Why Email Is the Perfect Interface for Practical AI Agents
AI-powered “agents” hold big promise—but many still struggle to work independently, avoid hallucinations, and integrate into day-to-day workflows. Enter Mixus, a startup betting that email—the tool most professionals use daily—could be the breakout interface for consumer-friendly AI agents.
AI Agents With a Human Touch
Industry leaders like Andrej Karpathy and Ali Ghodsi have emphasized the importance of humans remaining in control of AI systems. Mixus follows that philosophy by letting users interact with agents directly from email or Slack. “We’re meeting customers where they are today… For the most part, they’re on email,” says co-founder Elliot Katz.—a design decision aimed at democratizing access to AI agents.
No-Code Creation Via Email
Mixus launched its beta from Stanford in late 2024 and has already raised $2.6 million in pre-seed funding, with early clients across retail, finance, and tech. Their platform enables users to define AI agents using chat prompts—or simply by emailing instructions to. The agents can then perform tasks in the inbox, such as generating reports, drafting emails for review, and even scheduling tasks—every action managed via familiar email threads.
A Demo Worth Emulating
In a live demo, agents were employed to:
- Pull open Jira tasks flagged “overdue.”
- Summarize these tasks and draft reminder emails.
- Pause for human approval before sending.
- Automatically send once approved—either immediately or on a schedule.
This example highlights Mixus’ flexibility: organizations can set checkpoints (especially on external email) and easily loop in another team member via CC or chat tag.
Shared Memory And Team Collaboration
Beyond task automation, Mixus offers “Spaces”—shared work environments that retain agent histories, user files, chat logs, and agent definitions. This gives teams a unified AI “memory” for collaborative workflows—a feature not yet fully supported by competitors like ChatGPT or Claude.
Agents That Operate Where You Work
Mixus can connect with tools like Gmail, Jira, and spreadsheets. It also uses Anthropic’s Claude 4 and OpenAI’s o3 models, along with real-time web access—like “Google Alerts on steroids”—to conduct live research or monitor content. In demos, agents navigated context intelligently, identifying task owners and managing reports on their own, signaling a leap toward AI bots that truly function as digital coworkers.
What It All Means
By embedding AI agents directly into email, Mixus removes barriers: no learning new apps, no coding, no context switching. For busy teams submerged in tools, the simplicity of “just email my agent” is compelling. Though still in early days, Mixus is pioneering what accessible, human-in-the-loop AI agents should look like—embedded in familiar workflows, smart enough to take initiative, and safe enough to pause for human approval.
