System-of-Systems Integration

Our team has a new article in #HealthSecurity arguing that epidemic preparedness requires a shift from siloed models to a systems-of-systems (SoS) approach to epidemic intelligence. Drawing on lessons from COVID-19 and a case study of highly pathogenic avian influenza (H5N1), we argue that integrating information about epidemiology, supply chains, behavior, policy, and economic systems—while respecting their autonomy—can improve situational awareness and decision support during outbreaks. The central claim is that pandemics are not single systems, and our intelligence infrastructure shouldn’t treat them as such. https://lnkd.in/eNv_UUMZ Happy to discuss implications for public health agencies, modeling teams, and funders. If you encounter a paywall, please send me a private email for access.

From Requirements to Customer Product, or the Benefits of Integrating Systems Engineering and Product Engineering Many product development challenges start with a disconnect: Requirements are defined in one tool, systems are designed somewhere else, and the engineering product structure lives in yet another system. The result is lost traceability, unclear responsibilities, and product structures that do not reflect the intended architecture. A more effective approach is to bring together Systems Engineering and Product Engineering in a continuous, integrated environment: Requirements → System Breakdown Structure (SBS) → 150% EBOM → Configured 100% products. The journey starts with requirements. These capture what the product must do: Performance targets, regulatory constraints, operational needs, and customer expectations. Requirements describe capabilities, not components. From these requirements, systems engineers develop the System Breakdown Structure (SBS). The SBS decomposes the product into systems and subsystems based on functional responsibility; propulsion, control, energy, structure, electronics, and so on. Each system becomes responsible for fulfilling a specific set of requirements and defining the interfaces to other systems. Here the product architecture begins to take shape. Product engineering then translates this architecture into the physical product structure. Each system defined in the SBS is implemented as a module or assembly in the Engineering Bill of Materials (EBOM). To support product families and variants, this is typically represented as a 150% EBOM, containing all modules and variant options across the platform. From the 150% EBOM configuration logic then selects the appropriate modules to create a specific 100% product EBOM for a customer order, region or production variant. When this process is executed in an integrated environment, powerful benefits emerge. Requirements remain traceable to the systems that fulfill them. Systems remain linked to the modules and assemblies that implement them. Changes in requirements or architecture can be traced directly to the affected product structures and configurations, and determining technical and financial impacts becomes quick and easy. This integration also supports better modularization based on changing requirements. Systems engineering defines clear functional boundaries and interfaces, which translate into well-defined product modules in the EBOM. In short, integrating systems engineering with product engineering creates a continuous digital thread: Requirements → Systems → Modules → Product Family → Customer Specific Product Configuration. And that integration is what ultimately enables companies to build complex, configurable products faster, with better control over architecture, variants, and lifecycle changes and ultimately quickly configure a product that meets specific customer requirements.

Every integration you add is a new universe you’re agreeing to live inside. And each universe has its own laws, rituals, bugs, broken edges, paperwork, and weather patterns. Before you even reach the API layer, you’re already fighting gravity: – creating the right kind of developer account – verification loops and onboarding flows – hidden settings you must toggle for the API to work – docs scattered across multiple pages – rate limits you discover by accidentally hitting them – sandboxes that behave nothing like production And this is before writing a single line of integration code. Then comes the real work: mapping endpoints, generating test data, validating schemas, handling inconsistent responses, debugging errors that surface only under specific user states, reproducing failures you didn’t even know were possible. And after shipping, the universes keep shifting: API deprecations, permission rewrites, breaking changes with zero notice, silently introduced fields, dashboard settings that suddenly matter, partner switches flipped without warning. This is what makes multi-integration systems brutal. The volatility of living inside dozens of external realities at once. Integrations are ecosystems. And building across ecosystems is never simple. That’s the world we’re trying to tame at Composio - turning these chaotic universes into something consistent, testable, and manageable so teams don’t drown in the complexity behind the scenes.

