Turn Any LLM into a Powerful Assistant (All About Tools) (bittere.substack.com)

The post argues the crucial reason modern LLMs feel so much smarter than early chatbots like Eliza isn’t just size or modeling advances but the ecosystem of “tools” they can call. Tools — e.g., web browsers, file readers, calendar create/search, or codebase search functions — provide the necessary external context and action capabilities that turn a passive text model into an interactive assistant. With the right toolset, a model can fetch live data, modify user resources, or run domain-specific actions, making it useful for real-world workflows rather than only generating static text. Technically the approach is simple and powerful: expose each tool with a name, inform the model those tools exist, have the model output a wrapped tool call when it wants to use one, then parse and execute those calls externally. The model never needs to know implementation details — only that invoking get_football_score returns the current score — which makes the pattern modular and broadly applicable. The author emphasizes that “context is king”: tools supply the context LLMs need, and the next installment will show a browser-extension implementation that can give almost any LLM tool access to the web and user data. This pattern has major implications for building agentic assistants and composable, secure integrations across AI systems.