Amazon has rolled out a new version of its Seller Assistant AI tool, which now includes advanced features such as automated price monitoring, stock management, and account management. This update is integrated into two popular AI platforms, Quick and Claude, through a new plugin. Initially launched in 2023, Seller Assistant was limited to answering questions and displaying sales data. However, it has since evolved to include agent capabilities, allowing it to perform tasks behind the scenes. According to Chris Jones, director of Seller Assistant and an AI specialist with nearly two decades of experience, the tool has expanded from a few AI interactions to covering a full range of seller activities. Now, Seller Assistant retains each seller’s pricing habits, stock cycles, and growth goals, and it can access this information even outside of Amazon’s Seller Central. It also includes a “canvas,” a visual interface where the AI can simulate scenarios, compare pricing strategies, or analyze advertising campaigns using real-time data. Each recommendation includes reasoning, and a subtle prompt reminds sellers to check for possible errors. One of the most notable new features is that Seller Assistant can now operate at night, but it requires permission to do so. This marks a shift from a reactive tool to one that monitors, interprets, and acts continuously, even when the seller is offline. Amazon also offers pre-built models for tasks like monitoring account health, summarizing orders and returns, analyzing prices, and discovering new products to sell. These models can be customized with plain language instructions. For example, a seller could set up an alert if one of their top-selling products drops below four stars in reviews or track competitive pricing in their main category to adjust their listings. Despite these advancements, some sellers may hesitate to let AI handle pricing decisions while they sleep. Chris Jones, however, emphasizes that sellers have full control over the autonomy level of each task, from simple suggestions to automated actions. Every action is recorded in an audit log, and for high-impact decisions, seller approval is required. The goal is to reduce the daily workload on sellers, not to remove their strategic input. Amazon plans to gradually increase the level of autonomy as trust builds. However, the risk of errors—common in AI systems—remains. To mitigate this, Amazon has added a secondary AI agent that acts as a critic, reviewing and correcting the first agent’s actions. Other monitoring systems track the AI’s performance over time. Jones notes that Seller Assistant is not just a single model but a complex set of models working together. French sellers will have to wait for the full rollout of the plugin, as it is currently available only for sellers in Amazon’s American stores. Amazon has chosen to integrate its tools into third-party AI platforms through a plugin called Amazon Selling Partner, which allows data, recommendations, and action capabilities to be used by external assistants. The plugin is first available in Quick, Amazon Web Services’ professional AI assistant, and in Claude, a well-known AI assistant from Anthropic. The integration with Claude takes about sixty seconds and requires no coding. Amazon assures that its security measures remain in place, including limited access, human validation of actions, and full traceability. Other integrations are expected in the future. To demonstrate the system, Jones highlights Richard, a small sneaker accessory company owner who created his own AI agent named Lisa using Amazon Bedrock. Lisa helps Richard prepare for upcoming product releases by analyzing web data and estimating stock needs. Setting up such an agent takes only a few minutes, and Jones assures that anyone can do it. For now, Amazon is proceeding cautiously, aiming to build trust with sellers before granting more autonomy to their AI assistants.