The rise of Large Language Models (LLMs) has ignited both excitement and skepticism, as people continue to explore their capabilities. One area where confusion persists is how these models handle seemingly simple tasks, like comparing numbers. For example, it's not uncommon to see posts highlighting how asking an LLM "Is 9.11 greater than 9.9?" can yield the wrong answer due to tokenization. Yet, rephrasing the same question into a sentence like "Which is bigger, nine point one one or nine point nine?" produces the correct result. This discrepancy is often cited as a limitation of AI, but that misses the bigger picture.
Focusing too much on edge cases like tokenization side effects distracts from the true potential of these tools. LLMs are not calculators or fixed-purpose software like spreadsheets—they represent a fundamentally new class of technology. Trying to force them into the mold of traditional tools or expect them to mimic human thought processes is a mistake. The real power of LLMs lies in their ability to revolutionize how we approach complex tasks, particularly in software development.
At voiceofreason.ai, where we create artificial intelligence tools to help people with dementia, brain injuries, or learning disabilities improve social interaction, we've had to rethink how we build software entirely. Traditional software development approaches—long cycles of coding, testing, and refining—simply don't make sense when working with LLMs. Instead, we've embraced a new paradigm where AI plays a central role in every phase of the process.
A New Approach to Software Development
The typical software development lifecycle involves painstakingly writing code, testing it, debugging, and repeating the cycle until perfection is reached. LLMs have flipped that model on its head. By leveraging their capabilities, we've been able to produce working prototypes in a fraction of the time it would traditionally take.
Our process is highly iterative and incremental. Rather than trying to perfect code from the start, we use AI to generate synthetic test data, build prototypes, and even evaluate and modify code as the tests run. The role of the software engineer has evolved from being a master coder to a systems designer. The LLM becomes the "master coder," rapidly generating code and refining it based on the system's needs. Guiding an AI in this capacity is fundamentally different from managing human engineers, but the advantages are immense.
LLMs allow us to work at an unprecedented pace. Within a month, we've gone from concepts to working prototypes that we can actively test and refine. The agility we’ve gained in development is invaluable, especially when building AI-powered tools to support people with cognitive challenges. This shift required us to abandon traditional software development processes entirely.
The Role of the Engineer as Systems Designer
So, what does this new approach mean for software engineers? The engineer's role shifts from writing and perfecting every line of code to orchestrating the overall system design, ensuring the components work harmoniously together. Engineers must learn how to guide the LLM effectively, focusing on high-level structure, architecture, and systems thinking.
Instead of handcrafting every function, the job becomes more about feeding the LLM the right prompts, test cases, and specifications to generate code that meets the project’s needs. This involves thinking of the problem in layers—guiding the AI on how the parts of the system should interact, understanding when to accept AI-generated code, and recognizing when refinement or human intervention is necessary.
The Advantages of Thinking Outside the Box
When engineers embrace this approach, the advantages are enormous:
- Speed: Prototypes and test environments can be created in hours or days, not weeks or months.
- Scalability: LLMs can handle repetitive, large-scale coding tasks, freeing engineers to focus on innovation.
- Flexibility: With LLMs, we can make changes and improvements to the system in real time, adapting to new insights from user testing without massive reworks.
- Collaboration: Systems engineers can focus on the bigger picture—integrating AI across different platforms or designing more intuitive, user-centered experiences.
The AI revolution in software development is just beginning. By moving away from the traditional coding mindset and embracing AI as a partner in the development process, we unlock levels of productivity and creativity that were previously unimaginable. At voiceofreason.ai, we’ve seen firsthand how transformative this new approach can be, and we're only scratching the surface of what's possible.
Conclusion
As we continue to experiment and refine how we use LLMs, one thing has become clear: these tools require a fresh mindset. They're not just faster coders—they're enablers of a new way to build software. The traditional approaches to software development, which worked for fixed-purpose software and even human coders, don't apply in this new AI-driven world.
In this new paradigm, the software engineer becomes a systems designer, orchestrating the work of AI to build, test, and refine products iteratively. Those who embrace this shift will see unprecedented speed and flexibility in their work. Those who don’t will be left behind, stuck trying to fit this revolutionary technology into old boxes.