AI Builder gives the builder workspace a real AI layer without making it feel complicated.

It already covers runtime control, provider/auth flexibility, and reusable prompt profiles. Use powerful memory-aware commands and stronger team policy without exposing private implementation details.

Mode-based presentation

Start from the view that matches the person reading the page

This page adapts for everyday product users, hands-on implementers, and company-level buyers without changing the underlying product direction.

Guided setup

Choose a runtime, provider, and profile without rebuilding the AI setup from scratch every time.

Reusable behavior

Prompt profiles and future command layers make the builder feel more consistent from session to session.

Native workspace

Lives inside the builder instead of sending users off into a separate AI tool.

Scales forward

Leaves room for stronger memory and governed workflows when the product grows into them.

Standard mode

AI Builder as a cleaner everyday workspace

Explain the value without going deep into runtime internals: better context, reusable setup, and a more natural place for AI inside the build flow.

Use AI Builder as a guided workspace instead of stitching together separate AI tools and one-off prompts

Pick a runtime, connect a provider, and reuse saved agent behavior without redoing setup every session

Move toward agents that remember more project context and follow more reliable default commands

Keep the interface understandable for product teams while still leaving room for deeper technical controls later

Start from a practical builder flow now, then grow into more advanced automation when your team is ready

Make AI-assisted building feel like part of the product instead of a sidecar chat window

What is already in place

Real capabilities in the current AI Builder baseline

Builder-native workspace

AI Builder mounts into the existing builder workspace instead of feeling like a separate bolt-on tool.

Runtime abstraction

Switch between Codex SDK and OpenCode-compatible runtime paths behind one consistent builder surface.

Provider flexibility

Use OpenAI, Anthropic, or Google model routing where the selected runtime supports it.

Auth mode control

Handle API-key and host-auth/device-auth flows without pushing the complexity into the user session itself.

Prompt profiles

Save reusable system prompt profiles and bind them to specific runtime/provider connections.

Forward-ready route surface

Dedicated health, generate, cancel, and Codex auth endpoints keep the plugin ready for deeper workflow migration.

What the next layer adds

The upgraded memory-and-command direction

Upgraded memory layer

Project memory is planned to flow into system prompts so the agent starts from stronger context instead of re-learning the project each session.

Default command packs

Teams will be able to define default command behavior so the builder arrives with the right baseline actions already wired in.

Custom command authoring

Create team-specific commands for repetitive workflows without exposing implementation details to every user prompt.

Enterprise prompt governance

Give companies a cleaner way to assign custom system prompt layers and policy framing across teams or client workspaces.

AI Builder plugin

Give your build workflow a real AI control layer

Start with the current plugin base, then evolve toward memory-aware commands and enterprise prompt policy when you are ready.

Expansion track

The next product layer is more than a better prompt box

The approved direction is a managed AI operating model for build work: commands, memory sources, versioned prompt behavior, approval controls, and reusable templates.

Command library and reusable packs

Ship with default commands, team-authored commands, and reusable prompt-command bundles so common workflows stop depending on one-off prompting.

Memory source controls

Choose which project, team, client, environment, and session memory layers feed the agent so context becomes intentional instead of accidental.

Prompt profile versioning

Compare, roll back, and safely roll out prompt profiles so teams can improve agent behavior without losing control of what changed.

Policy, approvals, and audit trail

Add role-based model and tool policy, approval gates for sensitive actions, and a clear audit layer for who used what profile and command.

Workspace templates

Start from repeatable build modes and team-specific operating presets instead of rebuilding the same setup across projects.

Session-aware continuity

Session resume already exists, so the product should surface continuation and state recovery more clearly instead of treating it like future work.

Product truths

Two things the page should say plainly

These are important parts of the AI Builder direction already established in the product, so the marketing layer should present them accurately.

Preview sessions stay isolated

Preview environments are intended to run on their own platform runtime and session database so experimentation does not leak into the live surface.

Session continuation is already part of the story

Resume and continuation should be presented as an existing capability that becomes more visible and useful as memory and command layers expand.

Built for the next layer too

AI Builder should feel powerful before it feels complicated

That means a clean workspace first, then stronger memory, command packs, and company controls layered in the right way.