This AI already writes 20% of the Salesforce code. This is why developers do not worry

When Anthrop CEO Dario Amodei declares that AI will Write 90% of the code Within six months, the encoding world has been preparing for mass extinction. But inside SalesforceA different reality has already been shaped.

“About 20% of all APEX code Written in the last 30 days came from Agentforce“Jaes Govindarajan, Senior Vice President of Salesforce AI, told me, during a recent interview. His team not only tracks a code generated, but the code is actually unfolded in production. The numbers reveal an acceleration, which is impossible to ignore: 35,000 active -uses, 10 million users, 10 million. a month.

Still, Salesforce developers don’t disappear. They develop.

“The bigger part of the development – at least what I call the first code project – will be written by AI,” Govindarajan admitted. “But what the developers do with this first project has changed dramatically.”

From code to strategic control: How developers turn into technological pilots

Software engineering has always mixed creativity with Tedium. Now AI is dealing with the latter by pushing the developers to the first.

“You are moving from a purely technical role to more strategic,” Govindarajan explained. “Not only” I have something to build, so I will build it “, but” What should we build? What does the client really want? “

This change reflects other technological interruptions. When the calculators replaced the manual calculations, the mathematicians did not disappear – they dealt with more complex problems. When digital cameras killed dark rooms, photography expanded and not concluded.

Salesforce believes the code works the same way. As AI reduces the cost of creating software, developers earn what they always miss: time.

“If the creation of a working prototype once took weeks, it takes hours now,” Govindarajan said. “Instead of showing customers a document describing what you can build, you just hand them through working software. Then you repeat based on their reaction.”

“Vibration Coding” is here: why software engineers already orchestrate AI rather than enter any command

Coders began to accept what is called “vibration“-Teminm created by Openai co-founder Andrei Carpathi. Practice involves giving high-level AI instructions, not precise instructions, then refining what it produces.

There is a new type of encoding that I call “vibration coding”, where you completely succumb to vibrations, embrace exponions, and forget that the code even exists. It is possible because the LLMS (eg composer of the cccore W Sonnet) is getting too good. Also I’m just talking to a composer with Superwhisper …

– Andrej Karpathy (@Karpathy) February 2, 2025

“You just give him some high -level direction and let AI use his creativity to generate first draft,” Govindarajan said. “It won’t work exactly the way you want, but it gives you something to play with. You refine parts of it, saying,” This looks good, do more than that “or” These buttons are janjes, I don’t need them. “

He compares the process to the musical collaboration: “Ai sets the rhythm while the developer refines the tune.”

While AI is distinguished by the generation of direct business applications, Govindarajan admits that there are restrictions. “Will you build the next -generation database with vibration coding? Unlikely. But could you build a really cool user interface that makes calls in the database and creating a fantastic business application?” Absolutely. “

The new quality imperative: Why test strategies should develop when AI generates more production code

AI not just writes code differently – it requires different quality control. Salesforce developed its Agentforce Testing Center Following the discovery that the code generated by the machine requires new check approaches.

“These are stochastic systems,” Govindarajan explained. “Even with very high accuracy, there are scenarios where they can fail. Maybe fails in step 3 or step 4 or step 17 of 17 steps it performs. Without the right testing tools, you won’t know.”

The indefinite nature of AI’s outputs means that developers must become experts in border testing and security setting. They need to know not only how to write code, but how to appreciate it.

Beyond Code Generation: How AI compresses the entire life cycle of software development

The transformation extends beyond the initial encoding to cover the full life cycle of the software.

“In the construction phase, the tools understand the existing code and expand it intelligently, which accelerates everything,” Govindarajan said. “Then comes testing – generating regression tests, creating tests for a new code – all of which AI can handle.”

This complete automation creates what Govindarajan calls a “significantly tougher cycle” between idea and implementation. The faster developers they can test and refine, the more ambitious they can become.

Algorithmic thinking still matters: why the basics of computer science remain essential in the AI ​​era

Govindarajan often raises alarming questions about the future of software engineering.

“I am constantly asked if people still have to study computer science,” he said. “The answer is absolutely yes, since algorithmic thinking remains significant. Breaking up big problems of manageable pieces, understanding what software can solve what problems, modeling the needs of the user -these skills become more valuable, not less.”

What is changing is how these skills are manifested. Instead of introducing any nature of a character solution, developers direct AI tools to optimal results. Man gives judgment; The machine provides speed.

“You still need a good intuition to give the right instructions and evaluate the production,” Govindarajan stressed. “It takes a real taste to look at what AI produces and to recognize what works and what doesn’t.”

Strategic Raising: How developers become business partners, not technical contractors

As the coding itself is coditized, the roles of the developers are more directly associated with the business strategy.

“The developers are taking supervisory roles by directing agents who do the job on their behalf,” Govindarajan explained. “But they remain responsible for what is unfolding. The buffer still stops with them.”

This elevation puts the developers closer to the decisions and further than the details of the implementation-promotion rather than the elimination.

Salesforce supports this transition with tools designed for each stage: Agentforce for developers processes code generation, agent creator allows personalization, and the AgentForce test center guarantees reliability. Together, they form a platform for developers to grow into these enlarged roles.

The company’s vision presents complete contrast to the story “The developers are doomed”. Instead of encoding in aging, software engineers who adapt may be more important than ever.

In a field where the rediscovery is routine, AI is the most powerful compiler so far – it trains not only how the code is written, but who writes it and why. For developers wishing to upgrade their own mental models, the future looks less like termination and more as transcendence.