This is a product vision draft, not a delivery checklist.
For phased rollout, see roadmap.md.
For implementation backlog, see bugs.md.
TermAl started from the terminal because that was the fastest practical entry point into real AI coding workflows.
That does not mean the product should stay bounded by terminal assumptions.
The terminal was the entry point. It is not the product boundary.
TermAl is an operating environment for coding agents.
It should give a developer one place to run, supervise, review, steer, and eventually collaborate around long-running software work performed by AI agents.
The product should feel purpose-built for agent-driven engineering work:
Working effectively with coding agents requires new product primitives.
Generic terminal and chat interfaces are useful starting points, but they are not sufficient on their own for the workflows that matter most:
TermAl should be designed around those primitives directly.
The right mental model is not “AI replaces software engineering.”
It is “humans and agents work together inside a system that protects correctness.”
Agents expand execution bandwidth. Humans hold intent, judgment, and responsibility for whether the work is actually correct.
That makes correctness the real bottleneck, not code generation speed.
Agentic development can make a strong senior engineer dramatically more productive.
That leverage is only real if the surrounding SDLC is strong enough to keep the work correct.
In an agent-heavy environment, SDLC is not secondary process overhead. It becomes the control system for accelerated engineering work.
That includes:
Without those controls, agent speed just increases the rate at which mistakes spread.
If coding agents become part of normal engineering work, developers will need something closer to a control room than a terminal window.
That control room should let people:
Pair programming becomes more important, not less, when agents increase individual throughput.
One senior developer with strong agent workflows can otherwise occupy too much project space too quickly. The codebase changes faster than peers can track, review, or meaningfully shape.
Pair programming helps close that gap by keeping two humans inside the same fast-moving context.
This is cooperative overlap:
In that sense, pairing is not just a coding style. It is a synchronization mechanism for high-velocity agent-assisted development.
It can also help narrow the gap between senior and junior engineers if the workflow keeps the junior inside the reasoning and review loop instead of turning them into a passive consumer of generated code.
The UI should model real agent actions directly: prompts, tool calls, commands, diffs, approvals, review comments, replies, and queued work.
The product should reduce ambiguity.
Users should be able to see:
Correctness needs to stay legible even when output speed increases.
Agent work is often multi-turn and multi-hour. Sessions, review comments, audit history, and queued work should survive interruption and restart.
Even when the first product is local-only, the core models should survive remote access later.
Messages, sessions, approvals, reviews, and audit trails should not depend on being on the same machine as the UI.
Remote pair programming and multi-user workflows only work if the single-user product is already clear, trustworthy, and structured.
The system should help high-output engineers stay aligned with peers instead of becoming isolated through speed.
Shared review, shared session context, shared audit, and pair-oriented workflows are part of the product, not extra polish.
A developer can leave a machine running agent sessions, open TermAl later from another device, review diffs, leave comments, ask for follow-up work, approve safe actions, reject risky ones, and bring in another engineer when collaborative judgment is needed.
The entire flow should feel like operating an intelligent engineering workspace, not wrestling with a terminal transcript.
TermAl began at the terminal, but its real destination is broader:
an operating environment for software engineering work done with AI agents.