Exploring the Causes of the Broad Disenchantment with Artificial Intelligence

The Reality of AI Development: Beyond the Hype and Propaganda

In today’s discourse surrounding artificial intelligence, it’s easy to become desensitized or disappointed. Much of the public perception is shaped by market-driven narratives, often propagated by entrepreneurs and tech enthusiasts more interested in selling solutions than delivering genuine innovation. The landscape is riddled with what could be best described as superficial hype—short-sighted “magic bullet” promises and quick-fix AI tools that promise much but deliver little.

Behind the scenes, however, a quieter revolution is taking place. Dedicated developers and researchers are leveraging AI to create custom automations tailored to specific tasks, streamlining workflows and enhancing productivity. Unlike one-size-fits-all solutions, these implementations are highly adaptable but usually not interchangeable, often requiring significant restructuring when adapted for different contexts.

One key insight is that the most significant progress won’t come from automating entire jobs outright. Instead, the future lies in automating the process of automation itself. Typically, a task involves 1 to 5 discrete steps, which may include loops, leveraging memory, and interacting with various APIs. This layered complexity suggests that, to truly harness AI’s potential, we need to focus on automating the orchestration of these steps.

This may sound simple on paper, but in practice, it involves managing a hierarchy of specialized agents. One agent might craft prompts, another designs the architecture—including memory integration—and yet another interacts with external APIs to execute tasks and process data. The key is orchestrating these components seamlessly.

Fortunately, much of this capability already exists. AI systems have been generating their own prompts for some time. Recent advancements, such as the 2023 paper on the MCP protocol (https://arxiv.org/abs/2310.08101), have laid the groundwork for more sophisticated integrations. The MCP protocol allows instructions to be embedded directly within API calls, streamlining communication between components.

Further, tools like AgentForge now support YAML-defined architectures, enabling large language models (LLMs) to autonomously build complex systems from scratch—sequencing prompts, managing memory dynamically, and interacting with APIs—all without requiring manual coding.

The horizon is clear: what remains is largely a matter of time. The essential, transformative workflows that can automate the process of automation are within reach. While the path isn’t without complexity, it holds the promise that many of the repetitive, labor-intensive tasks we face today could eventually be delegated or even fully automated—marking a pivotal shift in how we harness