ShipStation Global builds Deckhand, a Slack IT help desk agent on AWS

ShipStation Global has built Deckhand, an AI-powered IT help desk agent for Slack, on Amazon Bedrock AgentCore. According to AWS, the agent resolved 97.5 percent of incoming channel queries and earned a 96 percent positive feedback rating from employees over three months.
- ShipStation Global built Deckhand, an AI IT help desk agent for Slack, on Amazon Bedrock AgentCore.
- According to AWS, it resolved 97.5 percent of incoming channel queries with a 96 percent positive feedback rating over three months.
- Before launch, employees posted more than 1,500 threads in 12 months, and over a fifth of questions came outside US business hours.
- Deckhand uses the Strands Agents SDK, Anthropic's Claude in Amazon Bedrock, AgentCore Gateway with Lambda-based MCP tools, and a Bedrock knowledge base.
- Resolved Slack threads are mined with Recursive Language Models to add verified fixes to the knowledge base.
The problem: repeated questions and scattered answers
ShipStation Global describes itself as the leading intelligent shipping and logistics platform for small and mid-sized shippers. Its IT conversations took place in a busy help desk Slack channel, where employees asked and IT staff answered one thread at a time. The answers often already existed in a Confluence how-to article, on a company intranet page, or in a thread from months earlier. Finding them was the hard part.
The channel data showed the scale of the issue. In the 12 months before launch:
- Employees posted more than 1,500 threads, and volume kept growing.
- With staff in the US, UK, Spain, Poland, Australia and the Philippines, more than a fifth of questions arrived outside US business hours.
- Answers were split across Confluence articles, intranet pages and PDFs, IT-maintained guides, and the channel's own history.
Chris Collins, Senior Director IT at ShipStation Global, said the company believes that "tickets are an indication of operational friction not productivity," according to the AWS post.
What Deckhand does and how it is built
Deckhand answers routine IT and company questions from a curated knowledge base, reads error-message screenshots, and creates and updates Jira Service Management tickets. When it cannot resolve a question, it asks a clarifying question or offers to create a ticket from the conversation.
The solution has four main components:
- Slack service: a FastAPI and Slack Bolt application hosted on AWS App Runner. It receives Slack events, decides whether the bot should respond, and builds a user-context payload from the requester's Slack profile, including name, email, time zone and manager.
- Agent service: a Strands agent deployed as a container on Amazon Bedrock AgentCore runtime. It manages the multi-turn tool loop using Anthropic's Claude in Amazon Bedrock.
- Tools and knowledge: a single AWS Lambda function exposed as Model Context Protocol (MCP) tools through AgentCore Gateway. It classifies requests, creates and updates Jira tickets, and analyzes images. An Amazon Bedrock knowledge base indexes content from Confluence, the company intranet, Slack and marketing pages, stored in Amazon S3.
- Operations and feedback: Amazon DynamoDB stores feedback, AWS X-Ray traces request paths using OpenTelemetry, and services log to Amazon CloudWatch.
When an employee posts in a channel where Deckhand is a member, the Slack service assembles the full thread history and user context, then invokes the agent through the AgentCore runtime API. Because the employee's profile is included with every request, the agent can complete ticket details automatically.
A lightweight agent and a tool layer that changes independently
The team used the open source Strands Agents SDK, in which an agent consists of a model, a set of tools and a system prompt. The SDK manages the reasoning loop, letting the model select tools, process results and iterate until it produces an answer. According to the post, Deckhand's agent service takes only a few lines of Python. The model ID is set through an environment variable, so models can be switched without changing application code.
Knowledge base retrieval is a native Strands tool. Other functions, including classification, ticket management, user resolution and image analysis, are handled by the Lambda function registered on AgentCore Gateway, which routes each invocation by tool name. This separation allows tool logic to be updated independently, and the agent can access new tools from the Gateway without a rebuild. AgentCore runtime hosts the agent as a container, which avoids server or session management.
Knowledge base design and mining Slack history
Deckhand answers from a single Amazon Bedrock knowledge base with S3-backed data sources: Confluence articles, internal pages, PDFs and events from the company intranet, and resolved support threads mined from Slack. Each document carries metadata such as title, source URL, content type and event dates, which lets the agent filter content, for example when users ask about upcoming events or weekly lunch menus in the Austin office.
The post calls the archive of resolved support threads the most valuable data source, since it holds years of employees describing issues and IT walking them through fixes. It is also noisy, mixing banter, automated alerts and threads that never got resolved. Indexing an unverified fix would make the knowledge base surface an incorrect answer with the same confidence as a correct one, so the team says extraction needs to be precise. The mining uses Recursive Language Models (RLMs) in a three-stage pipeline, which begins with a script that exports the channel's full history through the Slack API.
What it means for IT teams
The case shows a pattern other IT teams may recognize: an agent that lives where employees already ask questions, draws on existing documentation and past resolved threads, and hands off to a ticket only when needed. ShipStation Global's own figures, 97.5 percent of channel queries resolved and 96 percent positive feedback, are reported by the company and AWS over a three-month period. The prerequisites listed in the post include an AWS account with access to Amazon Bedrock, AgentCore runtime and AgentCore Gateway, a Slack app, a container registry such as Amazon ECR, an S3 bucket with a Bedrock knowledge base, and IAM roles and Secrets Manager entries for credentials.


