Copied from https://gettriptip.com/blog/honeymoon-travel-planner/
Moving Beyond the Honeymoon Phase
Martijn Veldkamp
“Strategic Technology Leader | Customer’s Virtual CTO | Salesforce Expert | Helping Businesses Drive Digital Transformation”
June 14, 2024
Lucidworks, in its 2024 Generative AI Global Benchmark Study released last Tuesday, declared that “The honeymoon phase of generative AI is over.” source
While business leaders remain optimistic about AI’s transformative potential, the initial excitement has evolved into a more cautious and much needed measured approach.
It is great that we start to see more research being done with very interesting result. And I love that Lucidworks research was across all of the world and not just the US tech sector.
ROI?
One issue highlighted is the lack of financial payoff from generative AI projects. According to the study, “Unfortunately, the financial benefits of implemented projects have been dismal,” with 42% of companies yet to see substantial benefits from their generative AI initiatives. The costs involved in strategically using generative AI can be significant, whether companies are hosting their own large language models or using commercial APIs.
Interestingly, I start to wonder what were your expectations? What kind of financial benefits were you looking for?
The survey reveals that qualitative applications—those generating text and providing narrow responses—have seen the most success, accounting for about a quarter of successful implementations. These include projects such as generating FAQs and providing HR support, which are relatively straightforward to implement.
This is also the case with Salesforce. Our AI is great for creating Knowledge Articles based on certain cases and what was needed to solve them. It is a real productivity boost and helps companies by making the first step easier.
In contrast, applications involving quantitative tasks—such as monitoring, predicting, analyzing, optimizing, and prioritizing—have faced more challenges, with less than 15% achieving successful implementation. These complex projects include optimizing search results, screening job applicants, and supporting financial result closures.
What I expected
A notable area of success for generative AI is in code generation. Code generation stands out as a prime use case, demonstrating the potential of AI copilots to significantly enhance developer productivity. By automating repetitive coding tasks and providing intelligent suggestions, AI can free developers to focus on more complex and creative aspects of their work. This increased efficiency can lead to faster development cycles, reduced time-to-market for new software, and overall improved productivity.
For example, AI-powered tools can assist in writing boilerplate code, debugging, and even suggesting improvements or optimizations. These capabilities allow developers to spend less time on routine tasks and more time on innovation and problem-solving. Furthermore, by reducing the cognitive load on developers, AI copilots can help maintain a higher quality of code and reduce the likelihood of errors.
My take on it
With this focus on code generation I think people are missing the point. Historically seen developing an application was 30% of the cost, the rest was maintenance and LCM (Lifecycle Management). With the move towards DevOps teams and having more ownership of the services that you create and host and maintain. Do you feel ownership of code that was generated with a GPT? Did that GPT really understand the full problem domain that you as a dev want to address?
In the past I worked with Uniface, a 4GL. You created the Data Model first and it generated the screens and buttons to manipulate records. The time it took to debug why a certain piece of code was generated in that place and not over here was not funny.
If I look at previous projects and see teams having trouble understanding and adapting each others code (NIH syndrom), this is an initial productivity boost, but the long term results may be that we slowly start to not own, support or even understand any of the code now in production.
What is your opinion? Are the ‘witte bruidsweken’ over?
