AI corporations push agents as the next great interference in the workplace, but experts say they are still not ready for the best time. AI agents often fight to make decisions themselves, often hallucinations, cannot cooperate with other agents, Failure of confidentiality awarenessAnd integrate poorly with existing systems.
The pioneers of the industry like Andrej Karpathy and Ali Ghodysi said that, like the implementation of autonomous vehicles, People must be in the loop That agents succeed.
Startup called Mixus He wants to solve this thanks to the AI agent platform, which not only keeps people in the flow of labor, but also allows users to interact with agents directly with their e -mail or slack.
“We meet with clients where they are today,” said Techcrunch Elliot Katz, co-founder of Mixus. “Where is every person in the working force today? He is mostly on E -Mail. And because we can do it via e -mail, we think that in this way we can democratize access [to agents]. “
If the mixus works reliably, this can solve a big problem in the AI agent space. Most AI corporations either give a pre -built assistant, à la chatgpt or gemini, or programmers must build non -standard agents using RAMs akin to Langchain, AutoGen or Cretai.
Mixus introduced Beta outside Stanford only at the end of 2024, but has already collected $ 2.6 million financing before the seeded and brought some customers, including a chain of Rainbow Shops clothes shops, in addition to others in finance and technology.
Startup claims that its largest sales point is the ease of use, from how he helps to create agents to interact with them. Users can use text prompts to configure their agents on the MIXUS platform via the chat function or simply E -Mail of instructions to [email protected]. Then Mixus will build, launch and manages single or multi -stage agents directly from the inbox.
TechCrunch event
San Francisco
|.
October 27-29 2025
For example, a customer support manager can use a poem that reads:
Create an agent who finds all open tasks in JIRA in Project Mixus Mumnequin, and send me a report with information about all late tasks. Develop E -Mail for all assigneles who have late tasks, and ask me to look at them in the chat and with easy vivid e -mail formatting (without attachments/documents). After verifying, send E -Maile. Start now. And go forward, run it every Monday at 7 am PST.
Katz and his co -founder Shai Magzimof demonstrated TechCrunch agents, showing how to add human verifiers for their agents, simply instructing at which stage the agent should ask you for supervision.
For example, they conducted an agent for research on TechCrunch reporters before they were broken up. The agent identified and collected technological trends and trends, analyzed information to discover the potential pages of the story and developed a research report summarizing the arrangements. At the last stage, the agent was directed to send information to Katz for verification. After approval, the agent will send a accomplished test report to Magzimof.
The founders noticed that people may be in the loop as many or as little as required-Magzimof said that organizations can establish the principles of the whole company, akin to ensuring that the e-mail will be checked by man if it is sent outside the company.
The introduction of other colleagues into the flow of labor is so simple as the designation of them in chat with the AI agent or copying them on E -Mail to the agent. This is one other unique compared to agents in the markets: most models are one -way and although the concept of artificial intelligence and slack allow users to cooperate in common spaces, they do not allow artificial intelligence to manage conversations and tasks between members of the team in real time.
Another basic function of MIXUS is the ability to remember files, chats, hints and agents.
“We created spaces so that every team, every person, every group of people could have a common memory,” said Magzimof. “Then all my agents, all my files, all people can be remembered by this very specific space.”
While Chatgpt and Claude support memory, their company plans do not yet support the agent’s memory among users.
What else can Mixus do?
In our interview, the founders went through an hour demonstration version showing a variety of use and skills. Mixus agents seem capable, reflecting the high degree of autonomy and memory, which he places the company in the direction of a higher end of the AI agent spectrum. This means that the product works as reliably as in a demo.
Like other agents, Mixus can integrate with other tools, from Gmail to Jira, and users can free agents to act immediately or according to the schedule. Agents can run and edit documents or spreadsheets in the line – like Chatgpt, Microsoft Copilot and Google Gemini, but they are often limited to sandblasted environments.
Mixus also allows agents to autonomously navigate in an organizational context – for example, determining who is the owner of the task in the organization, browsing JIRA tickets.
Built on the combination of Anthropika Claude 4 and O3 OpenAI, Mixus agents also have access to the Internet, which according to Magzimof can be used for tasks akin to live or monitoring research. He described it as “Google alerts on steroids.”
To sum up, Mixus seems to be a smaller productivity tool, and a more tireless digital colleague – one other ambitious attempt to re -expression AI as a collaborator. If he works as advertised, one other “colleague” may not be human, but he can go through your inbox faster than you.