We've spent the better part of a decade backing companies that build the foundational layers other software depends on, data platforms, workflow orchestration systems, integration infrastructure. The pattern is consistent: when a new computing paradigm emerges, the first wave is tooling. The second wave is the infrastructure that makes those tools scalable.
We're watching that pattern play out again in enterprise AI.

The first wave of generative AI tooling has arrived, and it's working.

Copilots accelerate individual tasks. Chatbots surface information instantly. Code assistants compress development cycles. These tools are delivering real productivity gains and becoming essential parts of how knowledge work gets done.
But here's what we're hearing from the enterprises deploying them: "We have lots of different AI tools across the organization. Nobody knows what's being used where, or how they're supposed to work together."

Individual productivity tools succeed when they fit naturally into how someone works. But enterprises need that productivity lift to compound across teams, departments, and thousands of employees. That's where agentic AI comes in, autonomous systems that can orchestrate multi-step workflows, make decisions across organizational boundaries, and operate without constant oversight.
The potential is substantial. McKinsey projects $2.6-4.4 trillion in productivity gains from agentic AI*.

But most enterprises aren't capturing that value yet.

To date, enterprises have experimented with different approaches to deploying agentic systems, but few have realized the expected value

Many select pre-built agents, roll them out enterprise-wide, and hope they map to how work actually happens. This achieves scale but can force standardization, agentic workflows get designed around generic processes that don't reflect how individuals actually operate. Adoption stalls when agents don't fit naturally into existing work patterns.

BAAPs: A new architectural pattern

We're starting to see a new architectural pattern emerge to solve this, and it's coalescing into a distinct category: Behavioral Agent Automation Platforms, or BAAPs.

BAAPs are the orchestration infrastructure for agentic AI. Instead of predicting workflows and configuring agents to match those predictions, BAAPs observe how work actually happens, then orchestrate agentic capabilities based on behavioral evidence. Instead of forcing standardization or requiring manual assembly, BAAPs discover patterns autonomously and deploy individualized agents at scale.

It's observation-driven orchestration, not configuration-driven deployment.

Liminal is pioneering this approach, and that's exactly why we invested. The Liminal Platform delivers secure AI capabilities, observability into how users leverage AI and their organizational memory, then autonomously identifies and deploys agentic workflows without requiring prediction or manual configuration. It is infrastructure that adapts to how organizations actually operate, and we believe it represents the orchestration layer enterprise agentic AI has been missing.
Liminal has published a comprehensive whitepaper exploring how BAAPs work and why observation-driven orchestration represents a fundamental architectural shift. Worth reading if you're thinking about what comes next in agentic AI infrastructure.


* McKinsey & Company. The economic potential of generativeAI. McKinsey & Co, June 2023