Engineering Workflow Integration Challenges

𝗘𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗔𝗜 𝗜𝘀 𝗮 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲 Much of today’s conversation around AI agents focuses on #graphs, #models, #prompts, #context, or orchestration #frameworks. These topics matter, but they rarely determine whether an AI system succeeds once it moves from prototype to enterprise production. The real challenges appear when AI systems operate inside long-running business workflows. Consider a workflow that analyzes documents, retrieves data from multiple systems, calls APIs, and produces a structured decision. Such processes may run for twenty or thirty minutes and involve dozens of steps. Now imagine something routine happens: a network call fails, an API times out, or a container restarts. No problem, the agent says. It starts the workflow again. That may be acceptable for chatbots. It quickly becomes impractical for enterprise processes such as financial analysis, document processing, underwriting, or claims review. These workflows are long-running, resource-intensive, and deeply connected to operational systems. In these situations, the limitation is rarely the model’s intelligence. More often, the challenge lies in the #engineering #discipline around the system. At Cognida.ai, our focus is on building practical enterprise AI systems rather than demos or PoCs. We consistently find that several principles from #distributedsystems engineering become essential once AI moves into production. Here are three such constructs: 𝗗𝘂𝗿𝗮𝗯𝗹𝗲 𝗘𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻 Agent workflows should not be treated as temporary requests. Each step should persist its state so that if a failure occurs, the system can resume from the last successful step rather than restarting the entire process. In practice, this means workflow orchestration with checkpointed state, deterministic execution, and event-driven recovery. For long-running processes, this is often the difference between a prototype and a production system. 𝗜𝗱𝗲𝗺𝗽𝗼𝘁𝗲𝗻𝘁 𝗔𝗰𝘁𝗶𝗼𝗻𝘀 AI agents increasingly trigger real-world actions: sending emails, calling APIs, updating records, moving files, or initiating financial transactions. Retries are inevitable in distributed systems. If actions are not idempotent, retries can create duplicate or inconsistent results. Reliable AI systems must ensure the same action cannot run twice unintentionally. 𝗣𝗲𝗿𝘀𝗶𝘀𝘁𝗲𝗻𝘁 𝗦𝘁𝗮𝘁𝗲 𝗕𝗲𝘆𝗼𝗻𝗱 𝘁𝗵𝗲 𝗠𝗼𝗱𝗲𝗹 Large language models operate within limited context windows rather than durable memory. Enterprise workflows often run longer and across many stages. The system managing the workflow must maintain its own persistent state instead of relying on the model’s temporary context. It means treating AI workflows as structured state machines, not simple prompt-response interactions. Are you treating AI workflows more like state machines, event-driven systems, or traditional #microservices? #PracticalAI #EnterpriseAI

💡 Enterprise AI’s moat isn’t the specific model. It’s integration velocity — compounded. We’ve all experienced enough agentic pilots and demos over the last few months! (seen more Pilots than British Airways! 😂). Durable advantage is now a race to wire AI into identity, data, actions, and human workflows—safely, measurably, repeatedly. Value is cross functional and requires integration across silos - leading to a recent trend to centralize more into Centre's of Excellence (actually really into Centre's of Execution!). Across thousands of use cases over the last three years, one pattern is unmistakable: the edge now is how fast you integrate, not how loudly you experiment. Here’s what the leaders do differently technically based on our real experience of scaling into production: 1) Broker‑before‑bot Trust fabric first: SSO/SCIM mapped to entitlements, DLP/eDiscovery in the prompt path, auditable agent actions. If AI can’t clear your brokers, it won’t clear your board. 2) Knowledge with rights Governed RAG that respects ACLs, emits citations, tracks lineage. Answers that stand up in audit, not just in a demo. 3) An action mesh, not a chat box Typed, approved, journaled tools into systems of record (CRM/ERP/ITSM). Agents that do real work—read the contract, open the ticket, update the record—inside policy. 4) Agent SLOs and observable economics Tracing + evals + cost budgets. Model mix and caching beat model mythology. Quality up, unit cost down, week after week. 5) Workflow rewrites New KPIs, handoffs, and exception paths for human+AI teams. Training that changes rituals, not just skills. Our best engagements seek to measure three numbers: Time‑to‑Trust (days to clear identity, policy, DLP), Time‑to‑First‑Action (days to a safe write in a system of record), Unit Cost per Outcome (what it costs to achieve the business result). Together – we can define an ‘Integration Yield’: IY = (% of workflow steps safely automated × quality uplift) / unit cost. Raise IY and pilots should turn into P&L. If your AI roadmap doesn’t start with integration, it won’t end with value. #AI #GenAI #AgenticAI #Integration #LLMOps #EnterpriseSoftware #OperatingModel Fernando Lucini Alberto García Arrieta Gavin Stephenson Nick Millman Stefano Sperimborgo Azeem Azhar Laetitia Cailleteau Pankaj Sodhi

