AI: Are programmers cannibalizing each other?

According to McKinsey, up to $250 billion will be invested in generative AI (GenAI) to automate IT operations over the next three years. This mainly affects code generation, IT helpdesk and test automation. New areas of application for GenAI are aimed at creating agent-based workflows that replace other applications: “The integration of GenAI into established products can improve their range of functions, for example by creating interfaces for voice input or new AI-based co-pilots to support existing workflows,” it says In the report,

Advertisement

McKinsey analysts see the greatest potential of GenAI in the development of new application software in the IT sector. For this purpose, models are created or adapted for specific applications or vertical workflows; using these models, code for more specific use cases should be automatically generated. In addition, new tools can be developed for company-specific workflows, for example the automation of certain back-office processes or the creation of knowledge agents to be able to access company information more specifically. In addition, new tools will be introduced to support the new AI application layer, for example for managing base models and platforms for AI development. Other areas include AI governance tools, data storage, networks, data visualization, AIOps, and automation.

According to McKinsey, programming is another major application area for GenAI. It can increase the productivity of software developers by up to 50 percent. This mainly relates to documentation, adaptation to new requirements, and refactoring. This will lead to more in-house development and less standard software being used. This trend will be reinforced by the fact that citizen developers will soon no longer use low-code/no-code, but will program only in natural language and GenAI. In the new AI-based, direct communication between users and AI data and models should also play an important role. MIT has recently introduced GenSQL for such queries. “SQL allows you to ask queries of databases in a high-level programming language, but we believe that in the future you will have to move from pure database queries to querying models and their data, which can only be done in natural language, writes MIT researcher Vikash Mansinghka in his publication.

iX Workshop: OWASP® Top 10 – Understanding Security Risks to Web Applications

McKinsey is not alone in these views. But IDC thinks soGenAI-based tools will change the software world. An IDC report on the topic says, “GenAI will primarily automate software testing, significantly reducing manual effort and leading to better code quality.” In addition, their analysts also see a wave of AI-based modernization in IT services, for example helpdesk.

AI in itself is nothing new in IT

Enthusiastic forecasts are based, among other things, on the fact that AI has been supporting IT operations for a long time. Examples of this are HP’s Proactive Insights or Intel’s Data Center Manager (DCM), HPE is represented by Infosight and AppsRamp. What they all have in common is that AI is used to analyze a lot of operational data in real time to spot anomalies and automatically initiate countermeasures. HPE’s Rubis Central For example, HPE manages more than four million customer devices. AI can be used to determine what the best measures are for optimal network operation. This includes recommendations for network administrators. They can follow the recommendation or reject it. In many cases, a small checkbox will appear asking: “If the problem occurs again, will the system fix it automatically?”

Dell Technologies has that Apex AIOps Infrastructure Observability. It provides AI-powered insights, predictions and recommendations for IT optimization, cybersecurity and stability of the entire IT infrastructure. According to Dell, this results in numerous operational benefits: it takes only one-tenth the time to resolve infrastructure issues, one working day per week is saved on system administration, and it takes only three minutes for an automated cybersecurity scan of 1,000 systems. All of these products are based on classic AI and are being continuously developed independently of GenAI, so that IT administration continues to be greatly relieved in the future.

GenAI is currently hardly used

As far as the current status of GenAI in IT is concerned, its use is still relatively low. According to consulting firm Bain & Company Only 21 percent of companies use GenAI in the field of software development and 35 percent are experimenting in related development projects. It is not known how intensive and extensive the individual activities are. In IT operations, only 17 percent say they are already using GenAI and 28 percent are still in the project stage.

The error rate in these solutions is still quite high. “A coding agent can generate any code very quickly, but they cannot see what is wrong with it or how to test the code,” says Kanjun Qiu, CEO and founder of AI startup Imbue, which specializes in such GenAI solutions. Nvidia researcher Jim Fan also has his concerns. “ChatGPT can write code, but it is not able to understand long content. A human developer can navigate dozens, if not hundreds, of lines of code — something that no coding agent can currently do,” he experiences. Infosys uses GenAI for faster coding, including GitHub Copilot and an internal coding assistant. But we don’t know exactly how big the benefits are. “Costs are coming down in some areas, but new additional costs are emerging in others. Proper deployment requires significant investments, for example in research and development,” says Sunil Seenan, head of data, analytics and AI at Infosys.

The Upwork Research Institute determined in a recent survey that the use of generative AI will mean additional effort for 77 percent of users. 47 percent are wondering how they should achieve the required increased productivity. McKinsey analysts also put their statements in perspective at the end of the report: “Of course, it is difficult to predict the full extent of the long-term consequences, especially given the unpredictable evolution of new technologies and applications and the countless ways they may ultimately have an impact,” they say there.

(For)