run tests, and generate full documentation.
Today, we see a clear spectrum within our team. Some engineers rarely write code manually anymore — they specify, direct, and review. The review itself is AI-assisted: Sonar flags issues, Claude Code surfaces suggestions, and the developer makes the final call. Others are still copy-pasting from chat windows. The gap between them isn't technical. It's a difference in how they think.
Specification Is the New Critical Skill
One thing is increasingly clear across the industry — and we see it firsthand: writing the code is no longer the expensive part. A poorly written specification is.
When production gets cheap, bad specifications get very expensive. AI perfectly executes bad instructions at remarkable speed. The real value is created when the task is structured and well-defined — and the AI builds on that foundation. This is why we now treat specification as a core engineering skill, not a formality.
Building Skills, Not Just Better Prompts
Our next step wasn't to write better prompts — it was to build reusable capabilities.
We developed Jira and Confluence skills that can summarize tickets containing hundreds of comments and attachments, and transform complex bug reports into structured, actionable formats. These connect via MCP to both Claude Code and OpenAI Codex CLI. The AI doesn't just see code — it sees context.
AI Behind a Low-Code Platform
Appello development isn't standard backend/frontend work. Our low-code/no-code platform is heavily configuration- and metaprogramming-driven — which creates both a challenge and an opportunity.
We're building a proprietary agent-based AI assistant, currently in active testing and already integrated into parts of our development workflow. Connected via MCP, it goes beyond code generation. It can interpret system configurations, handle parameterization tasks, and connect platform-level modules.
We've started with our decision engine — a general-purpose rules engine — but the vision goes further.
Where We're Heading
Our goal is to build dedicated skills for every internal platform module: the data modeler, notification module, template engine, and others. Each skill will allow LLMs to configure, parameterize, and orchestrate these modules directly through CLI and API — the same way a senior developer would, but at AI speed.
The vision: an AI assistant that doesn't just understand code, but understands the Appello platform itself. One that can take a structured requirement and translate it end-to-end — from data model to business rule to notification — without the developer having to manually bridge each layer.
This aligns with a broader shift we believe in: as the cost of building approaches zero, the bottleneck moves to specification and judgment. The engineers who thrive won't be the fastest coders — they'll be the ones who can define problems precisely, structure work so agents can execute it, and evaluate outcomes intelligently.
Where We Are Today
AI is now consciously embedded across our entire development workflow: specification preparation, ticket processing, code generation, review, documentation, and configuration.
And this is just development. Testing and test automation deserves its own post.
The question isn't whether AI will replace jobs. It's who learns to use it systematically — and who stays in the chat window.
We're building toward the former. Deliberately.