When you're a dork who's spent years focused on enterprise transformation, you notice the patterns and can become a bit cynical.
About six or seven years ago, it was blockchain. Every conference, every vendor pitch, every innovation team — blockchain was going to reinvent everything. Supply chains, healthcare records, financial services. Pilots launched by the hundreds. Most of them are gone now. Blockchain didn't disappear — it found a handful of use cases where it genuinely belongs. But the revolution that was promised? That didn't happen. What happened was a lot of money spent solving problems that didn't need a distributed ledger.
Then came RPA. Robotic process automation. This one hit closer to home for me because I've spent years transforming the back offices where these bots were deployed. RPA was supposed to be the silver bullet — software that mimics human clicks, no coding required, ROI in weeks. And honestly? For about eighteen months, it delivered. Bots processing invoices, moving data between systems, handling the repetitive grind that ate up entire teams.
Then the bots started breaking. Every time a UI updated. Every time a vendor pushed a patch on a Tuesday night. Suddenly you're spending more time maintaining the automation than the process took to do manually. The miracle became a maintenance treadmill.
I watched it happen from inside the engagements. Smart people, good intentions, real budget. But the bots were layered on top of broken processes with inconsistent data, and nobody wanted to do the hard work of fixing what was underneath before they automated it.
Sound familiar?
Now it's 2026, and the shiny new toy is agentic AI. Autonomous agents that reason, plan, and execute across your systems. Every vendor deck leads with it. Every product launch mentions it.
And the data is already telling us the same story. PwC's 2025 AI Agent Survey found that 79% of companies say they've adopted AI agents. Only 17% have full adoption throughout the company. MIT studied over 300 AI deployments and found that 95% delivered zero measurable return. Not low returns. Zero.
Three hype cycles in under a decade. Three different technologies. The same outcome.
Here's what I keep coming back to: the technology wasn't the problem in any of these. Blockchain works. RPA works. AI agents work. You can see it proven every day. The technology has never been more capable or more accessible.
The problem — every single time — is that organizations reach for the technology before they've done the work to understand what they're actually trying to fix.
And before someone calls this a pattern-matching fallacy — I'm not arguing that agents will fail because RPA failed. The technologies are fundamentally different. Agents can reason. RPA couldn't. That's real progress. What hasn't changed is how organizations adopt new technology: skip the process work, skip the data work, and expect the tool to solve problems it was never designed to diagnose. MIT didn't find that the AI models were broken. They found that organizations deployed them without defined business outcomes and without data ready to support them. That's not a technology pattern. It's a human one.
Nobody wants to hear this, but process reengineering and data analytics are still your best friend. They're not legacy disciplines that got replaced by AI. They're the foundation that determines whether AI delivers or becomes another failed pilot your board stops funding.
I've read the posts. "Focus on outcomes." "AI is just a tool." I've read them a thousand times, and they're not wrong — they're just not helpful. They don't tell you what to actually do on Monday morning when you're sitting in front of a process that takes your team 40 hours a week and someone's asking why you haven't automated it yet.
Here's what to actually do: understand the process first. Not at a whiteboard level — at the level where you know where errors cluster, where handoffs create delays, what the data looks like going in and coming out, and what "good" actually means for that workflow. That's process discipline. That's data literacy. And that diagnostic work isn't a prerequisite you check off before you get to the AI part. It is the work. It's what tells you whether you need a machine learning model, a rules engine, a workflow tool, or — eventually — an agent.
Without that understanding, you're guessing. And guessing at scale is how you end up in the 95%.
Here's a move that I think is underrated, and it works fast.
Take a high-volume process that's currently 100% manual review — loan documents, claims intake, compliance checks, reconciliation. Deploy a machine learning model as a validation layer. Not replacing anyone. Not automating the whole thing. Just running AI as a first pass. The model handles the straightforward cases — the 80% that are routine — and flags the exceptions for human review.
What happens next matters more than the efficiency gain. Your team starts seeing the AI get it right. Day after day. They stop questioning whether it works and start questioning why they're still manually reviewing what it already validated. That's trust. Not the kind you build in a steering committee presentation. The kind that's built when people rely on something every morning when they sit down to work.
And while that trust is building, the data flowing through the model is getting cleaner. The exceptions it flags create a feedback loop — each one makes the next iteration smarter. You're building the data foundation and the organizational confidence at the same time, without a separate initiative for either.
Within months — not years — you've got a production system delivering real ROI, a team that trusts AI because they've watched it work, and a foundation that's ready for bigger deployments. That's when agents become a real conversation. Not because a vendor told you it was time. Because your organization earned its way there.
If you're running a regional bank, an insurance carrier, or a manufacturer doing $50 to $500 million in revenue — this is actually where you have an advantage. You're closer to your operations than the Fortune 500 will ever be. Your leaders know the people running the processes. Your data lives in fewer systems. Your decision cycles are weeks, not quarters.
The companies that will get real value from AI aren't the ones that adopted agents first. They're the ones that understood their processes, understood their data, and chose the right technology for the right problem.
Process reengineering and data analytics aren't the old way of doing things. They're still the way. The technology on top changes every few years. The disciplines underneath don't.
It's Groundhog Day out here. But it doesn't have to be.
Few tokens were harmed in this writing. ✌️