AI for code generation – now also in Salesforce

The use of Generative AI (GenKI) is already having a significant impact on software development. Many companies using these tools report significant increases in productivity. One According to a McKinsey study With the help of GenKI, software development can be accelerated by a factor of two. In particular, providers of application software and database systems have jumped on the bandwagon and provided appropriate support.

advertisement

While Oracle introduced SQL query generation using natural language input last fall, Salesforce followed that up with an extensive portfolio designed specifically for developers at the recent TrailblazerDX (Salesforce Developer Conference) conference. This is where voice input contributes to code generation Anypoint building codeThe Salesforce Flow workflow tool and the Apex programming language are used. According to the company, for example, business processes can be formulated in natural language, which are then fully automatically mapped into Salesforce Flow. It should also be possible to generate Apex code in the same way. Voice input is also supported by auto-completion, so the wording can be typed more quickly.

Generation is not limited to application code: associated test scenarios can also be generated automatically with the click of the mouse. They're usually not perfect right away, but are intended to serve as a starting point from which, according to Salesforce, any testing environment can be set up much more quickly than if developers started from scratch. “We're not interested in making developers and administrators redundant; instead, we want to relieve them and speed up their work,” Alice Steinglass, executive vice president and general manager of the Salesforce platform, emphasized in an interview. Ninth.

Bridging the gap between the IT world and the business world

In this context, Steinglass sees another advantage of generating code using natural language: “We know that there is still a gap between the business world and the IT world. With artificial intelligence, this gap can finally be bridged, because if I “can” formulate a workflow A business or business process works in natural language, then both sides understand it and the IT department can continue working directly with the generated code,” the Salesforce manager is convinced.

She illustrates this using two examples. The first involves mailing, where a series of data must be retrieved from a customer relationship management (CRM) system to then be used in an advertising form suitable for a highly personalized email. This approach requires a combination of business experience and IT/data knowledge. This means you need someone who understands the business and knows exactly what you want to achieve with this email. It then requires someone who knows the data and knows how to retrieve it. GenKI can serve as a bridge through which the two sides can reach a concrete agreement.

The second example includes formulas, some of which may be very complex. To that end, Salesforce has expanded its Einstein software — including Einstein 1 Studio, which is also introduced in TrailblazerDX and helps integrate AI into Salesforce applications — to include formula drafting in natural language. This means that employees from different specialized areas can “speak” the formulas by which new code is created. Developers then have to check this generated code and improve it if necessary, for example to better reflect exceptions. Here too, GenKI acts as a bridge to improve communication between business and development.

Silvio Saravez, Executive Vice President and Chief Artificial Intelligence Scientist at Salesforce, goes one step further: “Natural language provides an ideal opportunity to lower the inhibition threshold for using IT tools. This means that it is not just about programming, it is about every tool and every complex program – Everything can be used more easily when used by voice.”

An important step, but still far from perfect

But all is not well so far, on the contrary. Oracle admits that its SQL code generation accuracy is only 70 to 75 percent. Salesforce managers also stress that “developers should definitely look at generated code very carefully.” The extent to which code generation can be further improved is uncertain. However, experts believe that the high error rate will still exist because natural language is not clear enough – especially when it comes to nested queries with many exceptions.

So Salesforce doesn't focus so much on perfection in future development, but instead focuses on designing new roles. Saravese has his own prediction: “Developers will become designers. This means that AI will take over many manual tasks, while developers focus on tuning individual units and putting them together — just as a composer specifies his musical instrument work.”

However, this does not make programming unnecessary. Saravez even explicitly recommends that schools continue to teach coding. “Coding knowledge will remain essential in the future, because as long as you cannot trust AI, every automatically generated code will have to be checked manually – and this is only possible with in-depth knowledge,” he explains his assessment. Aside from that, he is convinced that learning a programming language is still the best training for logical thinking.

(a map)