AI companies are pushing agents as a barrier to the next great workplace, but experts say they are still Not ready for prime time. AI agents often struggle to make decisions, often deceive, cannot cooperate with other agents, Failed to be aware of privacyAnd insert poorly with existing systems.
The industry’s pioneers like Andridge Carpathi And Ali Goodsi has said, like the deployment of autonomous vehicles, Humans need to be in the loop To succeed agents.
Called a startup Mix It wants to solve this with its AI agent platform that not only keeps humans in the workflow, but also allows users to interact with their email or slowly with agents.
“We are meeting consumers where they are today,” said Elite Kitz, co -founder of the mix. “Where is every person in the workforce today? For most parts for h, they are on email. And because we can do this by email, we believe that this is a way we can make access (to agents) democratic.”
If the mix works reliably, it can solve a major problem in the AI agent’s place. Most AI companies today either give you a pre -created assistant, Ã lachet or gym, or developers have to make a custom agent using a framework such as Langchen, Autojan, or Croatia.
The mix just launched a beta out of Stanford at the end of 2024, but has collected 6 2.6 million in pre -seed financing and brought some consumers, including the clothing store China Rainbow Shops as well as finance and other tech members.
Startup says its biggest sales point is ease of use, how helps you make an agent how you can communicate with them. Users can use text prompts to configure their agents in the Mixes platform via chat function, or just by emailing instructions on the agent@Mixus.com. Then the mix will operate, operate and manage directly from the inbox directly-or multi-phase agents.
Taxkarnch event
San Francisco
|
October 27-29, 2025
For example, a customer support manager can use a prompt that reads:
Create an agent who finds all the open tasks in the project mix dummy, and send me a report that contains information about all the tasks that are owed. Prepare emails to all the assistants who have the task, and review me in the chat and easy clear formatting for email (no attachment/document). Once I confirm, send emails. Now run it. And moving, run on the PST every Monday at 7am.

Kitz and its co -founder, Shay Magazmov, demo the agents for the taxchchic, showing how to include human certificates for your agents, which only instructed the agent to ask which step you should ask.
For example, he run an agent to do research before researching the Tech Crunch reporters. The agent identified and collected technology news and trends, analyzed information to identify possible angles of the story, and compiled a research report, summarizing these results. In the last step, the agent was instructed to send Katz to Katz for confirmation. Once approved, the agent will send a thorough investigation report to the magazimov.
The founders noted that a human being can live in a maximum or less loop-Mutmov said that organizations can set the company’s wider rules, as an email is checked by a human if it has been sent out of the company.
It is as easy as bringing other colleagues into a workflow, tagging them in a chat with an AI agent, or copying the agent on an email. This is another stand out compared to agents in today’s markets: Most models are single users, and when the idea allows AI and Slack users to cooperate in shared places, they do not allow AI to manage dialogue and tasks among team colleagues in real time.
Another fundamental feature of the mix is the ability to remember files, chats, indicators and agents.
The magazmov said, “We created places so that every team, every person, every group of every person can have a joint memory.” “Then all my agents, all my files, all people can be in memory of this particular place.”
Although Chat GPT and Claude support both memory, their enterprise projects do not yet support joint agents memory in consumers.
What else can mix?
In our interview, the founders fled through an hour -long demo, showing a range of different use issues and abilities. Mix agents look worthy, reflecting advanced sovereignty and memory that places the company towards the high end of the AI agent spectrum. , Namely, If The product does as much reliable as it did in the demo.
Like other agents, mix can be connected, from Gmail to Jira, with other tools, and user agents can mobilize to run immediately or on schedule. Agent documents or spreadsheets can run and edit them – like Chat GPT, Microsoft Colat, and Google Gemini, but they are often limited to the sandboxed environment.
Maxis allows agents to go to the independent organizational context – as to find out that looking at Jira tickets in an organization.
Made on the combination of Anthropic’s Claude 4 and Openi O3, mix agents also have access to the web, which Mizmov says can be tapped for direct research or monitoring. He described it as “Google Alerts on Steroids”.
Together, the mix looks like a low and more tireless digital partner in a productive tool. If it works according to the ad, your next “fellow worker” may not be a human, but it may be found faster than you in your inbox.
Correction: This article was updated to reflect on how much Maxis has increased to date.
Are a sensitive tip or secret document? We are reporting about the internal works of the AI industry – from companies to those affected by their decisions, from companies to those who form their future. Rebean to the Rebecca Ballon Rebecca.bilan@TechCrunch.com And Maxwell Zef Maxwell.zev@TechCrunch.com. Safe Communication L you, you can contact us via signal @Rebekabelin.491 and @Mizf.88.