AWS re:Invent 2025 - Slack securely powers internal AI dev tools with Bedrock and Strands (AIM3309)

Slack's Journey with Generative AI and Agents

Slack's Developer Experience AI Journey

• Slack's developer experience team, consisting of 70-80 people, was tasked with building tools to make Slack engineers more agile and productive.
• The team started by building a "buddy bot" to help with documentation and answering developer questions more quickly.
• Slack then experimented with prototypes and hackathons using Amazon SageMaker, proving the art of the possible with generative AI.
• In 2024, Slack moved to Amazon Bedrock, a fully managed foundation model layer, which simplified infrastructure management and reduced costs by 98%.
• Slack's first bot built on Bedrock was the "Buddy Bot", which leveraged knowledge bases to provide faster answers to developers.
• Developers then requested coding assistance, leading Slack to experiment with Anthropic's Clot Code and Cursor models on Bedrock.

The Impact of AI-Assisted Coding

• With AI-assisted coding using Cursor and Clot Code on Amazon Bedrock, Slack saw:
  • Accelerated developer productivity
  • Faster innovation and reduced prototyping time
• Metrics showed:
  • 99% of Slack developers using AI assistance
  • 25% consistent month-over-month increase in PR throughput
  • 5,000 escalation requests per month handled by the AI assistance bot
• Slack also faced challenges, such as increased code review time due to the increased surface area, which they are actively working to address.

Slack's Journey into Agents

• Slack's move to agents was driven by the need to automate workflows, enable more complex reasoning, and leverage their existing tools and data sources.
• Slack leveraged the agentic capabilities in Clot Code, including sub-agents and planning, as a starting point.
• Slack also built their own MCP (Model Context Protocol) servers to standardize access to their tools and data sources.
• Slack explored various agentic frameworks, ultimately choosing to use the open-source Strands framework.

Strands: Slack's Agentic Framework

• Strands is an open-source, model-agnostic, and flexible framework for building agents, developed and open-sourced by AWS.
• Key features of Strands include:
  • Model and deployment choice flexibility
  • Highly flexible with built-in guardrails and observability
  • Integration with MCP for accessing tools and data sources
  • Support for multi-agent patterns like swarm, graph, and workflow
• Slack used Strands as the orchestrator agent, with Clot Code sub-agents performing specialized tasks.

Slack's Escalation Bot: A Strands-powered Solution

• Slack built an escalation bot using Strands as the orchestrator agent and Clot Code sub-agents.
• The Strands-powered escalation bot:
  • Leverages Temporal for reliable workflow orchestration and state management
  • Integrates with internal MCP servers for secure access to sensitive systems
  • Optimizes performance by running sub-agents in parallel and managing token usage
• Slack's vision is to establish fully automated agentic workflows across the entire development cycle, integrating more internal tools via MCP.

Key Takeaways

• Slack's journey demonstrates the power of generative AI and agents in enhancing developer productivity and innovation.
• By adopting Amazon Bedrock and the open-source Strands framework, Slack was able to simplify infrastructure management, increase flexibility, and drive adoption.
• Slack's escalation bot showcases how agentic workflows can automate complex tasks, improve reliability, and optimize performance.
• Slack's experience highlights the importance of a model-agnostic, extensible, and observability-focused approach when building agent-based solutions.