Engineering transformation is not optional anymore, it’s a race against irrelevance! For years, we’ve all seen the same patterns in product development. - Mechanical, E/E and software teams working in isolation. - Complexity growing faster than our ability to manage it. - Errors discovered too late. - Interfaces that don’t fit. Integration often feels like assembling a puzzle… only to realize that half the pieces were built from entirely different pictures. Weeks, sometimes months, lost not because of bad engineering, but because of fragmented engineering. And yet, despite knowing these problems for years, many organizations are still waiting. Waiting for the “right moment.” Waiting for clearer standards. Waiting for others to move first. That moment is gone. Global competitors have already picked up speed and are exerting pressure. With Model-Based Systems Engineering (MBSE) and AI reaching real maturity, we finally have the tools to fix what we’ve been complaining about for a decade. The question is no longer if transformation will happen. The question is: how fast can you move? Here’s how Vlad and I currently think about it in 9 concrete steps: 1. Adopt and mature MBSE - Build system models that truly reflect your product, not just documentation. 2. Derive domain-specific models from system models - Create consistent, hierarchical product structures across all domains and disciplines. 3. Capture all engineering artifacts From requirements (RFLP) over testing to homologation, make everything explicit and create development templates. 4. Link all artifacts via a knowledge graph - Enable impact chain analysis based on a solid engineering ontology. 5. Standardize and accelerate component development - Align tools, data and processes for each discipline and component 6. Build cross-domain CI/CD pipelines - Enable fast, automated iteration across requirements, architecture, design, simulation and testing. 7. Rationalize the toolchain (APIs over UIs) - Tools must be controllable from the outside enabling agent-based workflows. 8. Make engineering knowledge machine-readable - Document not just the what, but the how and why. Only then can agents effectively navigate engineering-specific challenges. 9. Define the future work split - Clarify what engineers do and what AI agents should handle. Establish strong human-in-the-loop validation. The core message is simple: Engineering excellence in the future will not come from better tools alone. It will come from how well we connect systems, data, people and agents. Companies that start building this foundation now will gain speed. Those who wait will struggle to catch up. What’s missing from your perspective? Which steps would you add to make this transformation truly work? Timmo Sturm | Daniel Spiess | Sebastian Linzmair | Sascha Bach | Rick Bouter

Why Many PLM Evaluations and Improvement Projects Start in the Wrong Place I see the same pattern in many PLM evaluations and improvement projects: Companies start by defining dozens of individual use cases and hundreds of functional requirements in various capability areas: ✔ Document management ✔ Change management ✔ BOM management ✔ Requirements management ✔ Etc All important. But the wrong starting point. 🔹 The Core Mistake Many organizations don’t first ask a much more fundamental question: 👉 Which end-to-end processes matter most to our business, and which of those must be tightly integrated to unlock real value and efficiency gains? Without first answering that question, PLM becomes a checklist exercise: Feature A vs. Feature B Tool X vs. Tool Y Best-in-class capability comparisons The result? A technically impressive solution that optimizes individual tasks, but not the overall flow of work. 🔹 Why This Matters As I discussed in previous posts, the biggest efficiency gains come from process integration, not from isolated functional excellence. PLM is not just a collection of tools. It is the process backbone of product development. If you don’t first understand: - Where handoffs occur - Where data is recreated or reconciled - Where delays, loops, and rework originate …then no amount of detailed requirements will save you from: - Broken process chains - Excessive integrations - Productivity losses - Low ROI from PLM investments - User frustration 🔹 The Right Way to Approach PLM Evaluations 1️⃣ Identify your critical end-to-end processes (e.g., requirements → engineering → change → manufacturing → quality) 2️⃣ Determine where tight integration is essential Not everything needs to be unified, but some workflows are critical for the business and absolutely need to be integrated. 3️⃣ Define architectural principles What must be native? What can be federated? Where is latency acceptable? 4️⃣ Only then define detailed use cases and requirements Now they serve a purpose, supporting process flow, not fragmenting it. 💡 The Key Takeaway PLM architecture decisions should be driven by process integration first and tool preference second. When companies reverse that order, they often end up with individual best-in-class tools automating disjointed tasks. And that’s a very expensive way to miss the point of PLM and a huge lost opportunity. #PLM #Evaluation #Process #PLMadvisors

