# Antigravity Knowledge Work: The Two-Gear Merchant Guide **Plan carefully. Execute bounded work. Keep the merchant in control.** OCTOplux · 11 November 2026. **YouTube description hook:** Switch to the document-first layout, learn live voice, and copy the two-gear workflow for product pages and blog batches. ## The quick layout switch In an eligible Antigravity App account on v2.30.0 or later, open **Settings → General → Interface Layout → Knowledge Work**. If Settings opens a project-specific panel, select the global/general scope. The layout emphasizes conversations, documents and folder outputs and hides developer-oriented controls. It does not change the agent’s underlying permissions or make its actions read-only. Documentation differs on some business-account availability; if the selector is missing, check your installed version and account-specific documentation rather than assuming the feature is broken. [Settings](https://antigravity.google/docs/settings), [getting started](https://antigravity.google/docs/cli/tutorial/). The switch itself can be quick; “ten seconds” assumes you are already installed and signed in. Installation, updates, account access and connected-service permissions are separate steps. ## Voice is a conversation, not an automatic approval In the conversation composer, choose **Start Voice mode** and select a voice if prompted. Gemini Live enables a spoken exchange while the underlying agent performs the work in that conversation. Dictation is different: it transcribes a prompt for you to review and send. Keep the screen visible when reviewing a store change, and ask the agent to show the target and preview before acting. [Voice mode](https://antigravity.google/docs/voice-mode). Try saying: > “Help me improve this one product description. Show the plan and preview first. Keep the current store unchanged until I approve the exact update.” Voice access does not grant permission to browse unrelated files or publish to a store. Confirm that the store connection and selected target are the ones you intend. ## Turbo is a permissions choice Turbo is documented as allowing unrestricted command execution and broad filesystem access. It is not a model-speed setting, quota-saving switch or requirement for Knowledge Work. You can find permission presets under **Settings → General → Global Permissions → Permission Preset**, with account/version-specific variations. A simpler interface can conceal developer controls without reducing what the agent can do. [Agent settings](https://antigravity.google/docs/agent-settings), [permissions](https://antigravity.google/docs/permissions). For a merchant’s first trial, keep the normal permission preset and a clearly scoped project. If you deliberately choose Turbo for trusted local work, understand its effect first and preserve backups; do not describe it as the safety mechanism for a live store. This guide does not change your settings. ## Gear A: decide the hard parts once Google announced Gemini 4 Carbon in Antigravity on 10 November 2026, with access depending on plan. Carbon is suited to deeper planning/verification; Argon is another capable choice for complex work. Use whichever model your account actually provides. The two-gear method below is an OCTOplux workflow recommendation, not a Google guarantee of a particular quota reduction. [Carbon announcement](https://antigravity.google/blog/gemini-4-carbon-in-google-antigravity). Ask Gear A to produce a short, durable specification: 1. Exact store and resource boundaries. 2. Approved facts and missing evidence. 3. Section order and image requirements. 4. Link/card rules and schema policy. 5. Acceptance criteria and restore procedure. 6. One complete sample for merchant review. **Gear A prompt:** > Plan one product makeover and a ten-topic supporting content plan using the approved catalog. Decide the layout, fact constraints, card rules and validation steps. Produce a compact execution brief and one preview sample. Flag unsupported claims, external costs and missing permissions. Do not publish or launch the remaining batch. ## Gear B: execute a fixed brief Gemini 3.8 Flash appears in the current model catalog. Choose the reasoning level available in your account; **High**, when offered, is appropriate for a task that needs more checking, but the visible picker may default to Medium. Do not assume the exact label “Gemini 3.8 Flash - High” appears identically in every interface. A model change normally takes effect for the next turn rather than replacing an already-running turn. [Model selection](https://antigravity.google/docs/models). Give Gear B the approved brief, sample and a small, explicit batch. Avoid making it rediscover the strategy for each article. Use a concise exception log to return ambiguous decisions to Gear A. **Gear B prompt:** > Execute the approved brief for these selected items only. Match the approved sample, preserve source facts, apply the card/link rules and report validation results per item. Do not broaden the batch, invent missing data, purchase services or publish. Stop an individual item when its facts or target are ambiguous, record the exception, and continue only independent approved items. ## A practical quota ledger Record task, model, reasoning level, selected batch size, reported usage, accepted outputs, revisions and elapsed time. Compare **usage per accepted item**, not usage per first draft. A cheaper model that needs many retries can be a false saving. Keep the shared brief small and include only the necessary product facts. Use deep reasoning for planning and failures, not every repetitive formatting step. Do not start a swarm solely to save quota; Google warns parallel agents can consume it faster. ## Your non-coder handoff Before a batch, you should be able to read five plain-language statements: what changes; which store/items; what it costs; what must pass; and how to restore. Ask for screenshots or an HTML preview if code is not useful to you. Hidden terminals do not replace these checks. **Next step:** explore [OCTOshopX](https://octoplux.com) for the merchant workflow and use this guide with the supported build. Single-store launch pricing is $399.90 once or $99.98 × 4 ($399.92 total), then $0/month software subscription. Antigravity plan access, quotas and applicable model/API charges are separate. No upgrade is required merely to read or use the planning worksheet.