.png)
This article was originally written for and published by Business Insight Journal.
Ross Meyercord, as CEO at Propel, you’ve closely witnessed the evolution of AI from automation tools to autonomous systems. What has shaped your perspective on building a balanced relationship between human expertise and AI-driven decision-making?
My perspective has been shaped by three decades in enterprise technology, watching companies adopt transformative tools and either capture the value or get burned by the hype. The pattern repeats: early returns disappoint, organizations overcorrect, and the technology ultimately delivers when businesses treat it as a structural shift rather than a shortcut.
With AI, I’m seeing manufacturers move faster than I’ve ever seen in previous technology cycles. It took them a decade to trust the cloud and roughly a year to start trusting AI. But speed without discipline is dangerous, especially in regulated industries where a bad decision in a change order or quality process has real consequences. The balance I advocate for is to let AI handle the high-volume, rules-based work, the tasks that grind down your best people, and preserve human judgment for the decisions that materially impact downstream accountability.
Gartner predicts that a significant percentage of work decisions will soon be handled autonomously through agentic AI. What does this transition realistically look like inside product-focused organizations over the next few years?
Realistically, it’s not a sudden handoff. It’s a staged expansion of agent authority tied directly to demonstrated trust. In the near term, manufacturers are deploying agents to run simultaneous processes that have always been painfully serial. Take a change order: today, dozens of approval steps run sequentially, and a routine engineering change can take weeks to approve. With agents assisting that change order, that same process can run in a fraction of the time. That’s not a futuristic vision, that’s happening now.
Over the next few years, you’ll see organizations expand agent scope from task execution to multi-step orchestration across product lifecycle management (PLM), quality management (QMS), enterprise resource planning (ERP), and supply chain data simultaneously. The companies that get there fastest are the ones who’ve already unified their product data. The ones who haven’t will find that agentic AI exposes their architecture problems with brutal clarity.
Conversations around agentic AI often swing between excitement and fear. Why do you believe the industry still misunderstands the role humans will continue to play in AI-powered environments?
The misunderstanding comes from framing AI as a replacement engine rather than a capability multiplier. Fear tends to dominate when the conversation stays abstract. Exuberance takes over when people see a demo. Neither response is particularly useful. What I tell the executives I work with is this: agents aren’t self-governing systems, they need to be managed like any high-performing employee. They need coaching, feedback, and ongoing calibration. We had an early customer whose quality agent was pulling data from superseded documents. The agent wasn’t broken, the instructions weren’t specific enough. We adjusted the prompt in minutes and it was back in production the next day. The human role isn’t disappearing. It’s shifting from doing the work to directing and validating the work. That requires different skills and a different mindset, but it’s still fundamentally human judgment that determines whether the output is trustworthy and the business outcomes are sound.
Many businesses are rushing to integrate AI into existing operations. What foundational mistakes are companies making when they treat AI as a plug-in rather than a structural transformation?
The biggest mistake is assuming AI will be like waving a magic wand – dropping AI on top of fragmented, disconnected systems and expecting it to magically begin to perform. Agentic AI doesn’t function in silos. An agent optimizing supply chain decisions needs simultaneous access to engineering specs, quality records, supplier performance, customer commitments, and financial data. If that information lives in five disconnected, inaccessible systems, the agent’s output is compromised. It’s an architecture problem, not an agent problem. The second mistake is treating AI deployment as an IT project rather than a business transformation. The business has to own the outcomes. Teams need to actively define agent roles, tune instructions based on real outputs, and monitor results against measurable targets. The true litmus test for any AI deployment is simple: are you confident enough to adjust key business metrics based on it? If not, the solution isn’t ready for production, or you frankly aren’t ambitious enough with your scope.
Human-in-the-loop systems are becoming central to responsible AI adoption. What practical safeguards should organizations establish to ensure humans remain actively involved in oversight, validation, and governance?
Start with data governance, it’s non-negotiable. If an employee doesn’t have access to sensitive data in your system, they shouldn’t be able to retrieve it through an AI-powered query either. Role-based permissions and access controls need to extend into your AI environment, not be left behind at the application layer. Beyond access, require citations. Every agent response should surface its sources so humans can verify the reasoning, not just accept the output. Then define agent authority explicitly before deployment: what can the agent recommend, what can it do with human confirmation, and what can it execute autonomously? Build structured review checkpoints at each tier, especially for any agent action that modifies records, triggers workflows, or influences procurement and quality decisions. These aren’t bureaucratic guardrails. They’re what allow you to scale with confidence, and they’re what give you the audit trails needed to demonstrate compliance when regulators come calling.
Product companies are under pressure to innovate quickly while maintaining trust and accountability. How can leadership teams balance experimentation with responsible deployment when implementing autonomous AI systems?
