Are coding agents hitting a wall? We are building a visual developer agent and want your feedback.

Most developer agents today operate entirely in a sandbox or via text-based CLI. They are excellent at writing isolated functions, but they are completely blind to the external tools we actually use to ship software. As soon as a project requires configuring an external service, generating API keys behind a login wall, or dealing with dynamic pricing tiers, the developer has to step in and handle the friction manually.

We wanted to see if we could bridge this gap. We are working on a visual developer agent-the Universal Operator. Before we write the actual code, we wanted to get some feedback from other engineers on our approach.

The Problem: We want to build, not babysit

As developers, we want to focus on high-level logic and architecture. But current AI coding assistants often require us to be the bridge between the code and the browser. If we want to set up a database, connect a payment gateway, or deploy, we have to do the manual GUI work ourselves. We are designing our agent to handle the boring manual steps by giving it eyes and hands.

What we are designing the agent to do

• Verify first, code later: It crawls the web to check the latest documentation and pricing tiers before starting, so you do not get hit with unexpected paywalls mid-build.
• Visual operations: It uses computer vision (screenshots) to log in, navigate dashboards, solve Captchas, and configure third-party tools exactly like a human would.
• Self-healing terminal: It manages local and cloud environments, automatically installing packages and resolving CLI dependency errors.
• Code hygiene: It continuously refactors the codebase to ensure the AI does not turn your clean project into a spaghetti-code mess.

We would love your input!

We want to build something that actually improves the developer experience, not just another flashy tool. We would love to hear your thoughts on these questions:

• What is the most annoying manual task you have to do when setting up or deploying a new project?
• How do you feel about giving an AI agent visual access to your browser and terminal? What security guardrails would you expect?
• Do you prefer having an agent that does everything autonomously, or do you want a strict prompt asking for your permission at every step?

Please drop your thoughts, critiques, or ideas in the comments below. We are looking forward to chatting with the community!