Most integration problems in enterprises today are not technical gaps. They are architectural misalignments that only show up when systems are pushed to operate in real time. What I continue to see across environments is this: Organizations invest in AI, automation, and cloud platforms, but the integration layer underneath still assumes a slower, human-orchestrated world. Workflows are predefined. Connections are tightly coupled. Change is expensive. That model holds until it doesn’t. The moment AI systems start interacting dynamically, calling services, responding to events, and orchestrating workflows on their own, the friction becomes visible. Latency increases. Dependencies break. Teams start compensating with workarounds that add more complexity instead of reducing it. This is why integration is quietly becoming the most critical architectural layer in the AI enterprise. Not as a platform you manage. But as a capability fabric embedded into how systems operate. The shift I find most important is this: from connecting applications to exposing capabilities. When business functions are available as reusable services, AI systems and applications can compose workflows dynamically. That changes not just system design, but how organizations operate at scale. In practice, getting there is not trivial. It involves unwinding years of tightly coupled integrations, dealing with hybrid environments, aligning API strategies across teams, and managing production risk while systems are still running. Most of the effort is not in building new integrations, but in rethinking how existing ones are structured and governed. This is not a one-off transformation. It is a repeatable pattern we are seeing across enterprises trying to move from digital systems to intelligent operations. We explored this in more depth in our latest blog: If you are investing in AI and automation, it may be worth stepping back and asking: Is our integration layer designed for systems that execute workflows, or for systems that define them? #EnterpriseArchitecture #AITransformation #CloudNative #SystemIntegration #DigitalTransformation #APIArchitecture #SageIT #ThoughtLeadership

Scaling energy transition is no longer a technology problem. It is a systems integration problem!! In scaling battery and energy storage platforms globally, I’ve seen this firsthand. We have made extraordinary progress—better batteries, more renewables, and increasing electrification. But scale is no longer constrained by innovation. It is constrained by integration. Today’s challenge is connecting: - Generation, storage, and demand - Hardware, software, and AI - Engineering, manufacturing, and operations - Data across fragmented value chains Without this, even the best technologies underperform. We see it everywhere: - Strong technologies, but inconsistent system outcomes - Ambitious targets, but slow execution - Abundant data, but limited decisions The bottleneck is no longer innovation. It is orchestration. The next phase of the energy transition will be led by those who can: → Integrate across the full lifecycle - from design to operations → Build digital backbones connecting physical and virtual systems → Orchestrate ecosystems, not just components → Apply AI grounded in physics and real-world data Industrial software, digital twins, and AI are not add-ons. They are becoming the operating system of the energy transition. The question is no longer: Do we have the right technology? The real question is: Can we integrate it into systems that scale? The winners will not be those who innovate the most but those who integrate the best. #EnergyTransition #AI #IndustrialSoftware #DigitalTwin #SystemsThinking #BESS #energystorage

System of systems thinking is rad. A complex system is vulnerable and limited. If it's destroyed, its constituent parts are non-functional. A system of systems is resilient. Constituent parts operate independently. Think of a giant supermarket vs a local farmers' market. The supermarket is a complex system—efficient but fragile. One power outage or supply chain disruption, and the whole thing grinds to a halt. The farmers' market is a system of systems—each vendor operates independently but works with others. If some vendors can't make it, the market adapts and carries on. Bad weather hits one farm's crops? Others fill the gap. I think a system-of-systems approach needs to be applied to AI on a broad scale. Sae Schatz has the right idea. On the Warfighter Podcast with Tom Constable and Colin H., she advocates for “a modular open systems approach, where we make each Lego block within the larger system its own standalone device.” For complex, dynamic environments, there's no single stand-alone product that will fit the bill. If your use case is basic, sure, go ahead and use ChatGPT or some other product. For environments that need a specific capabilities set assembled to spec, with the ability to rapidly deploy new capabilities or reconfigure, a system of systems approach is what you need. Plug and play interoperable modules into a larger ensemble, which itself can be a module in even a larger ensemble. It's mindblowing what capabilities this can enable. Take drug discovery. One system analyzes molecular structures and binding properties. Another processes genetic pathway data. A third examines clinical trial outcomes and patient data. A fourth scans scientific literature and research papers. Each runs independently, mastering its domain. But combine them into a higher-level system, and patterns emerge. That molecular structure matches this genetic pathway, connecting to those clinical results, backed by emerging research trends. Stack this ensemble into an even larger system that correlates with global disease patterns and antibiotic resistance, and you've got an AI that spots promising drug candidates while anticipating future needs. No single AI model could handle this complexity. At Talbot West, we've developed a term for system-of-systems AI ensembles. We call it Cognitive Hive AI (CHAI). It's the future, and we're proud to be part of it. I wrote an article expanding on the thoughts of this post. Link in comments. #artificialintelligence #systemofsystems #chai #cognitivehiveai #talbotwest