Software development is entering a new phase where developers are no longer just writing code they are coordinating AI systems that execute tasks across terminals, browsers, and runtime environments.
However, the current tooling ecosystem has not caught up with this shift.
Developers still rely on fragmented tools:
standalone terminal sessions
browser-based workflows
scripts glued together with limited orchestration
This talk explores a fundamental question:
What should developer tools look like when AI becomes an active participant in the development loop?
To answer this, we present the design and implementation of an open-source system that reimagines the terminal as a programmable workspace.
The session focuses on the technical architecture behind building such a system on Windows, covering:
designing a hierarchical workspace model (Window → Workspace → Pane → Surface → Panel)
implementing a multiplexed terminal runtime using ConPTY
handling raw terminal streams and real-time rendering constraints
integrating browser automation into the same control plane
designing a JSON-RPC interface for external control and automation
building an event-driven system for state tracking and recovery
enabling agent-agnostic interaction through CLI and protocol design
Rather than presenting a product, this talk focuses on system design decisions, trade-offs, and implementation challenges involved in building a new class of developer tooling.
The goal is to provide a technical perspective on how open-source systems can evolve to support AI-driven workflows while remaining composable, extensible, and developer-friendly.
A new mental model for developer tooling in AI-driven workflows
How to design a programmable terminal workspace architecture
Practical insights into building a terminal runtime using system APIs
Designing RPC-based control planes for extensible tooling
Applying event-driven patterns for state consistency and recovery
Trade-offs in building cross-runtime developer tools (terminal + browser + automation)