Overcoming resistance to process automation. What held up from 2022. What changed in 2026.

In 2022, Boyd McKenna wrote that process automation failed most often not because the technology was wrong, but because teams hadn't been brought along. The arguments were clear. Culture first. Value creation discipline. Then the bots.

Four years later, those arguments hold. Teams still resist change. Process improvement is still non-negotiable. The culture conversation still comes first.

But the scope of what automation can handle has widened. Kerolles Hany, who runs enterprise platform delivery across global markets, points to a critical shift. RPA handled repeatable, rule-based tasks. Agentic AI handles judgment calls, exception cases, and cross-system orchestration. The ceiling Boyd described in 2022 has moved up. The resistance hasn't moved at all. So the frame has to evolve.

Here are Boyd's original arguments, updated for the AI-agent era.

Culture. The argument hasn't changed.

Boyd's 2022 piece started with a hard truth. Teams resist automation because nobody told them what was coming. A bot that surprises people on Monday morning looks like a threat. A bot that people co-designed with you looks like an upgrade.

This is cultural, not technical. And it hasn't changed.

The team that finds out they're being replaced by bots without context will slow any implementation. They'll find reasons why the bot can't handle edge cases. They'll surface problems the bot might have missed. They'll be right most of the time. And they'll fight you every step of the way.

The team that co-designed the automation, that helped identify which processes actually matter, that tested the bot before it went live, that will defend it. They own it. And when the bot fails, they fix it. This culture doesn't come from the vendor. It comes from you.

For Kerolles's teams rolling out agentic systems in MENA markets and globally, the culture conversation got harder. Because the scope got bigger. RPA was narrow. It automated three people's work. Agentic AI can touch ten different teams. The cultural surface area expanded.

Which means the work upfront got more important, not less. You don't workshop an agent with ten stakeholders on short notice. You design it with them. You show them what the agent will do. You tell them what it won't do. You let them break it in controlled space. Then you run it.

The resistance softens when people understand the agent is an upgrade to their workflow, not a replacement for their role.

Value creation discipline. The path from automation to operation.

Boyd's second argument was that teams fail at automation because they automate the wrong processes. They pick what's easiest to automate, not what creates the most value. Then they end up with a faster version of their worst problems.

Value creation discipline means starting with outcomes, not tools. What outcome do you want? Faster processing? Fewer errors? Fewer escalations? Once you know the outcome, you pick the process. Once you pick the process, you fix it before you automate it.

Boyd's 2022 line was direct. "Embrace RPA without process improvement and there will be legitimate resistance born out of the fact that you are just automating bad processes."

This is harder when the tool is more powerful. An RPA bot has clear guardrails. You know what it can and can't do. An agentic system is more flexible. It can handle exceptions. It can make judgment calls. But that flexibility is worthless if the underlying process is broken.

Kerolles sees this in delivery. Teams get excited about what an agent can do. They skip the process audit. They deploy. Then the agent does things they didn't expect, because the process was ambiguous to begin with.

The discipline is the same as it was in 2022. Map the process. Fix the process. Then automate it. The tool changes. The sequence doesn't.

Boyd's framework held because it wasn't about RPA. It was about how to sequence work. Understand the culture. Understand the process. Then introduce the tool. That sequence still works for agentic systems.

What agentic AI changed. The 2026 update.

RPA in 2022 was narrow. It handled high-volume, repeatable, rule-based tasks. Structured data. Clear decision trees. When the exception hit, the bot got stuck. Someone took over. The bot moved to the next case.

The ceiling was real. And teams knew where it was.

Kerolles's perspective from global program delivery is that agentic systems broke that ceiling. An agent can now handle judgment calls. It can navigate unstructured data. It can orchestrate across systems. When it hits an exception, it can reason through it, escalate intelligently, and document what happened.

But it also broke something else. The integration architecture underneath agentic systems is more complex than RPA. An RPA bot connects to a system's UI. An agentic system needs a data fabric. It needs APIs with good contracts. It needs data lineage. It needs to understand context.

From Kerolles's Cairo Engineering Center and global delivery experience, the technical reason programs fail is usually architectural, not algorithmic. The agent is fine. The data underneath is a mess. Systems don't talk to each other. Historical data is scattered. Process metadata doesn't exist.

This is the 2026 constraint that didn't exist with RPA. You can't run a capable agent on bad data architecture.

This matters because it changes the work sequence. With RPA, you could fix the process, automate it, and go live. With agentic systems, you have to assess the data fabric before you design the agent. If it's broken, you either fix it or you scope the agent narrower. You can't automate judgment calls if the data the agent needs is hidden or inconsistent.

Cultural resistance to agentic systems sometimes masks data architecture problems. People say the agent isn't working. What they mean is the agent is asking for data they don't have. Or the data it's finding is inconsistent. That's not a culture problem. That's infrastructure.

Where RPA ends and agentic commerce begins.

The practical question that comes up in every planning conversation is the same one that came up in 2022. What should we automate with what technology?

The answer is usually both, and the sequence matters.

RPA is mature. It's proven. It's narrow. It handles the known, high-volume, repeatable cases. A transaction that hits the system 10,000 times a month and follows a clear path? RPA is your answer.

Agentic systems are newer. They're more capable in exception space. They can reason. They can handle variability. But they're also more expensive to operate and they require better data infrastructure. An edge case that happens 50 times a month and requires judgment? That's agentic.

The right answer is almost always both. Process the 10,000 cases with RPA. Process the 50 exception cases with an agent. The agent escalates to a human when it needs to. The RPA bot fails and queues for the agent.

This sequencing is what Boyd meant when he said automation is about systems running in production, not about pilots. You're not running RPA vs. agentic AI. You're running an orchestrated system that knows which tool handles which case.

Boyd was talking about RPA. But the broader point holds. The technology will keep evolving. Agentic systems will get more capable. New tools will emerge. What doesn't change is the principle. Start with culture. Fix the process. Then automate it. Know what your tool can and can't do. Run both in sequence. Measure the outcomes.

This piece shows that four years of evolution didn't break the original frame. It just expanded the scope of what gets automated, and what infrastructure you need underneath it.