By: Lisa Samuel
Agentic AI is simply the thing that is about to expose it.
Every major technology shift changes the interface through which people interact with information. The internet changed how information was distributed. Mobile changed how it was accessed. Artificial intelligence is changing how information is understood and acted upon. Each shift arrived wrapped in noise, and each time the noise obscured the same quiet truth: the technology itself was never what separated the winners from everyone else. What separated them was whether the work underneath was organized well enough to use it.
I think that is the lens worth bringing to the current moment, because the conversation in our industry has gotten loud and a little lost. The message from every direction is that artificial intelligence is about to transform recycling and you had better buy some before you are left behind. I want to offer a calmer read, and one I believe is more useful. Auto recycling does not have an AI problem.
It has a workflow problem. AI is simply the thing that is starting to expose it—and the next wave of it will expose it far more sharply than the chatbots have.
What People Mean by AI, and What is Actually Coming
Most of what gets called AI today is generative AI: you ask a question, you get an answer. It drafts, it summarizes, it responds. It is genuinely useful, and it is also fundamentally passive—it produces text and waits for a human to do something with it. That is the version of AI nearly every vendor in the space is selling, and it is worth being clear-eyed that it is a feature, not a transformation.
The shift the technology world is actually focused on is different, and it has a name:
agentic AI. Where a chatbot answers a question, an agentic system pursues a goal—it can gather information, make decisions, and take a sequence of actions across multiple steps with limited human intervention. The distinction is simple and it is everything. Traditional AI answers questions. Agentic AI executes processes. It is the difference between a tool that tells you what to do and a system that goes and does it.
The distinction is easiest to see outside auto recycling. Ask a chatbot to help you plan travel and it returns a list of suggestions you then go book yourself. Give an agentic system the same goal and it works toward the outcome—comparing options, making the arrangements, handling the steps in sequence—rather than handing the work back to you.
That is the leap: from a tool that produces an answer to a system that pursues a result. It is the direction the entire field is moving, which is why the people building frontier AI are far less impressed by chatbots than the market is. They understand the chatbot is the shallow end.
The Model is Becoming Infrastructure
There is a deeper shift underneath, and it changes where the advantage lives. For the past few years the AI conversation was about models—which was most capable, who had access. That era is closing. Capability is commoditizing quickly, costs are falling, and the gap between competing systems is narrowing toward irrelevance for most practical uses. Intelligence is becoming infrastructure: something that runs quietly in the background, like electricity or connectivity, rather than a product you point at and admire.
When a capability becomes infrastructure, the advantage moves off the technology and onto what you have built to run it on. This is the part the marketing inverts. The model is becoming the easy part. The hard part—the part that actually determines whether any of it produces value—is the quality of the operation the intelligence has to work inside. The wider market is learning this expensively right now: organizations are discovering you cannot simply bolt AI onto a disconnected business and expect it to work, and that if you treat AI as an add-on to a weak foundation, the foundation stops being neutral and becomes the impediment.
This is a pattern technology markets have followed repeatedly. As a capability becomes widely available, competitive advantage shifts away from access to the technology itself and toward how effectively an organization applies it. AI appears to be following
the same path. The question is becoming less “Do you have AI?” and more “What does your business actually enable AI to do?”
Agentic AI Raises the Stakes
This is exactly where auto recycling needs to pay attention, because agentic AI does not tolerate a broken operation the way a chatbot does. A chatbot can sit on top of a messy business and still answer a question, because all it has to do is talk. An agent has to act—and to act, it has to move reliably through the steps of a real process and trust what it finds at each one. Where a business runs on scattered tools and undocumented habits, where the same task is done five different ways depending on who is working, an agent has nothing dependable to act upon. It fails not because the intelligence is weak but because the ground beneath it is.
Early enterprise data already suggests many agentic AI initiatives will fail—not because the models fail, but because organizations struggle to operationalize them on top of the businesses they actually have. Intelligence works. The operation underneath does not support it. That is the pattern auto recycling is on track to repeat, at the exact moment the technology is getting powerful enough to matter.
AI is An Accelerant, Not a Strategy
So where does this leave AI? As an accelerant, not a strategy. Point it at a sound, well-run process and it makes that process faster and lighter. Point it at a broken one and it accelerates the breakage—the same flawed process at higher speed, with a more sophisticated interface bolted to the front. Putting AI on top of a broken workflow is like putting a turbocharger on an engine with missing parts. Technology is not the bottleneck. The underlying system is.
A Note on What to Buy, and When
Because capability is commoditizing so fast, recyclers should be cautious about long-term decisions made on short-term features. Chat, chatbots, drafting, summarizing, basic automation—these are becoming standard functionality, not durable advantages, and much of what commands a premium today will be ordinary within 12 to 24 months. The wiser place to put effort is the operation itself: how work flows, how consistent it is, how much of it depends on one person remembering to do something. Those improvements hold their value regardless of which model dominates next year, because a sound operation is what every future system—agentic or otherwise—will require in order to do anything at all.
Where the Advantage Actually Goes
The limiting factor for AI in this industry over the next decade will not be model capability. It will be operational readiness. As AI moves from answering questions to executing increasingly complex processes, the businesses that benefit disproportionately will be the ones whose operations were sound enough to support it. The recyclers who win are unlikely to be the ones who bought the most AI. They will be the ones who removed the most friction from how they work, so that when intelligence became infrastructure, they had something real for it to run on.
Auto recycling does not need more AI buzzwords. It needs better operational foundations. The unglamorous work—tightening how the business runs, making it consistent, letting people spend their time on the judgment only people can provide—is what turns the next generation of AI from an expensive disappointment into something that actually delivers.
That work makes you a stronger business today, on your own terms, with or without AI. And it happens to be the only thing that makes you ready for what comes next. The future belongs to the businesses that can move information into decisions and decisions into action—which has surprisingly little to do with AI, and almost everything to do with the work we choose to do before it arrives.
Lisa Samuel is the founder of PayBuddy and Sabhi. She has spent more than two decades working at the intersection of payments, commerce, risk, and technology, helping businesses navigate operational and transactional complexity..
