The AI conversation in manufacturing has reached an inflection point. After two years of pilots, proofs of concept, and boardroom enthusiasm, a pattern is emerging: the companies seeing real operational impact from AI are not the ones with the most sophisticated models. They are the ones applying AI within purpose-built systems that understand their specific workflows, data structures, and delivery requirements.
This distinction, between horizontal AI and vertical AI, is becoming the defining strategic question for manufacturers evaluating where to invest next.
Two Terms Worth Defining Clearly
Horizontal AI refers to general-purpose AI capabilities: large language models, computer vision frameworks, generative tools designed to work across industries and use cases. These are powerful technologies. They can analyze documents, generate text, process images, classify data, and surface patterns at scale. They are also foundational. Nearly every AI application in existence, whether consumer or industrial, builds on horizontal AI capabilities at some level.
Vertical AI takes those same capabilities and embeds them inside domain-specific workflows, data structures, and delivery systems built for a particular industry. The underlying model may be the same. What changes is everything around it: the context the AI operates within, the data it draws from, the structure it imposes on its outputs, and the systems it connects to upstream and downstream.
This is not a competition between two opposing approaches. Vertical AI is horizontal AI given a job description, a set of constraints, and a production environment. The most effective vertical AI platforms use horizontal AI principles at their core, an LLM as the interface layer, orchestration agents coordinating specialized tasks, generative models handling specific activities, while wrapping all of it in the domain knowledge, compliance requirements, output formatting, and workflow logic that a specific industry demands.
Horizontal AI is genuinely valuable for general-purpose productivity, custom integrations, data analysis, and gathering insights across broad domains. The question is not whether horizontal AI works. It is whether horizontal AI alone is sufficient for the mission-critical, last-mile workflows where industries need the most help.
Where Manufacturing Actually Lags
Manufacturing runs on three pillars: product, process, and people. The product pillar has been transformed by engineering systems, digital twins, and simulation. The process pillar has been transformed by smart factory automation, IoT, and advanced execution systems. Both have absorbed billions in investment and delivered real results.
Vertical AI is already emerging across all three. Purpose-built AI for predictive maintenance, quality analytics, and supply chain optimization is gaining traction precisely because manufacturing’s data is complex, proprietary, and deeply embedded in specialized systems like CAD, PLM, MES, and ERP. Generic AI tools were never designed to navigate those structures natively.
But the most urgent and underserved opportunity sits in the people pillar. The way engineering knowledge gets translated into guidance for the workforce has not fundamentally changed. Workers are still relying on static documents, procedures stitched together from screenshots and tribal knowledge, and instructions authored in isolation from the engineering systems upstream. Products and processes entered the modern era. Knowledge delivery to the people doing the work did not.
This matters because of what is happening to the workforce itself. The Deloitte and Manufacturing Institute project that 2.1 million manufacturing jobs could go unfilled by 2030. Experienced workers are retiring and taking decades of practical wisdom that was never formally captured. New workers are arriving expecting visual, interactive tools and encountering paper-era processes. The gap is widening from both directions simultaneously.
When AI entered the manufacturing conversation, it naturally gravitated toward the two pillars that already had sophisticated digital infrastructure: better predictive models for engineering, smarter scheduling for production, more refined analytics for quality. AI was wired into the systems that already worked. The pillar that was actually broken, the people side, received comparatively little attention.
Why Horizontal AI Cannot Close This Gap Alone
Manufacturing is not a domain where general-purpose tools translate easily into production-ready solutions. The data is proprietary and deeply structured: engineering models in CAD, product definitions in PLM, production logic in MES, quality records in specialized compliance systems. The workflows are regulated, safety-critical, and intolerant of error. The integration requirements run deep.
General-purpose AI tools are remarkably good at generating information. They can analyze engineering drawings, parse legacy documentation, transcribe videos, and produce structured text from unstructured inputs. That capability is real and useful. But it is not enough.
Information is not the bottleneck. The bottleneck is structuring that information into executable guidance within the specific context of each workflow, connecting it to the enterprise systems that govern the process, delivering it to the right person at the right moment, and capturing feedback that flows back upstream. A general-purpose AI tool can generate better content. It cannot, on its own, embed that content into the production environment where it needs to operate.
Consider a recent example from the AI industry itself. In March 2026, OpenAI discontinued Sora, its video generation model, just six months after launch. Disney had licensed characters for it. A billion-dollar investment was on the table. The technology was genuinely impressive. But OpenAI chose to redirect those resources toward reasoning and productivity, the areas where it was actually winning revenue and retention.
The lesson is not that Sora failed. It is that even the most well-funded AI lab on earth concluded that raw horizontal capability, without vertical depth, domain-specific workflows, and purpose-built delivery, does not translate into sustainable value. That pattern is instructive for every industry evaluating where to invest in AI.
