This startup thinks email could be the key to usable AI agents

In the rapidly evolving landscape of AI, companies are heralding agents as the upcoming transformative force in workplace efficiency. However, experts caution that these agents still have significant hurdles to overcome. Many AI systems struggle with decision-making, frequently produce inaccurate information, lack the ability to collaborate with other agents, and often fail to maintain confidentiality. Notable figures in the industry, such as Andrej Karpathy and Ali Ghodsi, emphasize the necessity of human oversight, drawing parallels to the deployment challenges faced by autonomous vehicles. Enter Mixus, a startup aiming to bridge the gap by introducing an AI agent platform that integrates directly with users' existing workflows, specifically through email and Slack. Co-founder Elliot Katz shared with TechCrunch, "We’re meeting customers where they are today. For the most part, they’re on email. By leveraging this platform, we believe we can democratize access to AI agents." Launched in beta from Stanford in late 2024, Mixus has already made waves, securing $2.3 million in pre-seed funding and attracting clients such as the retail chain Rainbow Shops, along with several finance and tech organizations. The startup's standout feature is its user-friendly approach, allowing individuals to create and manage AI agents through simple text prompts or by sending instructions via email to agent@mixus.com. For instance, a salesperson could instruct the agent to track overdue tasks in a project management tool and draft emails for follow-up. The flexibility doesn’t end there; users can also schedule recurring tasks with ease. Katz and his co-founder Shai Magzimof recently showcased the platform's capabilities, highlighting how users can dictate when an agent should seek human confirmation on specific actions. During a demo, they illustrated an agent conducting research on TechCrunch reporters, compiling findings, and seeking approval before distributing the information. This feature allows businesses to tailor oversight based on their needs, creating a balance where human verification is integrated into the workflow as required. Mixus also excels in facilitating collaboration within teams. Unlike many current AI solutions that operate in isolation, Mixus enables users to tag colleagues in chats or include them in emails concerning agent activities. Additionally, it boasts a unique memory function, allowing every team member to access shared information seamlessly. While competitors like ChatGPT and Claude are beginning to explore shared memory across users, Mixus claims to have taken this concept further. The platform can remember files, conversations, and agent instructions, creating a comprehensive shared memory space. The demo showcased a variety of use cases, with Mixus agents demonstrating a level of autonomy and memory that positions them at the forefront of the AI agent field. Beyond basic productivity tools, Mixus aims to be a relentless digital collaborator. With integration capabilities across various tools, including Gmail and Jira, it allows for immediate or scheduled agent actions, editing documents inline, and even navigating organizational contexts autonomously. Underpinning its technology is a combination of Anthropic’s Claude 4 and OpenAI’s o3, enabling Mixus agents to perform tasks such as live research and monitoring akin to an upgraded version of Google Alerts. If Mixus lives up to its potential, it could redefine the concept of workplace collaboration by introducing a digital colleague that might just outpace human efficiency in managing tasks.