Salesforce Launches Slack Code for Multi-Agent Programming
Salesforce CEO Marc Benioff announced that Slack Code is officially launched, allowing developers to collaborate with AI agents within the same Slack channel, initially integrating agents from Anthropic, GitHub, Cognition, and Vercel.
Benioff defined the product as "multi-person collaborative programming": human developers can raise requests, review results, assign tasks, and handle exceptions within the channel, while AI agents take on tasks related to code generation, codebase understanding, testing, deployment, or runtime environments. The product aims to shift code collaboration from one-on-one operations in IDEs and ticket systems to a traceable team communication interface.
The initial partners cover models, code hosting, agent-based development, and deployment chains. Anthropic provides Claude-related agent capabilities; GitHub manages code hosting, pull requests, and developer workflows; Cognition focuses on autonomous software engineering agents; Vercel is involved in front-end application development and deployment. Slack aims to serve as the human-machine collaboration and task orchestration layer among these different agents.
The key aspect of the product is not the generation of code by a single model, but rather enabling multiple tools to share context, collaboration records, and task statuses. Developers can retain discussions about requirements, code changes, review comments, testing results, and deployment feedback within the same conversation, reducing the need to transfer information repeatedly between Slack, GitHub, IDEs, CI/CD systems, and cloud platforms.
The launch of Slack Code coincides with Dreamforce, indicating that Salesforce is advancing Slack from a corporate instant messaging tool to an AI work entry point. If agents can read channels, access codebases, submit changes, and provide execution results through controlled permissions, Slack's value will extend from message distribution to a control panel for software development processes.
In market mechanics, development teams and corporate IT departments are the buyers of collaborative AI programming; model providers, code agents, code hosting, cloud deployment, and enterprise collaboration software are the suppliers. Salesforce can generate revenue through Slack seats, enterprise AI subscriptions, partner distribution, and workflow value-added services; tools that only provide isolated code completion, lack team collaboration entry points, or cannot access corporate permission systems are under pressure. Actual penetration will still depend on code security, access control, audit records, agent execution permissions, and corporate procurement cycles.
Source: Public Information
ABAB AI Insight
Salesforce's path in AI products has evolved from the Einstein prediction model, Einstein GPT, to Agentforce, which is based on Data Cloud. After acquiring Slack, it first used it as an internal collaboration entry point, then attempted to transform Slack Channels into execution interfaces for business agents. Slack Code continues this strategy: Salesforce does not need to win the foundational model competition itself but aims to integrate external models and agents into its software distribution system by connecting corporate identities, CRM data, message flows, and workflow permissions.
The focus of the capital path is on "entry control" rather than monopolizing models. Microsoft embeds GitHub Copilot into GitHub, VS Code, and Azure; Amazon connects models and cloud resources using Bedrock, CodeWhisperer, and AWS development tools; Google integrates Gemini into Cloud, Workspace, and the Android ecosystem. Salesforce's choice to collaborate with Anthropic, GitHub, Cognition, and Vercel indicates its attempt to become a multi-agent scheduling layer: clients initiate tasks within Slack, while models, code tools, and deployment platforms compete for execution shares in the background, with Slack managing conversations, permissions, and organizational relationships.
A historical analogy can be drawn from how Jira and GitHub transformed software delivery management. Jira standardized requirements, defects, and project management, while GitHub networked version control, code review, and open-source collaboration; Slack Code attempts to compress "requirement raising—code generation—review—deployment—feedback" into real-time channels. The industry is transitioning from the auxiliary programming phase to the agent collaboration phase, but once the code generated by agents can trigger production deployments, team collaboration tools must assume approval, accountability, and audit functions, rather than just serving as chat interfaces.
Essentially, this represents a restructuring of the industry chain. The value of the past development toolchain was dispersed across IDEs, code hosting, project management, testing, and cloud deployment; agent-based programming connects these stages into a continuous execution chain, concentrating value on platforms that possess identity permissions, work context, and task distribution rights. If Slack becomes the default interface for agents to receive instructions and report results within enterprises, Salesforce can regain part of the entry point into development workflows; however, if GitHub, Microsoft Teams, or AI-native IDEs control deeper code and deployment contexts, Slack's coordinating position will be limited.
ABAB News · Cognitive Laws
The more AI agents there are, the more valuable the scheduling entry becomes.
Writing code will become commoditized; controlling deployment is where power lies.
The ultimate goal of collaboration tools is not chatting, but execution.