Oh boy, it’s hard to overstate the change in the software ecosystem since my last post. While my statement:

machine learning is bringing a whole new approach to […] software creation.

is still accurate, the arrival of ChatGPT and LLMs into the general technology landscape has been nothing short of seismic.

The statement:

Software has always had a major funnel problem — too many ideas & problems for the team & budget capacity.

may have been undermined somewhat with all the chat-to-code tools, taking the GitHub Co-pilot assistant model to new levels of whole apps from a description.

Creating an app from a short requirements statement, may leapfrog over the previous pipeline concern. It may suffer many of the same rapid app dev and low-code downsides previously discussed. However, there’s scope for the problems to never appear for a large class of system.

If an application is generated from a human-language description, which now forms the source-code in effect, in combination with being able to read the existing data / code automatically, the whole edifice of software development and crafting maintainable applications may no longer really apply.

While prompt engineering may be currently over-hyped, the skill of driving an LLM (or a successor model type) to generate useable artifacts is surely a useful skill and one that may require a differentblend of attitudes and aptitudes from the traditional software engineer. The previously stated funnel problem, may have pulled a proverbial rabbit from the hat.

While the software sector is not completely on fire, AI is providing the biggest shakeup since the web. Interesting times.