AI Agents and Humans are getting better at completing individual tasks, but enterprise teams still face a coordination problem: their agents often live in separate tools, with context repeatedly copied between systems. SandboxAQ is taking aim at that gap with Switch, an open-source platform designed to bring humans and AI agents into the same collaboration channels.
The next challenge for enterprise AI may not be building smarter agents. It may be figuring out how to make multiple agents—and the people supervising them—work together without creating another layer of software.
SandboxAQ is betting on the latter.
The company has launched Switch, an open-source platform designed to connect AI agents with collaboration environments such as Slack, Microsoft Teams and Discord. Instead of requiring employees to move between an AI development environment and their normal workplace tools, Switch puts agents into shared rooms where people and machines can work with the same conversation history and context.
That is a subtle but important change in how enterprise AI could be deployed.
Most AI-agent platforms are focused on the agent itself: its model, tools, memory, prompts or ability to execute tasks. Switch addresses the layer around the agent—the place where humans decide what needs to be done, agents perform work, and colleagues review or extend the result.
The software is part of Flint AI, SandboxAQ’s portfolio of agent products. The company says Switch is platform- and model-agnostic, allowing organizations to connect agents built with technologies including Claude Code, Google ADK, LangChain and OpenAI rather than forcing teams into a single model ecosystem.
The project’s public GitHub repository describes Switch as infrastructure for creating organizations where AI agents and humans work side by side. Its current architecture supports collaboration across Slack, Microsoft Teams, Discord, Telegram and Mattermost, according to the repository.
That interoperability could be one of its more consequential features.
Enterprise AI is becoming increasingly fragmented. Developers may use one model provider for coding, another for research and another for specialized business workflows. Teams may also operate agents locally, in cloud environments or through different frameworks.
Without a coordination layer, those agents become isolated specialists.
Switch attempts to turn collaboration channels into shared workspaces for those specialists.
From AI Assistants to AI Teams
The concept reflects the industry’s broader movement from individual AI assistants toward multi-agent systems.
A conventional AI assistant generally has one user and one conversation. A multi-agent workflow can involve several specialized systems, each responsible for a different part of a task.
Imagine a software incident in which an engineer asks an AI coding agent to investigate a production problem. A second agent could analyze monitoring data, while another prepares a remediation plan. Instead of exchanging results through separate interfaces, the agents could operate in the same channel as the engineers managing the incident.
The same model could apply to marketing, legal, finance or product development.
A marketing team might use research, copywriting and design agents in a campaign channel. A legal team could bring document-review and drafting agents into a deal workspace. An engineering organization could assemble specialized agents around a live incident.
The advantage is not necessarily that any individual agent becomes more capable. The potential advantage is that context becomes persistent at the team level.
Switch uses rooms as the unit around which work is organized. The room retains its context, history, participants and rules as agents join or leave.
That addresses one of the less visible costs of enterprise AI: context transfer.
When work moves from one agent to another, humans often have to restate the problem, attach files or explain decisions already made. A shared workspace can reduce that repetition.
The idea is similar to how collaboration software transformed human teams. The difference is that the participants now include software agents that can perform actions.
An Alternative to the AI Platform Lock-In Model
SandboxAQ’s other major bet is interoperability.
The company says Switch does not require customers to replace their existing messaging platforms or AI agents. It provides a layer between them.
That puts the product in an interesting position relative to Microsoft, Google, Salesforce and other companies building increasingly integrated AI ecosystems.
Microsoft, for example, has a significant advantage because it controls both collaboration software through Teams and a broad enterprise software portfolio. Google can similarly connect Gemini to Workspace and its cloud platform. Salesforce is embedding agents directly into customer workflows through its Agentforce platform.
Those integrated approaches offer convenience and centralized governance.
Switch takes the opposite route: keep the existing collaboration environment and connect agents from different providers to it.
For enterprises concerned about model lock-in, that architecture could be attractive. A company could potentially change its underlying agents or model providers without rebuilding its collaboration environment.
But interoperability also creates a harder governance problem.
An enterprise deploying multiple agents needs to know which agent can access which information, what actions it can perform, what context is exposed to other agents and who is accountable when an automated workflow goes wrong.
The Switch GitHub project says guardrails and cost reporting are planned, while its current architecture includes roles, instructions, permissions and attached knowledge as part of its management layer.
That distinction matters for enterprise adoption. Open source can make infrastructure easier to inspect and customize, but it does not automatically solve governance.
Open Source, With a Commercial Caveat
Switch is available publicly through GitHub, and SandboxAQ says it can be deployed without a license fee. The repository is currently licensed under Apache 2.0 with a modified Commons Clause that restricts selling the software or derivatives.
That means calling Switch simply “open source” requires some qualification. The project provides source code and broad rights to use and modify it, but the Commons Clause limits commercial resale.
For enterprise technology teams, the more important question is likely deployment and control.
Switch can be self-hosted, allowing organizations to retain control over their infrastructure and data. Its architecture uses a Matrix-based room system, adapters for different collaboration platforms and an agent bridge that connects external agents to shared rooms.
That makes it closer to AI-agent collaboration infrastructure than another chatbot application.
The timing is significant.
Enterprise AI adoption is moving toward agents that can execute multi-step work rather than simply answer questions. As that happens, organizations will need infrastructure for coordinating agents, assigning permissions, retaining context and monitoring activity.
Switch is targeting that emerging layer.
The bigger question is whether enterprises will prefer an open, interoperable coordination layer or continue buying integrated AI ecosystems from the largest cloud and software vendors.
For SandboxAQ, the answer is clear: AI agents should not require teams to abandon the collaboration tools they already use.
If the company is right, the next enterprise AI interface may look less like a chatbot and more like a workplace channel—with humans and software agents working side by side.
Market Landscape
The enterprise agent market is separating into several layers:
- Foundation models: OpenAI, Anthropic, Google and other providers supply the underlying intelligence.
- Agent frameworks: LangChain, Google ADK and similar tools help developers build task-oriented agents.
- Enterprise AI platforms: Microsoft, Salesforce, Google and others combine models, data, applications and governance.
- Agent collaboration infrastructure: Emerging products such as Switch focus on coordinating multiple agents and people across existing workplace environments.
- AI governance and observability: Security, permissions, evaluation, monitoring and cost controls are becoming essential as agents gain autonomy.
Switch is targeting the fourth layer rather than attempting to replace the other three.
That could make it complementary to existing AI infrastructure, but it also means the product will need strong security, identity and governance capabilities as enterprise deployments become more complex.
Top Insights
- SandboxAQ’s Switch connects humans and AI agents inside Slack, Teams and other collaboration tools, creating shared workspaces instead of isolated agent sessions.
- The platform supports agents built with multiple frameworks and providers, targeting enterprises that want interoperability without committing their workflows to one AI ecosystem.
- Persistent room-level context could reduce the repetitive handoffs and briefing required when multiple agents collaborate on complex enterprise tasks.
- Switch enters a market dominated by integrated platforms from Microsoft, Google and Salesforce, making interoperability its central differentiation rather than model ownership.
- Enterprise adoption will depend on governance, permissions, security and cost controls as organizations move from individual assistants toward multi-agent workflows.
