Applications and integrations
Agents focused on project logic, data flows and dependencies help develop changes that fit the existing systems.
SOFTWIRED · MY APPROACH
AI-assisted development
How I develop with AI
I build my own applications around real needs. ChatGPT helps me clarify and plan the task, Claude Code works with me on the implementation, and specialist agents bring knowledge of individual areas into the process.
I define the goal, choose between solutions, verify the result and operate the system. AI is a tool in my work; I take responsibility for the decisions and the finished application.
ChatGPTClaude CodeAI agentsGit · GitHubDocker · Linux
01 / WORKFLOW
From an idea to lasting knowledge.
Development involves conversations, concrete decisions and changes that can be checked. Lessons from everyday use feed back into planning.
- 01
Idea and human decisions
I define the problem, the intended outcome and the constraints. I decide what to automate, which data the application should handle and where it should run.
- 02
Planning with ChatGPT
Questions and alternatives help refine the requirements. The specification separates the goal, current state, decisions, architecture, data model, testing and acceptance criteria.
- 03
Implementation with Claude Code
The actionable specification becomes code that follows existing repository patterns. Claude Code uses the project rules and documentation to help implement, change and fix the application.
- 04
Specialist agents
Agents suited to the task help with application logic, testing, operations and documentation. They can contribute at several points in the workflow.
- 05
Git and verification
Changes are made in a separate branch and worktree. I review the changes; project-specific tests and functional checks verify them before focused commits go to GitHub.
- 06
Deployment and operations
Each release follows the project’s own deployment procedure. I run applications on Linux servers and in Docker environments, using tests, health checks and monitoring to follow their behaviour.
- 07
Preserving knowledge
Proven procedures, key decisions and lessons go into version-controlled documentation. The next development cycle starts from the project README, shared rules and topic-based knowledge base.
02 / AGENTS
Shared context. Several areas of expertise.
Specialist AI agents work from roles and project knowledge documented in the repository. Separating their responsibilities helps bring the right context to each task.
Testing and operations
Specialists in testing, monitoring and backups also consider the runtime environment and verification procedures.
Documentation and knowledge
Dedicated roles maintain project knowledge, the development log and agent documentation so the next session has a useful starting point.
Human direction
I assess agent results against the goal and acceptance criteria. Priorities, trade-offs and the decision to publish remain my responsibility.
03 / A PRACTICAL EXAMPLE
Office Presence
Office Presence is a public beta office attendance planner with Google sign-in. It illustrates how an everyday organisational need can become a standalone web application.
From a need to rules
I first break the need to plan office days clearly into behaviour rules, interface requirements and data-handling decisions.
From rules to an application
The specification refined with ChatGPT forms the basis for implementation with Claude Code. During development, I try the interface and behaviour, then refine them based on the results.
From an application to a service
The application deployed from version-controlled source is available under its own domain. My work continues with checks after release and independently operating the service.
04 / OPERATIONS
Development continues after release.
After deployment, I check the application and use lessons from everyday operation to guide the next development cycle. My servers, containers and network provide the technical foundation described in a separate chapter.
Explore my infrastructure →