95% of IT leaders say integration is blocking AI. It's not the models—it's the interfaces, data contracts, and ownership decisions in the messy middle. In 2025, the biggest AI failures aren't happening in labs or boardrooms. They're happening in the unglamorous middle ground between proof-of-concept and production. From my work at DataPattern and with Fortune 500 clients, I see four integration barriers that consistently catch organizations off guard: • Legacy system compatibility: Your legacy systems-whether they're 15-year-old manufacturing platforms, financial services infrastructure, or healthcare records systems-weren't designed for AI APIs . The integration layer becomes a custom engineering project. • Change management complexity: Employees don't just need training—they need psychological safety. Fear of job displacement creates silent resistance that kills adoption. • Skill gap reality: You need people who understand both your domain AND AI implementation. That's not a hiring problem—it's a talent scarcity problem. • Workflow redesign impact: AI doesn't just automate tasks. It fundamentally changes decision points, approval chains, and accountability structures. Each barrier is harder than it looks on paper. Here's the framework I use to bridge this gap: 1․ Start with workflow mapping before technology selection 2․ Build integration teams with both AI/ML and domain expertise 3․ Design change management as rigorously as technical architecture 4․ Plan for 3x longer integration timelines than your initial estimate Execution matters more than exploration in 2025. What's your biggest AI integration challenge right now? Technical, cultural, or operational? #AI #DigitalTransformation #AIImplementation

Enterprise AI teams are struggle with agents that cannot reliably access and act on core enterprise systems and data, lack deep understanding of domain‑specific language and context, and sit on top of weak retrieval and grounding over internal knowledge, which leads to brittle answers and hallucinations. At the same time, organizations have poor observability, control and evaluation of AI behaviour, with limited guardrails, benchmarking and testing and they find it hard to compose multiple agents and tools into robust workflows, so initiatives stall in “pilot purgatory” instead of scaling into production. Part of the solution to these challenges is integrating domain‑aware models with enterprise systems and adding strong retrieval, monitoring, guardrails and orchestration for multi‑agent workflows.

PLM–ERP–MES integrations don’t fail overnight. They stall because companies get stuck in the wrong stage and assume it’s “good enough.” This is how integration actually matures in the real world. 𝗦𝘁𝗮𝗴𝗲 𝟭: 𝗙𝗶𝗹𝗲-𝗕𝗮𝘀𝗲𝗱 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻 Everything moves through Excel, CSVs, PDFs, and emails. PLM exports data, ERP and MES teams manually re-enter it. Changes are slow, errors show up late, and people - not systems - do the integration work. Reality: Integration exists on paper, not in systems. 𝗦𝘁𝗮𝗴𝗲 𝟮: 𝗣𝗼𝗶𝗻𝘁-𝘁𝗼-𝗣𝗼𝗶𝗻𝘁 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻 Systems start talking directly through APIs or middleware. EBOM flows to ERP, production orders reach MES. Some automation appears, but context is missing and changes still need manual coordination. Reality: Systems exchange data, but don’t truly understand each other. 𝗦𝘁𝗮𝗴𝗲 𝟯: 𝗣𝗿𝗼𝗰𝗲𝘀𝘀-𝗔𝘄𝗮𝗿𝗲 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻 Integration follows business workflows, not just data movement. EBOM → MBOM → production BOM transitions are controlled. Approvals, versions, and feedback loops are enforced across systems. Reality: Data moves with intent, ownership, and process control. 𝗦𝘁𝗮𝗴𝗲 𝟰: 𝗘𝘃𝗲𝗻𝘁-𝗗𝗿𝗶𝘃𝗲𝗻 𝗗𝗶𝗴𝗶𝘁𝗮𝗹 𝗧𝗵𝗿𝗲𝗮𝗱 PLM, ERP, and MES react to events in near real time. Engineering changes trigger downstream updates automatically. As-built data flows back into PLM. One product identity exists across the lifecycle. Reality: Systems operate as one connected product platform. Most organizations think they’re “integrated” when they’re actually between stages 2 and 3. The real value shows up only when integration becomes event-driven and lifecycle-aware. Which stage does your PLM–ERP–MES integration truly operate in today and what’s stopping it from moving up one level? For a deep dive into PLM, MES, or CAD and to elevate your understanding of PLM, connect with us at PLMCOACH and Follow Anup Karumanchi for more such information. #plmcoach #plm #teamcenter #siemens #3dexperience #3ds #dassaultsystemes #training #windchill #ptc #training #plmtraining #architecture #mis #delmia #apriso #mes