Composabl Inc. is an industrial artificial intelligence (AI) and software-as-a-service (SaaS) company headquartered in the United States. Founded by Kence Anderson, a veteran in the field of autonomous systems and the former lead of Microsoftâ??s Autonomous Systems team (which incubated Microsoft Bonsai), Composabl focuses on bringing advanced automation to the manufacturing, design, and heavy industrial sectors.
The company's core mission is to bridge the gap between traditional industrial automation (such as Programmable Logic Controllers and PID controllers) and modern artificial intelligence. Composabl achieves this through a proprietary platform that enables industrial engineers to design, train, and deploy "Autonomous Intelligent Agents" to manage complex, dynamic, and real-time operations without requiring deep expertise in data science or reinforcement learning.
Core Technology Philosophy: Machine Teaching
Unlike conventional artificial intelligence models that rely purely on massive historical datasets (Machine Learning), Composablâ??s platform is built on the concept of Machine Teaching.
Machine Teaching flips the AI paradigm by focusing on the subject matter expert (SME)â??such as a chemical engineer, plant operator, or packaging specialistâ??rather than the data scientist. The SME breaks down a complex industrial task into smaller, manageable "skills" (modular tasks) and teaches these skills to an AI agent using structured guidance, heuristic rules, and simulations.
Key pillars of this technology include:
* Deep Reinforcement Learning (DRL): The underlying technology that allows agents to learn optimal control policies through trial and error in a simulated environment.
* Modular Agent Architecture: Agents are composed of multiple specialized "skills" (e.g., a startup skill, a steady-state operation skill, and an emergency shutdown skill) governed by a master selector brain.
* Simulation Integration: Because training RL agents in a live physical factory is dangerous and costly, Composabl utilizes high-fidelity simulation environments to train agents safely in virtual replicas of industrial processes.
Products and Services
Composabl provides an end-to-end software ecosystem designed to transition industrial facilities from manual or semi-automated operations to fully autonomous controls.
1. The Composabl Platform
The flagship offering is the Composabl Platform, an integrated suite that allows developers and industrial engineers to build, train, and deploy AI agents. The platform consists of two primary layers:
- Composabl SDK (Software Development Kit): A Python-based developer framework that allows engineers to write code to define agent structures, register simulation environments, specify reward functions, and manage agent training cycles. The SDK is designed to be highly extensible and integrates cleanly into existing developer workflows.
- Composabl Console (User Interface): A visual dashboard and low-code interface designed for process engineers and operators. The Console provides real-time visualization of agent training, performance metrics, reward histories, and comparative benchmarks against traditional control baselines.
2. Simulation Connectors
To train reinforcement learning agents, the agent must interact with a simulator. Composabl provides seamless integration pipelines and pre-built connectors for industry-standard simulation software, including:
* MATLAB and Simulink (widely used in control systems engineering)
* AnyLogic (used for logistics and system dynamics)
* Microsoft AirSim and various robotics simulators
* Custom Python simulators (such as Gym/Gymnasium environments)
These connectors allow the Composabl agent to read states from the simulator, execute actions, and receive rewards in microsecond loops, accelerating the training process through parallelization in the cloud.
3. Industrial AI Agents (The Deployable Runtimes)
Once trained, the output of the Composabl platform is an Autonomous Agent. These agents are packaged as lightweight containerized microservices (typically Docker containers) that can be deployed:
* On-Premise / Edge: Running on industrial PCs or edge gateways directly on the factory floor, communicating with PLCs via protocols like OPC UA, Modbus, or MQTT.
* In the Cloud: Executing in hybrid cloud environments for higher-level optimization tasks (such as supply chain routing, energy management, or production scheduling) that do not require millisecond-level latency.
4. Professional Services and "Agent Academies"
To support enterprise adoption, Composabl offers structured professional services:
* Proof of Value (PoV) Projects: Collaborating with industrial enterprises to identify high-value automation bottlenecks, build custom simulators, and deliver working pilot agents.
* Training and Enablement: Training programs to upskill traditional control and process engineers into "Machine Teachers," allowing enterprises to scale AI agent development internally.
Primary Applications and Industry Use Cases
Composabl's products are designed for industries characterized by high variability, complex dynamics, and multi-variable optimization challenges where traditional automation reaches its limits.
1. Process Industries (Chemical, Food & Beverage, Oil & Gas)
In chemical processing, operators must manage multi-phase reactions, temperature fluctuations, and raw material variations.
* Application: Composabl agents manage closed-loop control of distillation columns, bioreactors, and industrial extruders. The agent can adjust valve inputs, heating elements, and feed rates simultaneously to maximize throughput and minimize energy usage while keeping the system within safety constraints.
2. Discrete Manufacturing and Robotics
Traditional industrial robots are programmed using rigid, deterministic scripts. If a part is slightly misaligned, the robot fails.
* Application: By deploying a Composabl agent to control robotic arms or CNC machinery, the system gains "eyesight" and adaptive capabilities. The agent can adjust toolpaths in real-time based on sensor feedback, compensate for wear and tear, and handle irregular geometries.
3. Packaging and Logistics
High-speed packaging lines often suffer from frequent downtime due to material jams, tension imbalances, or changeover delays.
* Application: Composabl agents optimize web tension and speed synchronization in packaging machines, predicting and preventing jams before they occur and automating the calibration process during product changeovers.
How Composabl Differs from Traditional Control Systems
| Feature | Traditional Automation (PLC / PID) | Composabl Autonomous Agents |
| :--- | :--- | :--- |
| Logic Type | Rule-based, mathematical, deterministic | Adaptive, goal-oriented, predictive |
| Variable Handling | Struggles with highly coupled, multi-input multi-output (MIMO) systems | Excels at balancing hundreds of variables simultaneously |
| Adaptability | Requires manual recalibration when environmental conditions change | Continuously adapts to sensor drift, material changes, and wear |
| User Persona | Control Engineers writing Ladder Logic / Structured Text | Process Engineers defining goals and constraints via Machine Teaching |
