Gorgias AI Agent 3.0 acts on orders. Check your platform first.

Gorgias positions AI Agent 3.0 around conversational commerce, and the claim is that it does not just answer questions, it acts: orders, refunds and returns, across channels, coached by Gaia.

That is a real shift. An agent that answers is a knowledge problem. An agent that issues a refund is a permissions problem, and those are configured very differently.

Get the vocabulary right first

This area has been renamed enough to break internal runbooks.

• The current version is AI Agent 3.0. If your notes say 2.0, they are a version behind.
• The framing is conversational commerce. There is no Gorgias product called a Conversational Storefront.
• AI Agent covers two jobs: Support Agent for post-purchase and Shopping Assistant for pre-purchase.
• Gaia is a separate product, an AI coach that audits your agent and generates Skills for it. It is not a synonym for Skills.
• Vision is a capability of AI Agent, not a standalone product. Since July 24, 2026 it reads images in chat as well as email.

The constraint that decides whether any of this applies to you

AI Agent is Shopify only. So is Order Management, the feature that lets an agent act on an order inside the ticket. Gorgias documents this explicitly as unsupported on BigCommerce, Magento, WooCommerce and PrestaShop.

Flows work on Shopify plus BigCommerce and Magento. Article Recommendation works on BigCommerce and Magento but not Shopify, which surprises people, because the pattern is not a simple tier where Shopify gets everything.

So if you are not on Shopify, an agent that acts on orders is a custom build: order, shipping and refund state pulled through the API, surfaced in the sidebar, with the actions your policy allows wired to the systems that own them. That is normal work. It just needs to be in the scope and the price before anyone starts.

What acting on orders actually requires

• A written policy on what the agent may do unsupervised. Refund under a threshold is a decision, not a setting.
• Guardrails that fail closed. When the agent is unsure, the correct behaviour is handing over, not attempting.
• Escalation your team can absorb. Containment without a clean handoff moves the complaint rather than resolving it.
• Knowledge the agent reads that is actually current, because an agent grounded in a 2023 policy page will apply a 2023 policy.

On deflection numbers

We do not quote a percentage before seeing your tickets. Across the Gorgias instances we have configured, containment lands between 26 and 56%. That is our own measured range out of our own builds, not a Gorgias figure. Gorgias's own published claim is higher. What moves your number is knowledge quality, escalation design, and how much authority you hand the agent.

Verified against

• Gorgias product launch page, AI Agent 3.0, built for conversational commerce.
• Gorgias docs, compatibility of automation features with ecommerce store providers.
• Gorgias docs, AI Agent explained, and Skills explained.
• Gorgias product updates, Vision in chat, July 24, 2026.

Checked against Gorgias's own pages on August 9, 2026. Gorgias does not publish a launch date on the AI Agent 3.0 page, so none is quoted here.