Don’t run pilots in a lab, put them in production with a defined scope and real stakes. The challenge is whether the business can adapt to it and whether the outputs align with your intended objectives. In manufacturing, where a misstep can trigger a recall or violate FDA compliance, you need a staged approach to expanding agent authority, but you need it running on real data from day one, not a sandboxed environment with clean inputs that don’t reflect how your business actually operates.
My advice to product company leaders is to define what trust looks like in measurable terms before you deploy. What outcome signals tell you the agent is performing correctly? What threshold triggers a human review? Build those criteria upfront. The customers I see winning are the ones using real production data, real workflows, and real business staff to tune the outputs from day one, not proof-of-concept teams working in isolation who hand off to operations six months later.
As AI begins influencing operational and strategic decisions, how should organizations rethink accountability, ownership, and decision authority across teams?
Accountability doesn’t move to the AI agent. When an agent makes a recommendation or takes an action, a human owns that outcome. What changes is where accountability lives operationally. In a traditional process, the engineer or quality manager who runs the analysis owns the result. In an AI-augmented process, the business owner who defined the agent’s instructions, approved its data access, and reviewed its outputs before acting, carries that responsibility. That’s actually a higher-order accountability, not a diminished one. Organizations need to redesign their decision authority frameworks to reflect this by putting business stakeholders in the driver’s seat on agent configuration and output review. Audit trails aren’t optional in this environment. They’re how you demonstrate compliance, resolve disputes, and continuously improve the system
One of the biggest concerns surrounding AI is workforce disruption. How can companies reposition AI as a capability enhancer that expands human potential rather than a replacement for talent?
The honest answer is that both things are true. AI will reduce certain roles and it will expand what others can accomplish. In our survey of 800 U.S. product company employees, 52% of those actively using AI reported productivity gains and 32% cited the ability to redeploy workers to higher-value tasks. I can speak to this firsthand. Our own engineering team is now 50% more productive as a result of AI, and that number is still climbing as we continue to embed these tools deeper into how we work. We have used that additional capacity to ship more software vs cutting back on staff.
What matters for leadership is being intentional about which outcome you’re driving. I respect manufacturers who are upfront with their teams and share that they are going to use AI to eliminate tedious, error-prone work, while redeploying people into roles that require judgment, customer context, and domain expertise that no agent has yet. That’s a credible strategy. What erodes trust, and ultimately adoption, is using productivity gains as cover for indiscriminate cuts while leaving behind the people best positioned to coach and improve the very agents that replaced them.
You’ve spoken about the importance of building high-impact human-AI partnerships from the start. What are the five strategic areas product companies should prioritize first to create sustainable AI adoption?
Based on what I’m seeing from manufacturers who are getting real results, these five areas set the foundation: First, unify your product data into a single connected record across PLM, QMS, ERP, CAD, supplier inputs. In order for AI agents to drive real benefit for your business, they have to be fed with coherent data. Second, define clear agent roles and “jobs to be done” before you deploy anything. Vague instructions produce unreliable outputs. Third, mirror your human access controls in your AI environment. Role-based permissions need to carry through, not be bypassed at the agent layer. Fourth, embed agents into existing workflows rather than building parallel systems. AI should reduce the number of tools your teams use, not add another one. Fifth, assign business owners to each agent who are accountable for tuning instructions, reviewing outputs, and improving performance over time. The agents that deliver the most value six months in are the ones that have had an active human hand on the wheel, coaching, adjusting and raising the bar on what good looks like.
Looking ahead, what mindset shift will separate organizations that successfully scale agentic AI from those that struggle to move beyond experimentation?
The organizations that scale agentic AI will stop asking “where can we pilot AI?” and start asking “how do we fundamentally redesign this process for an agentic world?” Those are very different questions. In 2026, the winners won’t be those with the most AI features, they’ll be the ones who combined AI agents with the governance, workflows, and connected data architecture that makes those agents trustworthy at scale. Forward-looking organizations are already operating on the next layer.
Model Context Protocol (MCP) is the emerging standard that allows agents to move fluidly across enterprise systems without hardwired integrations. When your agents can pull from PLM, ERP, supply chain, and CRM simultaneously in natural language, without an IT project to connect each one, the productivity ceiling rises dramatically. The companies still treating AI as a separate initiative, running it in isolated tools disconnected from their core systems, will watch their advantage erode. Standalone software that can’t participate in an agentic ecosystem is becoming irrelevant. The same is true for organizations that can’t rewire their operating model to work alongside intelligent automation. The mindset shift is from AI as an experiment to AI as infrastructure.
Author’s Advice: Get in the game, and get in fast, but do it with intention. Put real use cases into production with real data, and hold business staff accountable for the outcomes. Pay attention to MCP specifically. The days of nine-month IT integration projects just to get two systems talking are ending, and leaders who understand that now will make smarter platform decisions before the window closes. Above all, stay anchored to business outcomes. The manufacturers who keep their focus on reducing errors, shortening cycle times, and protecting margins are the ones who will define what this industry looks like five years from now.
.png)
