Amazon used its Accelerate conference to push its seller tools in two directions at once. On the software side, Seller Assistant has moved from a chat window that answers questions to an agent that performs multi-step work inside Seller Central, builds its own task screens through a feature called Canvas, runs recurring jobs in the background, and reaches outside Amazon through Amazon Q (Quick) and a private-beta plugin for Anthropic's Claude Code. On the physical side, Amazon Global Logistics (AGL), the company's first-party freight forwarding service, reports a 20% year-over-year cut in China-to-US West Coast transit time, a new Amazon Managed Placement (AMP) at Origin option, expanded hazmat coverage, standalone customs brokerage, automated paperwork, and human-staffed support. For sellers who manufacture in Asia and sell in the US or Europe, the combined effect is that more of the routine decisions about inventory placement, shipping documents, ad budgets and listing fixes can be handed to Amazon systems, with the seller stepping in at approval points.

Seller Assistant moves from answering to acting

Amazon's AI lead for the Seller Assistant team describes the change as a shift from an advisory chat interface to an agentic engine that executes workflows on a seller's behalf. The company frames this as packaging 25 years of its own e-commerce experience into tools that give small and mid-sized sellers the kind of automation previously limited to very large companies.

Two demonstrations show what "acting" means in practice. In the first, Seller Assistant links a newer version of a product to its original ASIN (Amazon Standard Identification Number, the catalog ID for each listing). The task card shows the original and newer ASINs, a submitted status, and a processing time of up to 24 hours. Before publishing the change, the system runs backend checks on the product relationship and presents an Approve or Cancel prompt. In the second, Seller Assistant runs an excess inventory analysis and charts "weeks of cover" for product variants that need attention. It also turns raw sales and order data into an interactive dashboard with three-month sales trends, monthly revenue, monthly units sold and key findings.

The system also draws on seller habits. If a seller regularly launches lightning deals during Prime events, Seller Assistant can analyze the history, suggest a promotional schedule and forecast what repeating a past deal structure might produce. These actions work because Amazon feeds promotional engines, advertising APIs and other specialized tools into the model's action space.

Compliance questions are a quieter but useful case. Checking whether a pesticide-related keyword is allowed, or what the updated coupon eligibility rules say, previously meant digging through help documentation. Seller Assistant queries Amazon's internal knowledge bases directly and returns a plain-language summary.

The harness behind the model

Amazon attributes the accuracy of these answers to what it calls the "harness": the software layer that binds the language model to persistent memory, security guardrails, Seller Central APIs, internal database functions and catalog data. The company's position is that this setup keeps the model from inventing policy interpretations and instead grounds answers in the actual system rules and the specific seller's context.

Canvas builds the screen for the task

Canvas is the most visible new feature in Seller Central. Amazon's reasoning is that every seller has a different workflow and static navigation menus cannot serve all of them. Rather than sending a seller through nested menus, Canvas renders a custom interface in real time based on a natural-language request.

Demonstrated uses include diagnosing an ASIN linking error and generating a step-by-step resolution screen, and presenting a single-click approval modal for a catalog adjustment. During a retail peak, Canvas can assemble an end-to-end execution board that covers promotional setup, inventory replenishment milestones and ad campaign scheduling in one place. Configuring promotions, coupons and discounts through the standard menus has historically been tedious, so this is the kind of task where a generated interface appears to save the most clicks.

Interface panels assembling above a monitor as a helper figure places the last card, with a clock and envelopes behind

▲ Canvas generating a task-specific interface

Background workflows: the agent as a digital employee

Beyond one-off tasks, Seller Assistant now supports background Workflows. The idea is to treat the assistant like an autonomous employee: give it an ongoing assignment, let it run, and ask it to request approval only at critical decision gates. A demonstrated schedule screen shows a recurring instruction to send a dashboard report every morning at 7 a.m., with the system verifying data feeds and metrics on its own.

