SOFTWARE ENGINEERING · AI & MACHINE LEARNING
Visions Studio
An Electron based agentic development environment combining model backed code chat, a Monaco editor, controlled local tools, durable work, approvals, and fail closed security boundaries.
The problem
AI assisted development needs useful code and tool workflows without giving a model unrestricted authority over files, shells, approvals, or release decisions.
Why I built it
Justin built Visions Studio to explore a useful agentic development workflow while keeping application authority, approvals, and release decisions outside the model.
The solution
Visions Studio combines model backed code chat and editing with bounded file and shell access, durable tasks, recovery, activity management, human approvals, and fail closed behavior.
How it works
- Accept a bounded development task and relevant context.
- Use model backed chat and editor workflows within configured tool permissions.
- Require human approval for higher risk actions and retain durable activity and recovery state.
My contribution
Justin designed, directed, implemented, reviewed, tested, and iterated on the environment using agentic coding workflows while retaining responsibility for architecture, acceptance criteria, and release decisions.
Engineering decisions
- Keep application authority outside the model.
- Use explicit permission boundaries for local file and shell access.
- Treat real model and complete workflow certification gates as release requirements.
Validation
The browser workspace is temporary, uses a locally hosted Monaco editor, and binds server objects to the existing LinkedIn session.
Product screenshots
1 of 1
The browser based Visions IDE entry point with temporary workspaces and LinkedIn sign in.