Agentic Engineering System
A personal approach to building software with AI: project context, specialist agents, reusable skills, and verification. I use it across my projects; Stylefinden is one example of it in practice.
I built this approach to keep decisions and constraints connected to the work. An individual AI response is useful, but product development also needs continuity, clear scope, and evidence that a change works.
Select a node to explore its role in the system.
How I work with the system
Context before action
Each task starts with the relevant code, project constraints, and prior decisions. I keep context focused on the problem rather than loading everything. Useful findings return to project memory so the next session can build on them.
Specialized work
Work is scoped into manageable steps. Agents bring roles such as planning, implementation, and review; skills provide reusable methods for those roles. The combination depends on the task, with more scrutiny for changes that affect production.
Evidence before release
A generated change is a candidate, not a finished result. Tests, builds, browser checks, and review help establish what works. Findings can send the task back for revision before I decide whether it is ready.
System Loops
Feature Development
scope -> plan -> implement -> review -> test -> document -> ship
Content Operations
research -> draft -> review -> Sanity -> publish -> measure
Production Maintenance
detect -> reproduce -> patch -> verify -> deploy
Memory Continuity
decision -> document -> vault update -> future retrieval
Skill Matrix
Planning · Frontend · Code Review · Browser QA · SEO · Content · Research · Deployment · Memory · Verification
Connected Infrastructure
These are services behind Stylefinden, rather than the agent system itself. Scoped workflows can interact with them where the task and permissions allow.
- Sanity
- Structured content and controlled publishing workflows.
- Supabase
- Application data and operational state.
- Resend
- Transactional and lifecycle email infrastructure.
- Vercel
- Deployment, scheduled jobs, and production runtime.
- Google Analytics
- Product and content measurement.
Control Model
Scoped planning, implementation, review, testing, documentation, and selected content or service workflows.
Architecture, credentials, publishing rules, verification standards, and every production decision remain human-governed.
System in Practice
At Stylefinden, I use this approach for application development, content operations, and maintenance. The project combines a web application, editorial content, localization, and connected services, so a change needs more context than the file being edited.
For a localization change, the workflow starts with the relevant routes and market rules, then scopes the implementation and review. UI, API, and SEO checks help verify the result; confirmed decisions are recorded for future work. This is a recurring way of working, rather than a claim that every task follows an identical pipeline.
Content work follows the same principle: research and drafting are separate from editorial review and publication. AI supports the work; architecture, verification standards, and production decisions remain my responsibility.