AI companies are pushing agents as the next major disorders in the workplace, but experts say they are still Not ready for prime time. AI agents often struggle to make decisions on their own, they are often given, they cannot work with other agents, Failed to inform confidentialityAnd it is incorporated with existing systems.
The pioneers of industry such as Andrej Karpathy And Ali Ghodsi said that, like the development of autonomous vehicles, People must be in the loop to succeed agents.
A boot called Mixer He wants to deal with it with the AI Agent platform that not only keeps people in the workflow, it also allows users to interact with agents directly from their email or slack.
“We meet the customers where they are today,” said Elliot Katz, co -founder of Mixus, in TechCrunch. “Where are each person in the workforce today? [to agents]. ”
If the mixus reliably works, it can solve a big problem in the AI agent. Most AI companies today either give you a pre-made assistant, à la chatgpt or gemini, or developers must build custom agents using frames such as langchain, autogen or crewai.
Mixus started at Beta from Stanford only at the end of 2024, but has already raised $ 2.6 million in pre-funding and brought to some customers, including clothing stores, as well as others in all funding and technology.
Starting says that the biggest point of sale is ease of use than how it helps you create agents on how to interact with them. Users can use text suggestions to configure their agents on the mixus platform through a conversation function or simply by email with instructions at agent@mixus.com. Then Mixus will build, run and manage the factors of one or multiple steps directly from the inbox.
TechCrunch event
Francisco
|
27-29 October 2025
For example, a customer support manager may use a prompt that reads it:
Create an agent who finds all the open work on Jira at Project Mixus-Dummy and send me a report with information on all delayed tasks. Email messaging design to all versions that have delayed the work and review them in the conversation and with simple clear formatting for e -mail (without attachments/documents). Once I have verified, send the emails. Run it now. And move forward, run every Monday at 7am. Pst.
Katz and co -founder Shai Magzimof created the agents for TechCrunch, showing how to add human verifications for your agents simply by teaching what step the agent should ask you.
For example, they ran an agent to do research on TechCrunch journalists before placing them. The agent recognized and gathered technological news and trends, analyzed the information to identify potential history angles and drafted a research report that summarizes the findings. In the last stage, the agent headed to send the information to Katz for verification. Once approved, the agent will send a comprehensive research report to Magzimof.
The founders have noted that people can be in the loop as much as needed-Magzimof has stated that organizations can create rules throughout the company, such as securing an email controlled by a person if sent outside the company.
Attracting other colleagues in the workflow is as easy as their labels in the conversation with an AI agent or copying them to the e -mail to the agent. This is another standout compared to agents in markets today: Most models are a user, and while the concept AI and Slack allow users to work together in common, they do not let AI manage talks and tasks among teammates in real time.
Another key feature of Mixus is his ability to remember files, conversations, prompts and agents.
“We created spaces so that every group, every person, every group of people can have a common memory,” Magzimof said. “Then all my agents, all my files, all people can be in memory of this particular space.”
While Chatgpt and Claude support memory, they support, their business plans do not yet support all users.
What else can Mixus do?
In our interview, the founders ran through an hour of demonstration showing a series of cases of use and abilities. Mixus agents seem capable of reflecting a high degree of autonomy and memory that places the company towards the highest end of the AI Agent spectrum. Namely, if The product works both reliably and in the demo.
Like other agents, the mixus can be integrated with other tools, from Gmail to Jira, and users can activate agents to run immediately or in a program. Agents can perform and process inline documents or computer sheets – similar to chatgpt, Microsoft Copilot and Google Gemini, but they are often limited to Sandboxed environments.
Mixus also allows agents to navigate the organizational framework autonomously – such as finding who in an organization holds a project by looking via Jira tickets.
Built in a combination of Claude 4 of Anthropic and Openai’s O3, Mixus agents also have access to the web, which Magzimof says it can be used for work such as live research or monitoring. Described it as “Google notifications in steroids”.
Overall, Mixus seems to be less than a productivity tool and more like a tireless digital colleague – another ambitious attempt to redefine AI as a partner. If it works as advertised, your next “partner” may not be human, but it may pass from your inbox faster than you.
Correction: This article has been informed to reflect how much mixus has increased to date.
Do you have a sensitive advice or confidential documents? We mention the internal operation of the AI industry – from companies that shape their future to the people who are influenced by their decisions. Contact Rebecca Bellan to rebecca.bellan@techcrunch.com and Maxwell Zeff to maxwell.zeff@techcrunch.com. For safe communication, you can contact us through signal at @rebeccabellan.491 and @Mzeff.88.
