Current LLMs are intelligent enough for many tasks but fail because they lack access to complete context—emails, Slack messages, past data. The next step is building products that ingest this real-world context, making it available for the model to act upon.

The new software paradigm, driven by generative AI, moves away from complex interfaces. Instead, applications are designed to understand a user's natural language intent, removing the friction of learning how to operate the software and shifting the burden of learning from the user to the system.

Making an API usable for an LLM is a novel design challenge, analogous to creating an ergonomic SDK for a human developer. It's not just about technical implementation; it requires a deep understanding of how the model "thinks," which is a difficult new research area.

The significant leap in LLMs isn't just better text generation, but their ability to autonomously execute complex, sequential tasks. This 'agentic behavior' allows them to handle multi-step processes like scientific validation workflows, a capability earlier models lacked, moving them beyond single-command execution.

The current limitation of LLMs is their stateless nature; they reset with each new chat. The next major advancement will be models that can learn from interactions and accumulate skills over time, evolving from a static tool into a continuously improving digital colleague.

A non-obvious aspect of LLM training is that to accurately predict text describing reality (e.g., a lab result), the AI must learn to model the underlying physics or logic. This inherently trains it to be more capable than the human who merely observed the result, forcing emergent intelligence.

Instead of writing static code, developers may soon define a desired outcome for an LLM. As models improve, they could automatically rewrite the underlying implementation to be more efficient, creating a codebase that "self-heals" and improves over time without direct human intervention.

For decades, the goal was a 'semantic web' with structured data for machines. Modern AI models achieve the same outcome by being so effective at understanding human-centric, unstructured web pages that they can extract meaning without needing special formatting. This is a major unlock for web automation.

The fact that LLMs, designed to predict the next word (a writing task), spontaneously exhibit reasoning abilities provides empirical evidence for the long-held belief that writing and thinking are intertwined. A machine built to write inadvertently learned to think.