AI has successfully made a huge contribution to the amount of software being produced, but there’s a strange disconnect – an increasing percentage of software that is never deployed or used. The question I want to investigate in this blog is: why is there more software being produced using AI than ever before, but not as much as we would expect, given the need? The reason for looking at this is because there seems to be a disconnect between the software produced by AI and the software that is actually being adopted, so I wanted to see what that tells us about the real bottlenecks in building useful technology.
Introduction
This is a weird situation in the tech world right now. AI can create working code much quicker than any human development team in history; complete applications, scripts, and features can be developed in mere minutes versus weeks. Software output has gone through tremendous growth by all measures. But there are more and more of those lines of code that exist but never get used: half-finished internal tools, abandoned prototypes, features that were developed but never released.
With AI producing more software than ever, why is not more of it in use?
For years, this has been the puzzle we knew we had to resolve: the ‘big bottleneck’ between our ideas and our working product- writing code was the difficult, expensive part. AI has filled in a lot of the missing puzzle pieces. However, coding is not all there is.
Just because the code seems to emerge from the code of others doesn’t mean it’s been decided what to build, that it actually solves a real problem, that it has been integrated into existing systems, that it’s been tested correctly, and that a team or customers have adopted it. The distance between the “code exists” and “code is useful” worlds has grown even wider in recent years, as they used to be constrained by code generation.
Counting without Quality Control
If it is easy and inexpensive to create code, then there’s less reason to be picky about what is built. There is more experimentation by teams and individuals, sometimes well and good, but it also means that more ideas that weren’t really that good are turned into working prototypes without having a clear pathway for real use.
With no natural limits on developers’ time and resources that once existed, there is plenty of software created through the power of artificial intelligence, but the business case for it has been lacking.
The Trust Gap
Even if AI software performs technically, businesses do not always trust the software to be fully used in production. Security, maintainability, and performance in real-world application are all concerns that leave a natural pause in the thought, particularly if it’s customer-facing or business-critical. That hesitancy, even if the software is viable, hampers adoption.
Manual integration is still possible.
The most useful softwares don’t stand alone; it must integrate with existing databases, workflows, and systems that were developed years ago, and in many cases not created using A, at all. While AI can generate the new piece rapidly, it’s not enough, and human coordination, testing, and buy-in are necessary to stitch it into an organization’s current technology stack, workflows, and approval processes. That’s where the projects created by AI typically get stuck.
Reducing the Gap between Output and Adoption
The companies that are truly benefiting from using AI to aid in development have some common practices:
Beginning with a problem, rather than a capability. Constructing due to necessity, not because it is convenient.
Reviewing before deploying. Using AI-generated content as a rough draft, not a final product.
Engaging the end users right away. The adoption of a tool is greatly enhanced if it involves the end user, rather than being presented to the end user after its completion.
Investing in integration, not in generation. Making time for not-so-glamorous integration of new software into old systems.
corper mattis, pulvinar dapibus leo.
The Bigger Picture
AI is not making software development easy; it’s making one part of it easy. Those that are always difficult- judgment, prioritization, trust, integration- are still very human issues, and they are the true cause of many AI-powered software products being abandoned. Organizations that are moving forward are not the ones creating the most code; they are the ones bridging the gap between what gets built and what actually gets used.
FAQs
1. Why is the amount of software being created by AI greater than what is actually being used?
Writing code is not the only limiting factor because what to write, how to test the code, and how to integrate it into existing systems still needs human judgment and time that can’t be accelerated by AI.
2. What is the relationship between the amount of AI-generated code and higher software quality?
Not necessarily. More code can be generated faster, allowing for more experimentation, but not if there’s no clear problem to solve; a lot of that code is not going to be adopted.
3. What are the reasons why companies are reluctant to adopt AI software during the production process?
Security, reliability, and maintainability concerns are also a worry; teams are reluctant to use AI-generated code without review for anything that is customer-facing or business-critical.
4. Is it a technical or organizational problem?
Mostly organizational. Technical capacity has grown significantly, but uptake is dependent on integration, trust, and ‘buy-in’ all human and process-related.
5. How to prevent building an AI software that will never be used in a business?
Keep the problem statement simple and direct, build the problem with the end-users from the beginning, and consider the result of AI a draft to be reviewed, not a ready-to-ship product.