AI agents are getting smarter but in enterprise operations, they’re still surprisingly ineffective. The reason isn’t model capability. It’s memory.
That’s the core idea behind Interloom’s new announcement: a $16.5M seed round led by DN Capital to build what it calls a “memory layer” for enterprise AI. But beyond the headline, the company is tackling a deeper, more structural problem—how work actually gets done inside organisations.
The Problem: AI Without Context
Despite rapid advances, AI agents struggle in real-world operations because they lack the lived experience of human experts. As Interloom’s CEO Fabian Jakobi explains, most operational knowledge isn’t documented but learned over time and applied instinctively.
“Most of the actual knowledge is not written down. It’s just being done.”
This creates a massive gap. While companies rely on SOPs and documentation, the reality is that the majority of decisions—often cited as ~70%—exist only in the heads of employees or buried across emails, tickets, and conversations.
For AI agents, that’s a dead end.
Interloom’s Approach: Capturing Work as It Happens
Interloom flips the problem. Instead of trying to predefine workflows (which rarely works), it builds a system that learns from real operations in motion.
At its core is a new kind of workspace—part ticketing system, part AI collaboration layer—where humans and agents work together. But unlike traditional tools, every action is captured, structured, and reused.
“The system… is built to actually trace and capture every single action… to make sure that we remember how we did it last time.”
Each resolved case becomes part of Interloom’s Context Graph—a continuously evolving memory of how the organisation operates. Over time, this allows AI agents to move from assisting experts to autonomously handling similar cases.
From Static Software to Adaptive Systems
A key insight from the Interloom team is that traditional software was never designed for this level of complexity. Most enterprise tools are rigid and rule-based, while real-world operations are messy, dynamic, and constantly changing.
Jakobi compares the future of enterprise AI to Google Maps:
Static maps = traditional software
Real-time traffic data = human + AI feedback loop
Navigation = adaptive decision-making
In the same way, Interloom uses real-world actions to continuously refine how work is done—bridging the gap between deterministic systems and flexible AI agents.
Why This Matters Now
Two trends are converging:
The rise of AI agents in frontline operations
The loss of institutional knowledge as experienced workers retire
Interloom positions itself at this intersection. By capturing expertise as it happens, it ensures that knowledge doesn’t disappear—and that AI systems can actually use it.
As Jakobi puts it:
“If we’ve never written it down, how will the agent do anything? … It won’t be able to.”
A New Layer in the Enterprise Stack
Interloom isn’t trying to replace large models or existing systems. Instead, it’s building a new layer on top: one that grounds AI in real operational memory.
That distinction is key. The challenge in enterprise AI isn’t just better models, it’s embedding them into the messy reality of how organisations function.
And if Interloom is right, the winners in this space won’t just be the companies with the best AI but the ones with the best memory.