Two workflows Amazon highlights:

  • Monitor top-performing ASINs continuously, alert the seller when a budget constraint appears, and recommend adjustments to PPC (pay-per-click) advertising spend based on sales velocity.
  • Aggregate customer return reasons, identify recurring defects or points of confusion, and propose revised listing bullets that address them.

For brand operators, the return-sentiment workflow appears to be the more useful of the two, since it closes a loop that usually requires reading return reports by hand and then making separate listing edits.

Taking the agent outside Seller Central

Amazon says it does not want to force sellers into a single interface. Instead it is embedding Seller Assistant's capabilities into tools sellers already use.

  • Amazon Q, referred to as Quick, is an agentic workspace where users connect different data sources and automate tasks. Sellers who register by December 31 receive a free 12-month subscription to the Quick Plus plan for themselves and up to two team members.
  • A plugin for Claude Code, Anthropic's coding agent, is in private beta. It brings Seller Assistant into developer workflows, so a seller or their technical staff can combine Amazon store data with Claude's reasoning. Amazon has not given a public launch date but says broader availability is coming very soon.

Seller Assistant uses the same Selling Partner API (SP-API) that custom software builders use, layered with Amazon's own business intelligence and orchestration models. Amazon's collaboration with Anthropic dates to 2023 and, by the company's account, shaped how it designs agentic software for sellers. Looking further out, Amazon's AI lead expects business management to rely increasingly on interconnected groups of specialized agents working across multi-vendor software stacks.

Where seller data goes

Connecting an Amazon account to an external model raises an obvious question about data exposure. Amazon's answer has three parts:

  1. Processing inside Seller Assistant and its MCP (Model Context Protocol, a standard for connecting AI models to external tools and data) connections runs within AWS infrastructure.
  2. Seller data is used only on behalf of that seller to optimize their account and is not used to train public foundation models.
  3. The plugin architecture gives sellers granular, visible controls over which data points are shared with which external systems.

Amazon also advises sellers to stay deliberate about which third-party tools they authorize to access their catalog and performance metrics. That caution applies regardless of how secure any single integration is.

Logistics: faster lanes and a new placement option

AGL moves goods from overseas manufacturers directly into Amazon fulfillment centers, with key lanes from China and Vietnam into the US and routes into the UK, EU and Japan. The head of AGL describes the freight market over the past 6 to 12 months as turbulent, with rising prices and constrained ocean capacity. Against that backdrop, AGL reports:

Metric Figure reported by Amazon
China-to-US West Coast transit time Down roughly 20% year over year
Seller logistics costs after switching to AGL Down about 15%
Premium Ocean, China to UK and EU About 11 days faster than standard ocean
Seller-Managed Placement share of AGL bookings 30% to 40% in recent weeks
AGL's rank on the transpacific eastbound lane Largest NVOCC forwarder, first half of 2026

An NVOCC (non-vessel operating common carrier) is a forwarder that sells ocean capacity without owning ships. AGL's product lead stresses that the 20% figure is measured data rather than a target, and that most of the saving comes at the destination by choosing fulfillment facilities closer to ports. Faster transit also means working capital spends less time sitting on water or rail.

Container ship at a port with trucks fanning out to five warehouses while pallets are split at an origin warehouse

▲ Origin-based inventory split across US regions

Choosing between SMP and AMP at Origin

Inbound placement forces sellers to balance delivery speed across the network against placement fees. AGL frames this as a choice between keeping control and handing it off.

Option How it works Best suited to
Seller-Managed Placement (SMP) Seller chooses where inventory goes and ships directly into the five US fulfillment regions High-volume sellers moving multiple full containers
SMP Express Within the Seller Central booking flow, seller selects specific regional cross-docks with faster inbound processing Seasonal sellers who cannot miss check-in windows
Amazon Managed Placement (AMP) at Origin Seller makes a single booking into an AGL origin facility in China; Amazon watches US demand and splits, routes and documents shipments into multiple regions under one all-inclusive fee Sellers who want hands-off inbound replenishment

AMP at Origin has a second benefit. Warehousing at the China origin facility is substantially cheaper than US domestic storage, so a seller can hold buffer stock there and drip units into the US network just in time based on sales velocity, avoiding long-term storage surcharges. The service integrates with Amazon Warehousing and Distribution (AWD).

