AI companies are pressing agents as the next Great Workplace Disruptor, however experts say theyre still not prepared for prime-time show.
AI agents often have a hard time to make choices on their own, hallucinate regularly, cant comply with other agents, stop working at privacy awareness, and incorporate inadequately with existing systems.Industry pioneers like Andrej Karpathy and Ali Ghodsi have actually said that, like the release of autonomous vehicles, humans require to be in the loop in order for agents to succeed.A start-up called Mixus wishes to deal with that with its AI agent platform that not only keeps humans in the workflow, it likewise lets users interact with agents straight from their e-mail or Slack.Were conference customers where they are today, Elliot Katz, Mixus co-founder, informed A Technology NewsRoom.
Where is every person in the labor force today? For the many part, theyre on e-mail.
Therefore because we can do this through e-mail, our company believe thats a method we can democratize gain access to [to representatives] If Mixus works reliably, it might solve a big issue in the AI representative space.
Many AI business today either offer you a pre-built assistant, la ChatGPT or Gemini, or developers have to construct custom-made agents utilizing structures like LangChain, AutoGen, or crewAI.Mixus released in beta out of Stanford just in late 2024, however it has actually already raised $2.6 million in pre-seed financing and induced some consumers, including clothing shop chain Rainbow Shops, along with others across finance and tech.The startup states its biggest selling point is ease of usage, from how it helps you produce representatives to how you can connect with them.
Users can use text triggers to set up their representatives within Mixuss platform via a chat function, or by merely emailing instructions to [email protected].
October 27-29, 2025For example, a sales representative might utilize a prompt that checks out so: Create a representative that finds all open tasks in Jira in task mixus-dummy, and send me a report with details on all tasks that are overdue.
Draft e-mails to all the assignees who have overdue jobs, and have me evaluate them in the chat and with easy clear formatting for email (no attachments/docs).
Once I validate, send the e-mails.
Run it now.
And moving forward, run it every Monday at 7am PST.Image Credits: MixusKatz and his co-founder Shai Magzimof demoed the agents for A Technology NewsRoom, showing how to add human verifiers for your agents by simply advising at which action the representative need to ask you for oversight.
They ran an agent to do research on A Technology NewsRoom press reporters before pitching them.
The agent identified and gathered innovation news and trends, examined the details to determine possible story angles, and put together a research study report summing up the findings.
At the last stage, the representative was directed to send the info to Katz for verification.
As soon as authorized, the representative would send out the finished research report to Magzimof.The founders kept in mind that people can be in the loop as much or as low as needed Magzimof stated organizations can set up company-wide guidelines, like ensuring an email gets examined by a human if its being sent outside the business.
Bringing other colleagues into the workflow is as simple as tagging them in the chat with an AI representative, or copying them on the e-mail to the representative.
Thats another standout compared to agents on the marketplaces today: Most models are single-user, and while Notion AI and Slack enable users to work together in shared spaces, they dont let the AI handle conversations and tasks in between colleagues in real time.Another core function of Mixus is its ability to remember files, chats, triggers and agents.We developed Spaces so that every group, every person, every group of individuals can have a shared memory, Magzimof said.
All my agents, all my files, all the people can be in that very particular Spaces memory.While ChatGPT and Claude both support memory, their enterprise plans dont yet support shared representative memory across users.In our interview, the creators ran through an hour-long demonstration showing a variety of use cases and abilities.
Mixus agents do appear capable, reflecting a high degree of autonomy and memory that positions the business towards the higher end of the AI agent spectrum.
That is, if the item works as dependably as it performed in the demo.Like other agents, Mixus can incorporate with other tools, from Gmail to Jira, and users can set off representatives to run immediately or on a schedule.Agents can run and modify files or spreadsheets inline comparable to ChatGPT, Microsoft Copilot, and Google Gemini, but those are typically restricted to sandboxed environments.
Mixus also lets representatives autonomously browse organizational context like finding out who in an organization owns a task by checking out Jira tickets.
Constructed on a combination of Anthropics Claude 4 and OpenAIs o3, Mixus representatives also have access to the web, which Magzimof says can be tapped for tasks like live research or tracking.
He explained it as Google Alerts on steroids.Taken together, Mixus seems less of a performance tool and more like a determined digital associate yet another ambitious effort to reimagine AI as a collaborator.
If it works as advertised, your next colleague may not be human, but it might survive your inbox faster than you do.Correction: This post was updated to reflect just how much Mixus has actually raised to date.
Got a sensitive suggestion or personal documents? Were reporting on the inner workings of the AI market from the business forming its future to the people impacted by their decisions.
Reach out to Rebecca Bellan [email protected] Maxwell Zeff [email protected].
For safe interaction, you can contact us via Signal [email protected] [email protected].