We think most “AI agents” don't survive their first week on the job.
We're Jimmy and Kyle. We've spent more than a decade building autonomous systems that defend millions of endpoints, and we built onebee to do for AI agents what managed services did for security: run them for you, so you get the outcome without the babysitting.
How we got here
We met at MIT Lincoln Laboratory, sharing an office in the Cyber Systems Assessments group. The first thing we built together was for the DARPA Cyber Grand Challenge: software that could autonomously find and patch vulnerabilities in other software. The year was 2014. Autonomous systems finding and fixing real bugs, years before anyone was talking about LLMs.
We also collaborated on a hugely successful cyber testbed program called LARIAT, which simulated the internet end to end: root DNS, a full BGP backbone, and thousands of virtual users, each with their own browsing habits. (We had a soft spot for the ones who couldn't resist a phishing email.) On top of it, we helped build training and assessment programs for U.S. Cyber Command, putting both offensive and defensive teams through their paces.
When our Lincoln group leader and a colonel we'd worked with founded a new company called SimSpace, building cyber ranges for the Fortune 50, we followed. Those were the late-night years: a coworking space in downtown Boston, pushing six-figure server racks down the sidewalk between buildings, propping the windows open in the dead of a New England winter so the machines, running thousands of VMs and live automated attacks, wouldn't cook themselves. We trained some of the largest banks in the world, and a lot of very talented people.
Jimmy went to Carbon Black next and did something neither of us expected: he joined the marketing team. A great technical product that nobody can sell doesn't matter. A year later he went back to his threat-research roots, and then Kyle joined. Together we built automated malware-analysis platforms that could understand how a piece of malware behaved and turn that into detection rules. Kyle led the building of a binary emulation toolset called binee, open-sourced it, and gave a talk on the research at DEFCON. Jimmy helped ship a new generation of detection and prevention rules for Carbon Black's endpoint sensors. Those rules are still running today, protecting millions of endpoints.
Jimmy moved to Red Canary to help transform it from an endpoint-only MDR into one that also covered cloud and identity, and to pioneer the use of AI agents inside the Security Operations Center. Kyle joined a year later to scale the detection engine. Together we reworked how human-in-the-loop security systems were built, adopting new architectures that let Red Canary scale to thousands of customers while holding some of the highest accuracy in the market.
Kyle then became a founding engineer at 7AI, leading the system architecture for an AI-native SOC and proving the model could work. It absolutely could.
Then onebee.
Why we built onebee
Across all of it, we kept seeing the same pattern. A powerful new technology ships faster than teams can learn to operate it, so a market forms to run it for them, as a service. The customers never wanted another dashboard. They wanted the outcome, and someone accountable for it.
When LLMs arrived, we saw it begin again, except this time the technology could do real work, not just answer questions. We bet on AI agents early. And we watched the same story start over: standing up an agent that demos well is easy, keeping it alive is not. Agents drift, accuracy slips as the business changes, models get deprecated, costs creep, and capacity has to hold under real load. We watched capable teams burn quarters on automation that quietly grew a headcount of people keeping it alive.
Between us, the agents we've built have investigated tens of millions of security events, correlating across petabytes of data, grounded in years of work in machine learning and data science. We learned exactly what it takes to make an agent trustworthy enough to run a business on.
So we built onebee to do for agents what managed services did for security. We call it Managed Agents: we deploy, tune, secure, and monitor them for you, on a platform where you connect your data and ship an agent in a few clicks. Every agent we run is governed, logged, and yours to control. We like to say: if you can write a memo, you can build an agent on the Hive.
See what we'd automate first.
The free 10-minute audit is the fastest way to understand how we think.