32:57
The Next 2 Years...
4.0 Solutions
Overview
This video explores the evolving landscape of the industrial technology sector over the next two years, focusing on predictions and advice for end-users, OEMs, and systems integrators. It highlights the increasing importance of Model Context Protocol (MCP) and the shift in the build-versus-buy calculus, particularly with the rise of AI. The summary breaks down industry evolution into Industry 3, 4, and 5, emphasizing the convergence of human and artificial intelligence in Industry 5. Key takeaways include the need for full-stack fluency in new technologies, understanding the distinction between deterministic and probabilistic control, and adapting to new software licensing models.
How was this?
Save this permanently with flashcards, quizzes, and AI chat
Chapters
- The video addresses recurring questions about the next two years in the industry, specifically for end-users, OEMs, and integrators.
- The central theme is predicting industry trends and providing guidance for navigating these changes.
- The speaker's past prediction about the importance of Model Context Protocol (MCP) has proven accurate, but its applications have expanded beyond initial expectations.
Understanding the core questions and the speaker's foundational predictions sets the stage for the detailed analysis of future industry directions.
The speaker references a speech given 16 months prior where they emphasized the necessity of knowing MCP.
- MCP is emerging as a significant competitor to traditional RESTful APIs for serving documentation and enabling agent-based interactions.
- Software vendors are increasingly adopting MCP, sometimes replacing or running it in parallel with existing APIs, as seen with Google Cloud services.
- MCP is also being integrated into Software Development Kits (SDKs) by OEMs, allowing them to build tools and expose them via MCP for internal use.
- A novel application of MCP is its use as a conduit for agents to save and retrieve contextual knowledge for later use.
MCP's versatility and growing adoption signal a fundamental shift in how software and data interact, impacting integration strategies and development practices.
Google Cloud's parallel MCP integrations for services like Calendar and Gmail are cited as examples of its power and ease of integration compared to traditional methods.
- OEMs face a challenging future, with many expected to struggle or disappear, while new ones will emerge.
- Key questions for OEMs include 'Why wouldn't someone build this themselves?' and 'What is novel about our offering?'
- Successful OEMs will focus on being part of a larger ecosystem, leveraging open architectures, and enabling MCP and unified namespaces.
- Companies that prioritize sales and leveraging existing install bases over building the best product will face significant difficulties.
This section provides critical self-assessment questions for software and hardware manufacturers to ensure their long-term viability in a rapidly changing market.
The speaker contrasts companies that build the best product with those, like Microsoft (as an example of a strategy), that focus on compelling customers to buy through existing influence and install bases, warning that the latter approach is problematic.
- Industry 3 integrators (automation, PLCs, HMIs) will persist but face commoditization.
- Industry 4 integrators focus on merging IT and OT data for decision-making, requiring an understanding of human-AI convergence.
- Industry 5 integrators specialize in the human-AI convergence, understanding the nuances between deterministic and probabilistic control.
- Large-scale AI initiatives from major tech companies may initially falter due to a lack of manufacturing domain knowledge, creating opportunities for specialized Industry 5 integrators.
Understanding the distinctions between Industry 3, 4, and 5 integrators clarifies the evolving roles and skill sets required to bridge the gap between automation, data, and intelligent decision-making.
The speaker explains that while large companies like Nvidia and ServiceNow might offer AI teams, they may fail because their personnel lack practical manufacturing floor experience, unlike specialized Industry 5 integrators who understand the 'ever-changing nature of OT'.
- Deterministic systems (like PLCs) guarantee the same output for the same input within a specific time frame, crucial for safety and industrial processes.
- AI, particularly Large Language Models (LLMs), operates probabilistically, meaning outputs can vary even with identical inputs.
- Agents are well-suited for tasks like data contextualization, summarization, orchestration, and ambiguous judgment-based work where 'good enough, fast' is acceptable.
- Agents are not suitable for closed-loop real-time control, safety functions, or tasks with hard timing deadlines where 'probably right' is insufficient.
Distinguishing between deterministic and probabilistic systems is crucial for correctly applying AI agents in industrial settings, ensuring safety and reliability where it matters most.
The speaker uses the analogy of PLCs being 'always right' with 'nine nines of reliability' compared to LLM-based agents which, by design, can produce different paths or tool calls for the same prompt.
- The build-versus-buy decision has changed: buy infrastructure components (data infrastructure, databases, analytics layers) and build user-facing applications (analysis, visualization, pattern finding, reporting).
- Full-stack fluency is essential, but the 'stack' has evolved to include LLMs, context management, tokconomics, MCP, multi-agent workflows, and knowledge graphs.
- Key areas to study include model layers (LLMs, tokconomics), orchestration (MCP, tool calling), data layers (UNS, knowledge graphs), and engineering disciplines (agentic DevOps, eval).
- Evaluating vendor AI claims requires literacy, as many current claims are unsubstantiated.
This chapter outlines the practical implications for development and learning, guiding individuals and organizations on what to build, what to buy, and what skills to acquire.
The speaker explains that while components like databases and data transformers will be bought, the user analysis, visualization, and pattern-finding layers will be built, contrasting this with the past where entire MES systems were often built.
- The core problems remain 'data to decision' and achieving more with less, but the mechanisms for solving them are changing drastically.
- The tendency to pay for 100% of a product for only 1% of its features will diminish; users will seek value aligned with actual usage.
- Future software licensing models will likely be 'freemium,' where core functionality is free, and users pay for aspects requiring ongoing maintenance and support (e.g., connectors, integrations).
- End-users should expect AI to enhance human capabilities rather than replace them, focusing on improving decision-making and leveraging saved time.
This section addresses the future of software consumption and the ultimate goals for end-users, emphasizing efficiency, value, and the augmentation of human potential through AI.
The speaker suggests that platforms like Ignition might become free, with costs associated only with connecting that platform to external services, reflecting a shift towards paying for maintainable, integrated components rather than the entire software suite.
Key takeaways
- Model Context Protocol (MCP) is rapidly becoming a critical technology, challenging traditional APIs and enabling new forms of software interaction.
- The distinction between Industry 3 (automation), Industry 4 (data to decisions), and Industry 5 (human-AI convergence) is crucial for understanding future market roles.
- AI agents are powerful for probabilistic tasks like data analysis and orchestration but are unsuitable for deterministic, real-time control systems.
- Organizations should focus on buying core data infrastructure and building user-facing analytical and visualization layers.
- Acquiring skills in areas like LLMs, context management, tokconomics, and agentic workflows is essential for future relevance.
- The software industry is moving towards licensing models where users pay for maintainable integrations and services, not necessarily the entire platform.
- The primary goal of AI in the next two years will be to augment human capabilities and improve decision-making, not to replace workers.
Key terms
Model Context Protocol (MCP)Build vs. BuyIndustry 3Industry 4Industry 5Deterministic ControlProbabilistic ControlAgentsLarge Language Models (LLMs)Unified Namespace (UNS)TokconomicsRESTful APIs
Test your understanding
- How has the adoption and application of Model Context Protocol (MCP) evolved beyond its initial predicted use cases?
- What are the key differences between Industry 4 and Industry 5 systems integrators, and why is this distinction important?
- Explain why AI agents are considered probabilistic and how this characteristic makes them unsuitable for certain industrial control applications.
- According to the speaker, what types of software components should organizations prioritize buying versus building in the next two years?
- What is 'tokconomics,' and why is it considered an important area of study for the future?