Is the answer to everything AI?
It seems that whatever the question in 2026, the answer is AI, usually AI agents, and everyone’s solution is the best. However, is that really so? Or are we experiencing Hollywood style illusions? And anyway, do we really need AI, AI agents or an agentic operating system, and what does it all mean?
Let’s start with some context. Most blogs start with a definition of artificial intelligence (AI) along the lines of ‘a computer system that performs tasks that simulate human intelligence’. Then they discuss how computers operate differently to the human brain. The critical aspect is that the computer gives you an outcome you would have got, just in a different way.
Hollywood created AI perceptions
I want to explore two other areas: the marketing aspect of AI, and what we mean by agents. AI is a term the tech industry has latched onto. It strikes a chord with us from our Hollywood created perceptions of AI as a super advanced concept. AI is a deep rooted brand, originating in 1968 with HAL in Kubrick’s movie 2001. Through Star Trek, Star Wars, Terminator, WALL-E, Ex Machina, Iron Man and more, countless images form our perceptions of AI. Perceptions that are generally cutting edge, futuristic and advanced. Perceptions that differ depending on your generation. Hollywood has formed an AI brand perception for all of us. Often this is subconscious, generally positive, perception that AI can improve our work.
What is being called AI?
Because AI has strong positive associations, it has become a ubiquitous label. Silicon Valley, just north of Hollywood, has now wrapped many solutions into AI. Multiple different technologies that we can term traditional automation have been redefined to be part of AI. They include models that have run quietly for a decade or chatbots with a large language model (LLM) stapled on. One sure sign of an illusion is the supplier vendor who states they have had an AI solution for 20 years.
| Type | Abbr. | Overview | Pioneers |
| Robotic process automation | RPA | carry out high-volume, rules-based tasks across systems quickly, without manual input. | Blue Prism, UiPath, 2003 |
| Natural language processing | NLP | helps computers understand and make sense of human language | IBM Watson, Siri 2011 |
| Machine Learning | ML | learns from past trends to spot patterns and make predictions about what’s likely to happen next | IBM Watson, 2011 |
| Generative AI | Gen AI, LLM | type of AI that can generate output. A large language model (LLM) is trained, through deep learning, on a large amount of text to understand and create new, original content | ChatGPT, 2022 |
| Agentic AI | systems that can autonomously plan, take action, and adapt based on goals or changing conditions without needing step-by-step instructions for every task. | GPT-4, Claude, Gemini, 2024 |
What most of us actually experienced as new, was generative AI introduced by ChatGPT in 2022. What we now see the real value in is true AI agents. Today, you can hardly have failed to notice that not only is every solution AI, it is agentic AI, and probably AI native. Given the ubiquitous labelling and attractive connotations, agentic AI deserves tighter articulation.
So, what counts as an agent in 2026?
Leading IT research firm Gartner assessed that of the thousands of vendors claiming to offer AI agents, only around 130 offered the real thing. For the rest, Gartner politely calls ‘agent washing’. Most products marketed as agents today fall into the bottom two tiers. Now, not only do we see traditional automation (ML, RPA, NLP) rebranded as AI, we now see chatbots and agent assistants with no real autonomy rebranded as AI agents.
To spot a real AI agent, consider these 3 tests in the context of a request to find PO123:
- A real agent does not just find the requested PO123. It finds there is an invoice with overdue payment that needs action
- The agent puts the payment for the invoice into the next payment run. It advises the payment approver and the vendor of the expected payment date
- The agent gets it all done from one request. It uses the context you have provided, no ongoing chat or next steps
Enterprise Agentic Operating System
For the Enterprise, even this test is not enough. Yes, that gives you clever AI agents that any individual with a Claude Code licence can use for significant personal advantage. However, that is not going to move the dial for Operations. For the Enterprise, we want an agentic operating system. A system where we can orchestrate our procurement workflows with AI agents. Where Agents complete key activities within governed guardrails. Initially that will focus on efficiency gains in sourcing, contract reviews and supplier due diligence. Over time it may expand to strategic value creation and whole-of-life cost optimisation for the enterprise.
When selecting solutions, understand that an Enterprise agentic operating system differs from traditional automation, or a single AI agent. Those significant differences do matter and I will give them their own post later. For now, the AI agent test is simple: does it perceive, decide and act across your systems, within guardrails you govern? If not, you are looking at last decade’s automation in this year’s wrapping.
It is critical that we understand what we want, and how to evaluate the Hollywood illusions of smoke and mirrors in the marketplace.
How ready is your procurement function for AI, honestly? Take PRAIRA, our free 8-minute readiness assessment at spvalue.com.au/praira, and find the constraint setting your ceiling. Or contact me at SPV if you need support.
Material created by Paul Digweed with support by AI team member.
References:
Gartner Predicts 2025: Procurement Addresses Data Challenges and Embraces Rapid Change
Gartner estimates only about 130 of the thousands of agentic AI vendors are real.