AI companies are pushing agents as the next great work disintegration, but experts say they are still there Not ready for prime timeAI agents often struggle to make decisions by themselves, often halts, cannot cooperate with other agents, Failed in privacy awarenessAnd integrate poorly with existing systems.
Industry like a forearm Andrej karpathi And Ali Ghodasi has said that like the deployment of autonomous vehicles, Humans should be in loop To succeed agents.
Is called a startup Mixus With its AI agent platform, it wants to address that not only keeps humans in workflow, it allows users to interact with agents directly from their email or slack.
“We are meeting customers where they are today,” Elliot Katz, the co-founder of Mixus, told Techcrunch. “Where is every person in the workforce today? For the most part, they are on emails. And because we can do it via email, we believe that this is a way that we can democratization of access (for agents).”
If the mixus works firmly, this AI agent can solve a major problem in space. Most AI companies today either give you a pre-made assistant, a la chat or Gemini, or developers have to manufacture custom agents using framework such as Langchen, Autogen, or Kuwai.
Mixus launched only from Stanford to beta at the end of 2024, but it has already raised $ 2.6 million in pre-seeded funding and brought some customers, including the clothing shop chain Rainbow Shops, as well as others of finance and technology.
Startup says its biggest sales point is ease of use, how it helps you make agents how you can interact with them. Users can use the text prompt by emailing their agents within the platform of the mixus through a chat function, or by emailing the instructions at only agent@mixus.com. Then the Mixus inbox will make, run, run and manage single-or multi-step agents directly.
Techcrunch event
San francisco
,
27-29 October, 2025
For example, a seller can use a sign that reads:
Create an agent that finds all open tasks in cumin in project mixes-dimmy, and send me a report with information about all the tasks that are overdose. Draft emails for all assignments that have overdue tasks, and review me in chat and with simple clear formatting (no attachment/docs) for emails. Once I verify, send emails. Now run it. And moving forward, run it on PST at 7 am every Monday.
Katz and his co-founder Shai Magazimof demoned agents for techcrunch, showing how to add a human verifier for their agents just to give instructions to add a simply that the agent should ask you for the overseas.
For example, he ran an agent to research Techcrunch reporters before pitching them. The agent identified and collected technology news and trends, analyzed information to identify potential story angles, and briefly compiles a research report to the findings. In the final stage, the agent was directed to send information to Katj for verification. Once approved, the agent will send a complete research report to Magazimof.
The founders mentioned that humans may be in the loop or as much as the need, the less Magzimof said that the company can establish company-wide rules, such as an email is checked by a human if it is being sent outside the company.
Bringing other colleagues into workflow is as easy as tagging them in chat with AI agent, or copying them to the agent on email. Today it is another standout compared to agents in markets: most models are single-user, and while perception allows AI and slack users to cooperate in shared places, they do not allow AI to interact and manage tasks between teammates in real time.
Another main feature of Mixus is files, chat, signals and ability to remember agents.
“We created spaces so that every team, every person, every group of every person could have a shared memory,” Magazimof said. “Then all my agents, all my files, all people can be in the memory of that specific space.”
While both Chatgpt and Cloud support memory, their enterprise plans do not yet support the shared agent memory in users.
What else can mixes do?
In our interview, founders showed a series of cases and abilities used through one hour of demo. Mixus agents seem capable showing autonomy and a high degree of memory, which leads the company to the high end of the AI agent spectrum. he is, If The product works firmly as was done in the demo.
Like other agents, mixes can integrate from Gmail to cumin with other tools, and user agents can trigger agents to run immediately or on schedule. Agents can run and edit the document or spreadsheet inline – chat, microsoft copilot and similar to Google Gemini, but they are often limited to an environment with sandbox.
The mixus agents autonomally navigate the organizational context – such as finding out who in an organization is the owner of a task by looking through cumin tickets.
Mixed agents also have access to the web, built on a combination of cloud 4 of anthropic and O3 of OpenAI, Magazimof says that can be tapped for tasks such as live research or monitoring. He described it as “Google Alert on Steroid”.
Together, Mixus appears to be less than a productivity tool and a tireless digital partner – yet another ambitious attempt to re -prepare AI as a colleague. If it acts as advertised, your next “colleague” may not be human, but it can be done rapidly by you through your inbox.
Correction: It was updated to show this article how much Mixus has raised till date.
Found a sensitive tip or confidential documents? We are reporting on the internal functioning of the AI industry – from companies to people affected by companies that shape their future. Reach Rebecca Belan rebecca.bellan@techcrunch.com And on Maxwell Zaf maxwell.zeff@techcrunch.comFor safe communication, you can contact us via signal @REBECCABELLAN.491 and @mzeff.88.
