The problem I experienced while developing with AI
I started building AEGIS Trust because of problems I repeatedly encountered while using AI coding assistants in my own development work.
AI can be remarkably productive. It can help write code, analyze problems, suggest solutions, and accelerate development.
But I discovered that getting AI to perform a task and keeping AI consistently aligned with my intentions were two very different challenges.
Sometimes, I would give an AI assistant a specific instruction, only to find that it had done something different from what I requested.
It might:
- Expand the scope of work without being asked.
- Reduce or simplify work that was supposed to be completed in full.
- Repeat work that had already been finished.
- Skip requirements or steps that I had explicitly specified.
- Make changes based on its own interpretation rather than my actual instructions.
Another problem became increasingly apparent during longer development sessions.
AI assistants could lose track of earlier decisions, forget important instructions, or incorrectly remember what had already been completed.
As the work became more complex, maintaining consistency became increasingly difficult.
A task that initially seemed straightforward could gradually move away from its original purpose.
I sometimes found myself spending more time checking, correcting, and redirecting AI-generated work than I had expected.
The realization: AI needs boundaries
These experiences led me to a simple conclusion:
AI should assist the developer, not take control of the development process.
I wanted to use AI's capabilities without allowing it to make unrequested decisions, expand or reduce the scope of work, or proceed with actions that I had not properly reviewed.
The challenge was not simply making AI produce better code.
It was about maintaining human control, clear boundaries, accountability, and the ability to recover when things went wrong.
I began thinking about how to introduce practical safeguards into AI-assisted development.
That became the motivation behind AEGIS Trust.
Introducing AEGIS Trust
AEGIS Trust is a free, open-source VS Code extension developed by ProlixoTech Inc.
Its purpose is to provide practical tools that help developers maintain greater visibility and control within supported AI-assisted development workflows.
The current version, 0.1.1, includes:
1. Local file baselines and snapshots
AEGIS Trust supports local baseline and snapshot workflows for selected important project files.
These can help developers establish a known starting point and use recovery-oriented operations when eligible.
2. AI-suggested command review
Instead of immediately running an AI-suggested terminal command, developers can submit it for review through the AEGIS command-review workflow.
This creates an additional opportunity to examine the proposed action before deciding whether to proceed.
3. Reviewed terminal handoff
The supported Protected Terminal workflow can evaluate a proposed command before handing permitted text to a terminal.
Importantly, the command is not automatically executed. The developer retains the decision to run it.
4. Local work receipts
AEGIS Trust can generate local Markdown records from available AEGIS-managed activity.
These receipts provide a useful reference for reviewing supported actions, although they are not a complete record of everything an AI assistant has done.
What AEGIS Trust does not yet solve
I believe transparency about limitations is essential.
AEGIS Trust does not currently guarantee that an AI assistant will remember previous instructions correctly or remain within the original scope of every task.
It also does not automatically intercept every action performed by external AI agents, ordinary terminals, or other software.
Its file protection and review capabilities apply to specific supported workflows. It is not a universal sandbox or a replacement for independent backups.
The broader challenge of maintaining reliable human control throughout long, complex AI-assisted projects remains an important motivation for this work.
AEGIS Trust is a practical starting point, not a claim that the entire problem has already been solved.
The principle behind the project
I believe the future of AI-assisted development should not be measured only by how much code AI can generate or how quickly it can complete a task.
It should also be measured by how reliably developers can direct, understand, review, and control that work.
AI can propose.
AI can assist.
AI can accelerate.
But the developer should remain in control of what gets done, what gets changed, and what gets executed.
That is the principle behind AEGIS Trust.
Try AEGIS Trust
AEGIS Trust 0.1.1 is available as a free, open-source VS Code extension.
VS Code Marketplace:
https://marketplace.visualstudio.com/items?itemName=prolixotech.aegis-trust
GitHub repository and getting-started guide:
https://github.com/ProlixoTech/aegis-trust
I would especially like to hear from developers who have experienced similar problems with AI coding assistants.
Have you encountered unexpected scope changes, repeated work, forgotten instructions, or AI-generated changes that went beyond what you requested?
And what safeguards would make you feel more confident about using AI in real development projects?
I welcome feedback, ideas, and contributions.
— ProlixoTech Inc.
Top comments (0)