This young man believes that e -mail can be the key to useful intelligence customer

Artificial intelligence companies pay the agents as great compensation in the workplace, but experts say they are still Not ready for peak time. Artificial intelligence agents often struggle to make decisions themselves, and hallucinations can not often cooperate with other factors, FailureAnd it is badly combined with the current systems.

Industry pioneers like Andrej Karpathy Ali Godsi, such as spreading independent vehicles, said, Humans must be in the episode In order for the agents to succeed.

An emerging Mixus He wants to process this through the AI’s agent platform that does not only keep humans in the workflow, but also allows users to interact with agents directly from their email or recession.

“We meet where they are today,” Elliot Katz, co -founder of Mixus, told Techcrunch. “Where is every person in the workforce today? For the largest part, they are on the email. Therefore, because we can do this through email, we believe that this is a way we can cancel access to agents (for agents).”

If the Mixus is working reliably, it may solve a major problem in the area of the artificial intelligence agent. Most artificial intelligence companies today give you either pre -assistant, à la chaatgpt or Gemini, or developers to build custom factors using frameworks such as Langchain, autogen or Crewai.

Mixus was launched at Beta Out of Stanford only in late 2024, but already raised $ 2.6 million of financing before the seed and brought some customers, including Rainbow stores in the clothing chain, as well as others through financing and technology.

The starting company says its largest sale is the ease of use, from how to help you create agents how to interact with them. Users can use text claims to prepare their agents within the Mixus platform via the chat function, or once the instructions are sent via email to Agent@mixus.com. Then the Mixus will build, operate and manage one or multi -step factors directly from the inbox.

For example, a sales deputy may use a claim that reads this:

Create an agent who finds all open tasks in Jira in Project Mixus-Dummy, and send me a report on all the late tasks. Formulating emails for all the appoinals who are late in the tasks, and made me review them in chat and with clear clear format for email (no attachments/documents). Once you check, send emails. Run it now. Go ahead, run it every Monday at 7 am Pacific time.

Katz and his founder, Maggifov, collected agents for Techcrunch, explaining how to add human verification to your agents simply by the step that the agent should be asked to supervise.

For example, they operated an agent to conduct a search on Techcrunch correspondents before they were put. Identify the agent, collect news and technological trends, analyze information to determine the possible story corners, and collect a research report summarizing the results. In the last stage, the agent was directed to send information to Katz to verify. Once approved, the agent will send the complete research report to Magzimof.

The founders note that humans can be in the episode as much as it is required or less-Magazifov can that institutions can prepare bases at the company level, such as ensuring an email examination by the human being if it is sent outside the company.

Bringing other colleagues to the workflow is easy as putting signs in chatting with artificial intelligence agent, or copying them on the email to the agent. This is another prominent compared to the agents on the market today: Most models are one user, and although the idea AI and Slack allow users to cooperate in common spaces, it does not allow AI to manage conversations and tasks among teammates in the actual time.

Another basic feature of Mixus is its ability to remember files, conversations, demands and agents.

“We have created spaces so that every team, every person, every group of people can get a common memory,” said Magizov. “Then all my agents, and all my files, can all be in the memory of this very specific space.”

Although ChatGPT and CLADE support for memory, their institutions plans are not yet supported by a common agent’s memory across users.

What can Mixus do?

In our interview, the founders have fought an hour offer that displays a set of cases of use and abilities. Mixus factors seem to be able to fate, reflecting a high degree of independence and memory that puts the company towards the higher end of the artificial intelligence agent. that it, if The product works reliably as it was in the demonstration.

Like other agents, Mixus can integrate with other tools, from Gmail to Jira, and users can run the agents immediately or on a schedule. Customers can operate and edit documents or embedded data schedules – similar to Chatgpt, Microsoft Copilot and Google Gemini, but often limited to sand environments.

MIXUS also allows agents independently to move in the organizational context – such as knowing who has in the institution important by looking at Jira tickets.

Built on a mixture of Claude 4 and Openai’s O3, Mixus Agents can access to the web, which Magzimof says can be used for tasks such as live research or monitoring. He described it as “Google Alerts on Doping”.

Combined, the Mixus seems less than a production tool and more like an indisputable digital colleague – another ambitious attempt to re -imagine artificial intelligence as a collaborator. If he works as announced, the next “workshop” may not be human, but it may pass through your in your inbox faster than you do.

Correction: This article has been updated to reflect the amount of Mixus raised so far.

Do you have sensitive advice or secret documents? We report the internal business of the artificial intelligence industry – from companies whose future is to people affected by their decisions. Access to Rebecca Billan in ribecca.bellan@techcrunch.com Maxwell is false in maxwell.zeff@techcrunch.com. For safe contact, you can contact us by reference to RebeCabellan491 and MZEFF88.