AI companies are pushing agents as the next great workplace disruptor, but experts say they are not ready for prime time yet. AI agents struggle to make their own decisions, frequently hallucinate, fail to cooperate with other agents, fail to recognize confidentiality, and inadequate integration with existing systems.
Industry pioneers like Andrej Karpathy and Ali Ghodsi say that, like the deployment of self-driving cars, humans need to be in the loop for agents to succeed.
A startup called Mixus wants to deal with it with an AI agent platform that not only keeps people in the workflow but also allows users to interact with agents directly from email and Slack.
“We’re meeting our customers today,” Elliot Katz, co-founder of Mixus, told TechCrunch. “Where are all people in today’s workforce? Most of the time, they’re emailing. So we can do this by email, so we believe it’s a way to democratize access. [to agents]. ”
If Mixus works reliably, it could solve a major problem with the AI agent space. Most AI companies today need to provide either Chatgpt or Gemini, a pre-built assistant, or developers need to build custom agents using frameworks such as Langchain, Autogen, Crewai and more.
Mixus was released in beta by Stanford University in late 2024, but has already raised $2.6 million in seed funding, attracting clothing store chain Rainbow Shop and other customers in finance and technology.
Startups say their biggest selling point is ease of use, from how they help create agents to how they interact with agents. Users can use a text prompt to configure an agent within Mixus’ platform via chat functions or email instructions to agent@mixus.com. Mixus then builds, runs, and manages single or multi-step agents directly from your inbox.
TechCrunch Events
San Francisco
|
October 27th-29th, 2025
For example, the Customer Support Manager can use the prompt to read as follows:
Create an agent that found all open tasks in Jira in Project Mixus-Dummy and send a report with information about all expired tasks. Draft emails to all assigns who have late tasks, review them in chat, and let them review them in a simple format (no attachments/documents) in email. Once confirmed, we will send you an email. Run now. Then I move forward and run it every Monday at 7am PST.
Katz and his co-founder Shai Magzimof demonstrated how to add human validation agents to agents by demonstrating TechCrunch’s agents and simply instructing them to ask them to monitor.
For example, they ran agents to conduct research on TechCrunch reporters before pitching. Agents identified and collected technology news and trends, analyzed information to identify potential story angles, and compiled a research report summarizing the findings. At the final stage, the agent was instructed to send the information to Katz for verification. Once approved, the agent will send the completed investigation report to Magzimof.
The founder pointed out that humans can be in as many loops as they need or as few loops. Magsimov said organizations can set companywide rules to see if the organization is being sent outside the company.
Bringing other colleagues into your workflow is as easy as tagging them in chat with an AI agent or copying them to your agent via email. This is another standout compared to agents in the market today. Most models are single-user, and conceptual AI and slack allow users to collaborate in shared spaces, but AI does not allow them to manage conversations and tasks between teammates in real time.
Another core feature of Mixus is that you can remember files, chats, prompts and agents.
“We created a space so that every team, every person, every group of groups can have a shared memory,” Magimov said. “Then all my agents, all my files, all my people can be in that very specific space memory.”
Both ChatGpt and Claude support memory, but Enterprise Plans do not yet support shared agent memory across users.
What else can Mixus do?
In our interview, the founders ran through an hour-long demonstration showing various use cases and capabilities. Mixus agents appear to be capable, reflecting the high level of autonomy and memory that places the company towards the AI agent spectrum upper limit. That is, if the product works reliably like a demo.
Like other agents, Mixus can integrate with other tools, from Gmail to Jira, allowing users to trigger agents and run immediately or on a schedule. Agents can run and edit documents or spreadsheets inline, similar to ChatGpt, Microsoft Copilot, and Google Gemini, but these are often limited to sandbox environments.
Mixus can also allow agents to autonomously navigate the context of an organization. This includes examining Jira tickets to figure out who owns the tasks in your organization.
Built on the combination of Anthropic’s Claude 4 and Openai’s O3, Mixus Agent also has access to the web. This states that Magzimof can tap on tasks like live research and monitoring. He described it as “Google alerts on Steroids.”
To sum up, Mixus looks more like a tireless digital colleague than a productivity tool. This is another ambitious attempt to rethink AI as a collaborator. If it works as advertised, the next “co-worker” may not be human, but it could be faster than passing through your inbox.
Fix: This article has been updated to reflect the amount Mixus has raised so far.
Do you have sensitive tips or confidential documents? We report on the internal mechanisms of the AI industry. From companies shaping their futures to those affected by their decisions. Contact Rebecca Bellan and Maxwell Zeff at maxwell.zeff@techcrunch.com. For secure communication, please contact us via the signals @rebeccabellan.491 and @mzeff.88.
