This startup thinks email could be the key to usable AI agents

AI firms are pushing brokers as the following Nice Office Disruptor, however specialists say they’re nonetheless not prepared for prime time. They typically battle with autonomous decision-making, can’t cooperate with different brokers, fail at confidentiality consciousness, and combine poorly into current programs. 

Trade pioneers like Andrej Karpathy and Ali Ghodsi have stated that, just like the deployment of autonomous autos, people should be within the loop to ensure that brokers to succeed. 

Startup Mixus understands that and has constructed an AI agent platform that not solely retains people within the workflow, but in addition permits these people to work together with brokers instantly from their electronic mail or Slack. 

“We’re assembly clients the place they’re in the present day,” Elliot Katz, Mixus co-founder, informed Trendster. “The place is each particular person within the workforce in the present day? For essentially the most half, they’re on electronic mail. And so as a result of we are able to do that via electronic mail, we imagine that’s a approach we are able to democratize entry [to agents].”

Mixus solely beta-launched out of Stanford in late 2024, nevertheless it has already raised $2.3 million in pre-seed funding and introduced on some clients, together with clothes retailer chain Rainbow Outlets, and others throughout finance and tech. 

Ease of use is Mixus’s largest promoting level, from the way it helps create brokers to how customers can work together with them. Customers can create an agent or a number of brokers from easy textual content prompts. For somebody in gross sales, that immediate may appear like: 

“Create an agent that finds all open duties in Jira in mission mixus-dummy, and ship me a report with data on all duties which are overdue. Draft emails to all of the assignees who’ve overdue duties, and have me evaluation them within the chat and with easy clear formatting for electronic mail (no attachments/docs). As soon as I confirm, ship the emails. Run it now. And transferring ahead, run it each Monday at 7am PST.”

If Mixus works reliably, this can be a large unlock for the AI agent area. Most agentic AI instruments in the present day both offer you a pre-built assistant, a la ChatGPT or Gemini, or require builders to construct customized brokers utilizing frameworks like LangChain, AutoGen, or crewAI. 

With Mixus, customers can arrange their brokers inside Mixus’s platform by way of a chat operate – via written or vocal prompts – or by merely emailing directions to agent@mixus.com. Then Mixus will construct, run, and handle single- or multi-step brokers instantly from the inbox. 

“Many of the world, most of America, doesn’t even know what an AI agent is or why it’s useful for them, and so they’ve undoubtedly by no means used one,” Katz stated, noting that older employees might need an particularly laborious time studying the best way to use brokers. “We’re making an attempt to succeed in all these those who have by no means used [agents], however might very a lot profit from an AI.”

Katz and his co-founder Shai Magzimof demoed the expertise for me, displaying how straightforward it’s so as to add human verifiers on your brokers by merely instructing at which step they need to come to you for oversight. For instance, they ran an agent to do analysis on Trendster reporters earlier than pitching them. The agent would determine and collect the most recent expertise information and developments, analyze the data to determine potential story angles for a Trendster reporter, and compile a analysis report summarizing the findings. On the final stage, the agent was directed to ship the data to Katz for verification. As soon as accepted, the agent would ship the finished analysis report back to his Magzimof. 

The founders harassed that people could be within the loop as a lot or as little as a enterprise or enterprise dictates – Magzimof stated organizations can arrange company-wide guidelines, like guaranteeing an electronic mail will get checked by a human if it’s being despatched externally. Mixus doesn’t at all times require human oversight. So, for instance, if an agent has already run a Jira integration a whole bunch of instances and hasn’t messed it up but, a human could belief it to proceed that activity autonomously. 

Or as Katz put it: “We allow colleague oversight. We don’t mandate colleague oversight.”

Bringing different colleagues into the workflow is as straightforward as tagging them within the chat with an agent and even copying them on the e-mail to the agent. That’s one other standout in comparison with brokers on the markets in the present day. Most fashions are single-user, and whereas Notion AI and Slack GPT enable customers to collaborate in shared areas, they don’t take it that step additional of letting the AI handle conversations and duties between teammates in actual time. 

One other core function of Mixus is its capacity to retailer reminiscence. 

“We created Areas so that each workforce, each particular person, each group of individuals can have a shared reminiscence,” Magzimof stated. “Then all my brokers, all my information, all of the folks could be in that very particular Area’s reminiscence.”

Whereas ChatGPT and Claude each help reminiscence, their enterprise plans don’t but help shared agent reminiscence throughout customers.  

What else can Mixus do?

The founders ran me via roughly an hour-long demo displaying a spread of use instances and talents. Its brokers do appear miraculous, reflecting a excessive diploma of autonomy and reminiscence that put Mixus on the excessive finish of the AI agent spectrum. That’s, if the product works as reliably because it did within the demo.

Like different brokers, Mixus can combine with different instruments, from Gmail to Jira, and customers can set off brokers to run instantly or on a schedule. 

Brokers in Mixus can run and edit paperwork or spreadsheets inline, which has similarities to ChatGPT, Microsoft Copilot, and Google Gemini, however these are sometimes restricted to sandboxed environments. Mixus additionally allows brokers to autonomously navigate organizational context – like determining who in a corporation owns a specific activity by wanting via Jira tickets. That sort of cross-tool, org-aware reasoning continues to be uncommon amongst in the present day’s agent platforms. 

Constructed on a mixture of Anthropic’s Claude 4 and OpenAI’s o3, Mixus brokers even have entry to the net, which Magzimof says can be utilized for duties like stay analysis or monitoring. He described it as “Google Alerts on steroids.” 

Taken collectively, Mixus seems to be much less of a productiveness device and extra like a tireless digital colleague – one of the crucial bold makes an attempt but to reimagine AI as a real collaborator. If it really works as marketed, your subsequent “coworker” may not be human, nevertheless it may get via your inbox quicker than you do.

Received a delicate tip or confidential paperwork? We’re reporting on the internal workings of the AI trade — from the businesses shaping its future to the folks impacted by their choices. Attain out to Rebecca Bellan at rebecca.bellan@techcrunch.com and Maxwell Zeff at maxwell.zeff@techcrunch.com. For safe communication, you possibly can contact us by way of Sign at @rebeccabellan.491 and @mzeff.88.