A board and card game company illustrates the SMP side. Founded by a couple who first designed games for family, the business went viral on social media, outgrew shipping from a parent's garage, moved to FBA, and over the past 12 months adopted SMP to send inventory straight to regions rather than through consolidation hubs. Because roughly 80% of its annual sales land in the holiday quarter, securing space through SMP Express functions as insurance against missed check-in windows.

Hazmat, customs, documents and human support

Several smaller AGL changes remove friction that overseas sellers know well.

  • Dangerous goods: AGL now carries hazmat items such as aerosols and lithium batteries, covering about 97% of FBA hazmat selection, with a target of 99% by the end of 2026. Sellers no longer need to split catalogs across multiple forwarders for restricted goods.
  • Amazon Customs and Trade (ACT): a first-party, licensed, C-TPAT certified customs broker (C-TPAT is a US Customs and Border Protection supply-chain security program). In 2026 ACT became available standalone, so sellers using another forwarder can still use Amazon for clearance. Amazon says customs enforcement grew markedly stricter in 2026 and more importers moved to ACT for compliance, though it believes the biggest benefit comes from pairing ACT with AGL ocean transport.
  • Online Document Management (ODM): an international shipment typically needs 27 to 45 documents. ODM auto-generates core shipping documents and pre-fills known account and catalog data, cutting the documentation burden by about 50% and reducing errors that trigger inspections.
  • Connect Pro: threaded communication with human logistics operators in the US and China. Over the past year it shifted from reactive tickets to proactive monitoring that flags delays and customs exceptions before the seller notices. Amazon positions it explicitly as an alternative to chatbot support during critical escalations.

AGL's leadership acknowledges that the service is not superior in every scenario and that its software tooling still has room to improve, but says most sellers see measurable gains, and that despite its volume rank AGL still handles only a fraction of total FBA inbound freight.

What changes for sellers outside the US

For a seller outside the US who sources from China or Vietnam, the practical shifts look like this:

  • Inventory placement: instead of calculating regional splits, a seller can book once into the China origin facility under AMP at Origin and let demand signals drive distribution, or keep control with SMP if volumes justify full containers.
  • Shipping: Premium Ocean shortens the European lead time by about 11 days, which matters most ahead of promotional events; SMP Express targets faster cross-docks for peak-season check-in.
  • Paperwork and compliance: ODM and ACT reduce document preparation and clearance risk even when freight moves with another forwarder.
  • Brand operations: Seller Assistant workflows handle ad-spend monitoring and return-driven listing edits in the background, and Canvas consolidates peak-season setup into one board.

Key takeaways and next steps

Amazon is positioning Seller Assistant as an agent that works where sellers already are, backed by an AWS-contained harness and explicit data controls, while AGL is competing on measured transit and cost figures and on removing the hazmat, customs and documentation chores that used to require several vendors.

Sellers who want to act on this can:

  1. Claim the free one-year Quick Plus plan before December 31 and test agent workflows with up to two colleagues.
  2. Try Seller Assistant on a bounded task first, such as a compliance keyword check or an excess inventory analysis, then set up one background workflow with approval gates.
  3. Review which third-party tools and plugins have access to account data before adding MCP or Claude Code connections.
  4. Find Amazon Global Logistics in Seller Central or by searching for it online, and ask Amazon's logistics team to benchmark historical shipment data against AGL rates and transit times.
  5. Run a single trial shipment, choosing SMP, SMP Express or AMP at Origin based on volume and how much placement control the business needs.