Iterate is a notable example of an 'agent-as-infrastructure' play within the ecosystem. By centering their entire configuration around Git and the Model Context Protocol (MCP), they are actively pushing the industry toward interoperable standards rather than proprietary integration hubs. This makes them a key partner for anyone building MCP servers, as Iterate provides an immediate, multiplayer interface for those capabilities.
They occupy the orchestrator layer of the agent stack, focusing on the bridge between LLM reasoning and operational execution in tools like GitHub and Linear. For the agent community, Iterate represents a movement toward 'White Box' AI, where the agent's logic, prompts, and tool access are fully transparent and manageable via standard developer workflows.
Iterate is a bet on the idea that as software becomes cheaper and more customizable, traditional SaaS stacks become obsolete. Founded in 2024 by Jonas Templestein, the former co-founder and CTO of Monzo, the company is building what they describe as the "world’s most hackable AI agent." Based in London, the founding team brings a background in scaling high-compliance, operationally intensive startups to the problem of business automation.
At its core, Iterate is an AI agent that lives in Slack. Unlike most agents that provide a proprietary web dashboard for configuration, Iterate is controlled through a Git repository that the user owns. This design choice is intentional. It treats agent instructions and tools as code, allowing teams to use familiar version control workflows to manage how their agent behaves. If a team wants to change the system prompt or add a new rule, they push a change to their repo. This approach provides a level of transparency and auditability that is absent in most commercial agent platforms.
The product is built to be an active participant in a team's existing workspace. It supports the Model Context Protocol (MCP), which allows it to connect to any MCP-enabled server to pull in context or take actions. This enables Iterate to handle tasks that span multiple tools, such as syncing GitHub pull requests to Linear tickets, summarizing interview notes in Notion, or creating issues from bug reports.
By leveraging MCP, Iterate avoids the "walled garden" problem. Users can build their own local tools and expose them to the agent without waiting for an official integration. This extensibility is central to their "hackable" branding. The agent is multiplayer by default, meaning it can be summoned in any Slack channel or thread to assist the entire team simultaneously.
Iterate's pricing model is a departure from the standard per-seat SaaS subscription. They charge the raw token cost plus a 50% margin. They justify this by pointing to the visibility they provide; users can see every token used and every trace produced by the agent. By giving users full control over the system prompt, the company argues that users are responsible for their own token economy. If a prompt is inefficient, the user has the power to fix it.
The company is currently in beta, offering a $50 monthly usage credit to early users. While the technical barrier to entry is higher than a typical chat-based AI—requiring a GitHub repository and an understanding of system prompts—the target audience is specifically "operators" and technical teams who are already comfortable with these tools. The project is also open source, allowing teams to inspect the underlying logic or potentially self-host, further distancing itself from the "random web app" model of AI service delivery. By focusing on Slack as the primary surface, Iterate acknowledges that for an agent to be useful, it must exist where the work and the conversation are already happening.
An open-source, multiplayer Slack agent configured via Git and the Model Context Protocol.
Iterate is hiring.