Leobit
AI-Native SDLC for essencedesign
Using Claude to help essencedesign build a machine inspection solution
ABOUT the project
- Client:
- essencedesign, a software vendor
- Location:
-
Switzerland
- Company Size:
- 20+ Employees
- Industry:
-
Industrial software, field service and maintenance
- Solution:
-
Custom software
Technologies:
Leobit provided additional tech capacity to essencedesign, a Swiss software vendor building a field inspection and reporting solution for its end-customer, a manufacturer of large conveyor and packaging systems. The solution has been running in production since 2017, helping the end-client track the state of their manufacturing machines and generate comprehensive reports in different formats. The tool supports three distinct operating modes, each matching a real field workflow: maintenance for scheduled servicing, optimization for performance and upgrade assessments, and installation for commissioning new equipment.
The solution’s development model was rebuilt around Antropic’s Claude 2026, with a smaller team, a documentation-first process, and an agent-driven implementation and testing.
Leobit provides efficient and affordable services. The team is responsive and works effectively across multiple project management tools. Future collaboration is likely. Customers should expect to work in a fast-paced environment.
Customer
essencedesign is a Swiss-based branding agency that provides a variety of services to companies across industries. Among its services, the company offers software development to businesses, including the end-customer on this project — a Swiss industrial software vendor serving a manufacturer of conveyor and packaging systems.
Business Challenge
The customer was running a large-scale inventory management project for a Swiss manufacturer and needed additional capacity to accelerate delivery and handle multiple tasks in parallel. As a result, essencedesign established a long-term cooperation with Leobit, ongoing since 2018.
Why Leobit
In terms of quality and technology expertise, particularly with technologies like .NET and Angular, Leobit was one of the most fitting software vendors in the outsourcing market.
Project
in detail
This product was built between 2017 and 2024 by conventional means. In 2024, essencedesign decided to shift towards an AI-driven approach to product development. Leobit, which had been running the development team on the project since 2018, took on the transition with them. The change was phased through 2026
essencedesign and the Leobit team rebuilt the process around Claude Code, which allowed the client to reduce the development team. Engineers established an approach where implementations are derived from specification documents and carried out by Claude Code. In the new approach, they review, validate, and steer the code instead of writing it manually.
The testing model changed as much as the development model. Instead of a more traditional and manual approach to end-to-end product validation, we embraced an AI-driven approach. It is centered around Claude, which drives the browser directly against the running application, explores the interface, determines how the feature is actually meant to be operated, and verifies the outcome. Scenarios are run again on every change instead of being sampled before a release, and interface changes do not invalidate them.
A production system that has been extended continuously since 2017 carried a substantial technical debt. The AI-driven approach has simplified its management sufficiently. Refactoring work that had been deferred for years is now within reach of a normal sprint, and is being done alongside feature delivery rather than instead of it.
In particular, the new approach helped the development team tackle two long-postponed projects — a multi-version Angular upgrade and the merger of two drifted-apart front-end applications into one.
The team is currently developing the Machine Module, a centralized repository for data on a specific machine: installation details, upgrade history, parts replacements, status, and pending maintenance. Future plans include adding an AI layer to the system to automate some of its workflows.
Enhanced Delivery Loop with AI-native SDLC
For defects and small-to-medium evolutions, no developer writes implementation code at any point in the cycle under the new approach.
A ticket or a short specification is handed to Claude Code. It implements the change, runs the test suite, and pushes, while all the commits are analysed automatically. The change is then exercised interactively against the running application, deployed through the client’s CI/CD pipelines, and released.
Engineers intervene at two points: framing the problem before, and validating the result after.
The Machine Module Developed under the New Model
The Machine Module is the first component of this product designed under the new model. It serves as a single source of truth for an individual machine: installation record, upgrade history, parts replacements, current status, inspection results, and pending maintenance. It consolidates data that was previously spread across the inspection tool and the manufacturer’s own systems, and keeps it synchronised.
Architecturally, it exposes data and operations through an MCP server, making the machine record directly queryable by AI agents. The direction is a digital twin capable of introspection: machines that explain their own condition and anticipate their own maintenance. That target is only credible for a team that can build this fast and design AI-native architecture. Both of these capabilities came from the same shift.
Technology Solutions
- An AI-powered workflow centered around Claude, accelerating software delivery
- Testing workflows handled by AI
- Machine Module that exposes data across MCP server, providing a unified knowledge base on machines
- Continuous work on the AI layer aimed at making manufacturing machines capable of explaining their own condition
Value Delivered
- Accelerated development loops, eliminating existing limits in support, delivery of new features or modules
- Technical debt that had been unaffordable for years is being cleared, as the tasks became much cheaper to handle with a new approach
- Developers became leads of agent-based delivery, and testers became designers of test agents and BDD scenarios written for Claude
- Continuous collaboration with essencedesign aimed at making the inspection tool more self-sufficient and introspective through AI integration
- The end-client (Swiss manufacturing company) gets more effective and performing software at the same cost