What Vertical AI Actually Solves
The case for vertical AI in manufacturing applies across quality, maintenance, and supply chain workflows. But the clearest illustration of why vertical depth matters, and where the stakes are highest, is in the people pillar: how knowledge flows from engineering to the workforce and back.
This problem has two distinct ends. I describe them as the first mile and last mile of the digital thread.
The first mile is capturing knowledge from engineers and experienced workers and turning it into structured, reusable instructions. This step has historically been manual, labor-intensive, and disconnected from engineering systems. Someone takes CAD drawings, captures screenshots, photographs prototypes, and manually assembles all of it into step-by-step procedures. The effort compounds every time product complexity increases or a design revision comes through.
The last mile is delivering that knowledge to the technician or operator at the moment they need it, in a form they can immediately act on. And critically, capturing their feedback, inspection results, and performance data and flowing it back into the systems upstream.
Vertical AI, embedded within a platform that understands manufacturing workflows, can compress the first mile dramatically. AI co-pilots working alongside subject matter experts can ingest engineering documents, legacy service manuals, even videos of experienced workers performing procedures, and help generate structured, visual, interactive instructions. The expert stays in control. AI handles the heavy lifting of structuring, sequencing, and illustrating. For companies facing imminent retirements, the ability to capture decades of institutional wisdom in days rather than months changes the equation entirely.
On the last mile, vertical AI operating within a purpose-built delivery platform can adapt how instructions are presented based on the device, the task, and the worker’s experience level. A new worker gets complete step-by-step guidance. A veteran sees only what has changed for this specific job.
Safety steps and design changes get intentional friction requiring acknowledgment. The instruction adapts to the environment rather than the other way around.
None of this is possible with a general-purpose AI tool generating documents in isolation. It requires vertical depth: understanding how manufacturing data flows from engineering to the floor, how work instructions need to be structured for execution, and how feedback needs to travel back upstream to close the loop.
The Architecture That Makes It Work
The most effective vertical AI implementations in manufacturing are not built from scratch on top of a raw model. They combine horizontal AI capabilities with an established vertical SaaS platform that provides the domain context, workflow logic, and delivery infrastructure.
In practice, this means an LLM serves as the interface layer, enabling natural interactions for content creation and knowledge capture. Orchestration agents coordinate specialized tasks: ingesting a CAD model, analyzing a legacy document, structuring procedural steps, generating visual outputs. Each task may invoke more specialized, purpose-built AI processes tuned for that specific activity. And all of it operates within a vertical platform that controls the context, enforces the structure, formats the output for the target environment, and connects bidirectionally to enterprise systems like PLM and MES.
This architecture matters because it solves the production-readiness problem that trips up most enterprise AI initiatives. The horizontal AI provides the raw intelligence. The vertical platform provides the guardrails, the domain knowledge, the delivery mechanism, and the feedback loop that turn raw intelligence into reliable operational outcomes.
The Make-vs-Buy Reality
Some manufacturers, recognizing the people gap, attempt to build their own solution using general-purpose AI tools. The pattern is predictable. A team takes a general-purpose LLM, connects it to some internal data, and builds a prototype. The demo impresses leadership.
Then reality sets in. The prototype needs to connect to engineering systems, so it stays current when designs change. It needs a delivery mechanism that works on the factory floor. It needs device adaptation, role-based access, compliance tracking, and a feedback loop. Each requirement is its own engineering project.
Now the team needs developers building connectors to PLM and MES, a content management layer, security and access controls, and ongoing maintenance for all of it. The company is no longer in the manufacturing business. It is in the custom software business, building infrastructure that is not its core competency.
The calculus is straightforward. Horizontal AI gives you raw capability. A vertical platform gives you a production-ready solution with the domain architecture, enterprise integrations, and delivery infrastructure already built. Every month spent building custom AI tooling is a month competitors using purpose-built platforms are pulling ahead, because the knowledge advantage compounds. Faster onboarding, better quality, fewer errors, faster launches. Each cycle reinforces the next.
Where This Goes Next
The vertical AI wave in manufacturing is just beginning, but the trajectory is clear. Across quality, maintenance, supply chain, and workforce enablement, the organizations that move first will not gain a linear advantage. They will gain a compounding one.
Nowhere is that compounding effect more visible than in the people pillar. Every piece of institutional knowledge captured is knowledge that does not walk out the door with the next retirement. Every worker onboarded faster is a worker contributing sooner. Every feedback loop closed is a process that improves continuously.
The industry has spent two decades connecting machines, systems, and data streams. That investment was necessary. But the most important node in any factory is not a machine. It is the person standing in front of one.
Horizontal AI gave the industry powerful new tools. Vertical AI is what will finally point those tools at the problem that matters most: getting the right knowledge to the right person at the right moment.
Smart factories need connected knowledge, not just connected machines. The technology exists. The question is how quickly organizations will act on it.
