Sam Sammane is the CEO of TheoSym.
In the midst of all the GPT-5 hype, the release left many people puzzled. We were promised something close to magic. A move toward artificial general intelligence (AGI). What we actually got instead was a faster, bigger, more sophisticated version of the same streaming algorithm. Still hallucinating. Still overenthusiastic. And still trying too hard to impress.
Not that GPT-5 isn’t incredible. It is. As a piece of software, it’s a marvel. But we have to stop pretending it’s something else. These large language models are fundamentally predictive text engines trying to guess the next word in a sentence. And because they’re trained on the whole internet—a vast repository of brilliance and crap—you’re not getting a Ph.D. expert in your niche. You’re getting all of the “experts” talking at once, and that tends to create more noise than clarity.
That’s why I believe the immediate future of AI has less to do with chasing the dream of AGI and more to do with creating tailored AI for businesses.
The Problem With Generic AI
AI, as it’s built now, does not have thoughts. It makes informed guesses. The transformers underlying today’s models are merely data structures weighing words for significance. Pre-training doesn’t make them smart; it’s simply a human-guided process of assigning value to tokens.
That’s why hallucinations are inevitable. Sometimes the model is underfit; it doesn’t have enough data. Other times it’s overfit and has too much, and it starts spewing out random connections. Either way, the system isn’t thinking or reasoning. It’s matching patterns.
On top of that, generic AIs are intentionally designed to be positive. Whatever idea you feed them, they’re trained to tell you it’s a “great idea.” That might make people feel good, but it also means businesses can potentially make poor choices based on a machine designed to flatter.
If you’re relying on these generic models in isolation, you’re making a fundamental mistake. You’re pulling from the entire internet when what you really need is AI based on your own context, your own data, your own expertise.
The Evolution Of AI Eras
We’ve already seen several eras of AI come and go. First was the chatbot era with simple one-question, one-answer systems. There was the prompt engineering era, where people discovered that careful phrasing could coax better output from models. After that, we entered the RAG (retrieval-augmented generation) era, where grounding answers in a limited set of documents provided enhanced reliability.
And we’ve just entered the era I call context engineering.
Context engineering goes far beyond prompting. It’s not about one question and one answer. It’s about multistep reasoning, dynamic prompts with variables and guiding the AI through a process designed for your business. Think of it as programming the AI to think along a path you’ve designed, step by step.
This shift changes everything. It’s no longer enough to rely on a clever prompt. Businesses need to design systems where the AI works inside a defined context, using domain-specific knowledge that eliminates irrelevant noise. That’s how you get meaningful results.
Why Personalized AI Is The Only Way Forward
The one-size-fits-all AI is a myth.
Every company needs its own AI, customized for its vertical and grounded in its knowledge base. Otherwise, you’re just talking to the collective noise of the internet. Personalized AI empowers your people, making them the best version of themselves. It doesn’t replace human creativity, empathy or judgment; it amplifies it.
And the good news is, this is more affordable than ever. What would have cost millions just a few years ago can now be built for tens of thousands, sometimes less. With open-source models today and low-cost infrastructure, a company can spin up its own AI assistant tailored to its exact needs.
When I run AI on my own servers, the economics are obvious. Rather than paying $10 to $15 per interaction in the cloud, I pay virtually zero. Sure, the response takes 15 seconds instead of five, but for a business environment, that’s a trade-off I’m willing to accept.
It’s for this reason that I believe we’re on the verge of a new desktop era of AI. Just as we once installed massive databases on our own machines, we’ll soon be running personalized AI systems locally, in a powerful, affordable and private manner.
Hybrid Systems And Context Engineering In Practice
So, how does it work in practice?
It entails the best of both worlds, with structured databases alongside unstructured data search. People experimenting with vector databases and chunking are, in many ways, rediscovering what traditional databases did. The real opportunity lies in hybrid systems where AI can query your tables, process your unstructured data and combine both into coherent outputs.
It also means designing AI processes that mimic real business workflows and processes. Context engineering is not a single prompt. It’s a series of steps, with dynamic variables plugged in as you proceed. Instead of entering one question and hoping for the best, you guide the AI through 20 steps that simulate the way your business actually operates.
The result isn’t just better answers. It’s an AI that will be a part of your organization, working within your context, not outside of it.
Stop Waiting For AGI And Start Building Now
Many leaders are paralyzed by the idea of AGI, waiting for some mythical future when machines become truly intelligent. Let me be straight with you: We have no idea how to build AGI. Making models bigger and throwing more money at them will not suddenly create intelligence.
In the meantime, while others hang around waiting for the miracle to happen, the smart companies will act. They’ll develop their own proprietary AI, specific to their company, and stealthily take away market share from confused competitors who are still waiting for the fairy tale.
The real breakthrough will come from superintelligent humans making savvy use of AI, not from a superintelligent machine. That leveraging of human expertise with tailored AI is where competitive advantage really lies.
Build Your Own AI Or Fall Behind
GPT-5 and its peers are impressive, but they aren’t enough. Businesses don’t need the whole internet. They need their own AI that is personalized, contextual and grounded in their domain.
The cost barrier has fallen. The technology is available. The only real question is whether leaders will seize the moment or keep waiting for an illusion that may never arrive.
If you want to truly update your business for the future, stop relying on one-size-fits-all AI. Start building your own.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
