The models are good enough. The context is what’s missing.

AI researcher Andrej Karpathy made a striking observation in his year-end review: we haven’t realized anywhere near 10% of the potential of current AI models. Not future models—the ones available today. That’s an enormous gap between what’s theoretically possible and what’s actually being captured in production systems.

The reason for that gap? Context.

We’ve spent the last couple of years being amazed by what AI can do in general. The demos, the benchmarks, the capabilities that seemed impossible just a few years ago. But there’s a growing recognition that general capability isn’t the same as specific usefulness, and that gap is where most enterprises are stuck right now. The models are good enough. The context is what’s missing.

General intelligence, specific problems

The large language models we have access to are genuinely impressive as general reasoning systems. They can analyze, synthesize, write, code, and work through complex problems across almost any domain. But that generality is also a limitation when you’re trying to solve specific business problems.

Take any business-critical process in your organization. General-purpose AI has no idea about your specific requirements, your existing systems, your compliance obligations, or how your particular workflows actually function. It can help with isolated tasks if you frame them carefully, but it can’t operate within your business context autonomously.

Contextual AI is about closing that gap. It’s the application of AI within the specific data, workflows, systems, and constraints of a particular business context, so that it can operate meaningfully within it. Not a general assistant that you query with questions, but an embedded capability that participates in how work actually gets done.

The timing isn’t accidental

There’s a reason this is becoming the central question in 2026 rather than earlier. For the past few years, the focus has understandably been on model capabilities. Can AI do this task at all? Can it do it reliably? Can it handle edge cases? Those were the right questions when the technology was still proving itself.

But we’ve reached a point where the answer to most of those questions is yes, at least for a very wide range of business tasks. The models are capable enough. What they lack is the infrastructure that lets them operate within specific organizational contexts: access to the right data, integration with existing systems, understanding of particular workflows, appropriate governance and compliance controls.

Saanya Ojha wrote about this as scaffolding. The scaffolding is what turns general AI capability into specific business value. Without it, you get what she called pilot purgatory, where impressive technology never quite makes it into production at the scale where it would actually matter.

The work nobody talks about

When people discuss AI transforming business, they tend to focus on high-visibility applications. Customer-facing chatbots, content generation, data analysis and insights. Those are real use cases, but not where the biggest value sits.

The biggest opportunities are in the operational processes that consume enormous resources but rarely get attention because they’re a bit boring. Documentation workflows that eat hundreds of hours per project. Scheduling and coordination across large distributed workforces. Compliance tracking and reporting that requires constant manual effort. Asset management and maintenance logging that still runs on spreadsheets and paper forms.

This is the work that drains organizations from the inside. It’s expensive, it’s error-prone, it scales poorly, and it’s almost entirely invisible to anyone outside the teams doing it. But it’s also work that AI can actually do well, if you can connect AI to the specific context where that work happens and support the humans that need to audit, verify and be ultimately accountable.

The reason these processes haven’t been automated already isn’t that automation was impossible. It’s that traditional software development to automate them was too slow and expensive to justify, especially for processes that seem mundane. What’s changed is that the path from identifying a process problem to having a working solution has compressed dramatically. But only if you can build in context.

What contextual AI looks like in practice

Let’s look at a concrete example, because abstractions only take you so far.

Baneservice works on railway infrastructure and has to meet BREEAM environmental certification requirements, which traditionally meant hundreds of hours of manual documentation per project with significant risk of errors or omissions. They built an AI-assisted system that handles most of that documentation automatically, with humans reviewing and approving rather than doing the data entry. AI generation, human verification. The key was that the AI understands their specific certification requirements, their project structures, their documentation standards. And it is integrated with their existing systems and data. It operates within their context rather than offering generic assistance.

This is not an experiment. It’s a production system running business-critical operations. What makes it work is that it’s contextual, built around specific organizational needs rather than adapted from general-purpose tools.

What I expect to see this year

If 2025 was defined by model capabilities, 2026 will be defined by contextual deployment. The organizations that capture real value from AI will be the ones that figure out how to embed it within their specific business context, with proper integration, appropriate governance, and genuine understanding of particular workflows and requirements.

Enterprise transformation has always been about the unglamorous work of actually changing how organizations operate. That’s where the real value gets created. The models are good enough. The context is what’s missing.

At Appfarm, we’re building the platform that makes contextual AI possible, backed by visual transparency and enterprise governance. A platform where organizations can see, understand, and control how AI transforms their operations. This isn’t just another AI assistant. It’s the infrastructure that turns AI capability into business transformation.

2026 is the year context catches up with capability.