Meta Switches to Slack for Internal Agents, Citing Slack as the Current 'Strongest Agent Platform'
Meta's Chief AI Officer, Alexandr Wang, stated in an internal memo that the company will migrate its internal communications from Google Chat to Slack, citing Slack as the current 'strongest agent platform.'
Wang listed key reasons in the memo, including: Slack's conversational interface designed for agents, mature developer tools, and strong third-party integration capabilities. Meta's assessment is not merely that Slack is better suited for employee chatting than Google Chat, but that it more easily accommodates the development, invocation, and management of AI agents by different internal teams.
The memo noted that not all Meta employees are currently developing agents, but the company believes that establishing a reliable and practical agent ecosystem will benefit all employees. Meta has not disclosed the completion date for the migration, the number of employees affected, existing Google Workspace contract arrangements, Slack subscription levels, or the list of initial agents to be introduced.
Slack's third-party integration capabilities mean that agents can connect to code repositories, work orders, calendars, knowledge bases, data warehouses, and enterprise applications within message threads, retrieving information, generating results, or triggering authorized workflows after user commands. The value of this capability depends on the actual tools accessed, identity permissions, audit trails, and manual approval mechanisms, rather than the chat interface itself.
This choice elevates Meta's relationship with Salesforce from a typical enterprise software procurement to a collaboration on agent infrastructure. Slack is owned by Salesforce, which is embedding AI agents into enterprise collaboration scenarios through products like Agentforce and Slack Code; Slack Code has already supported teams in invoking coding agents such as Claude Code, Devin, GitHub Copilot, ChatGPT, and Vercel within project channels.
Meta still retains its self-developed Llama model, internal data platform, development toolchain, and multiple AI projects, so this migration does not signify that Meta is abandoning its own AI development or adopting Salesforce's models. A more accurate understanding is that Meta has chosen Slack as the messaging entry point for collaboration between humans and internal or third-party agents, rather than outsourcing its foundational models, training infrastructure, or core AI strategy to Salesforce.
From a market mechanism perspective, Meta's migration of internal communication tools as a large tech company will bring high-value enterprise seats, agent invocation volumes, and ecosystem demonstration effects to Slack; Salesforce benefits from gaining endorsement from a large client at the AI workflow entry point. Google Chat faces competitive pressure as enterprise collaboration tools shift from 'bundled meetings and messaging' to 'integrability of agents'; third-party software vendors developing tools for Slack, deploying identity controls, and managing enterprise agents may also see increased demand for access.
Source: Public Information
ABAB AI Insight
Meta has long built its own development and collaboration infrastructure internally and competes for the model layer ecosystem with its Llama open weight strategy; however, Wang's choice of Slack indicates that model companies do not necessarily need to build products at every workflow entry point. The most challenging part of enterprise agents is often not generating text, but integrating models into the conversations, permissions, tasks, documents, and approval systems that employees already use. Slack has accumulated channels, threads, application directories, OAuth permission models, bot interfaces, and enterprise auditing capabilities over years of development, and the migration essentially buys into these long-formed workflow assets.
The capital path shifts from selling collaboration software by seats to expanding based on agent invocation, connectors, automated tasks, and governance functions. If Slack becomes the default entry point for agents, model vendors, developer tools, databases, search, and SaaS products will need to pay integration, distribution, or usage costs to enter this messaging interface; Salesforce does not need to own every model but can accumulate value by controlling the middle layer of 'where agents receive tasks, which enterprise tools they invoke, and who reviews the results.'
This can be directly contrasted with Microsoft Teams and Copilot's path: Microsoft bundles collaboration, productivity, and AI invocation through Teams, Microsoft 365, Entra identity, and Azure; Google has native entry points with Workspace, Gemini, and Google Chat. Meta's migration from Google Chat to Slack indicates that enterprises are beginning to prioritize the developer experience for agents, the density of third-party tools, and multi-model access on par with or even higher than messaging stability when purchasing collaboration platforms. Slack Code's support for Claude, Devin, GitHub Copilot, ChatGPT, and Vercel is part of the competition for this model-neutral development entry.
Essentially, this represents a reconstruction of the industry chain: AI competition is extending from 'who has the strongest model' to 'who has the entry point for employees to initiate work, validate results, and invoke enterprise data.' Models will continue to evolve, but once teams embed channel structures, permission policies, business notifications, agent records, and approval flows in a particular collaboration platform, the switching costs will rise quickly. Slack is not just competing for a one-time chat software replacement order, but for the control plane in the era of enterprise agents; Meta's migration provides a rare validation from a large tech company for this positioning.
ABAB News · Cognitive Laws
Models determine capabilities, work entry points determine revenue
Agents are not lacking answers, but access permissions
Once chat tools take over tasks, they become operating systems