Join our daily and weekly newsletters for the latest updates and the exclusive content on AI coverage. Learn more
When Anthropic CEO Dario Amodei said the AI write 90% of the code In six months, the world coding has prepared for mass extinction. But inside DirtyA different reality has already taken shape.
“About 20% of all Apex code Written in the last 30 days has come from AgentForce“Said Jayesh Govindarajan, Vice-President Director of Salesforce AI during a recent interview. His team follows not only the code generated, but the code has really been deployed in production. The figures reveal an acceleration which is impossible to ignore: 35,000 active monthly users, 10 million lines of accepted code and internal tools saving 30,000 hours of development each month.
However, the developers of Salesforce do not disappear. They evolve.
“The vast majority of development – at least what I call the first draft code – will be written by AI,” said Govindarajan. “But what developers do with this first project has fundamentally changed.”
From lines of code to strategic control: how developers become technological pilots
Software engineering has always mixed creativity with boredom. Now has managed the latter, pushing the developers to the first.
“You go from a purely technical role to a more strategic role,” said Govindarajan. “Not only” I have something to build, so I’m going to build it “, but” what should we build? What does the customer really want? »»
This change reflects other technological disturbances. When the calculators replaced manual calculation, mathematicians have not disappeared – they addressed more complex problems. When digital cameras killed black rooms, photography has developed rather than contracted.
Salesforce thinks that the code works in the same way. While AI reduces the cost of software creation, developers earn what they always missed: time.
“If the creation of a functional prototype has taken once for weeks, it now takes hours,” said Govindarajan. “Instead of showing customers a document describing what you might create, you simply give them work software. Then, you will be based on their reaction.”
The “atmosphere coding” is there: why software engineers now orchestrate AI rather than type each order
The coders began to adopt what is called “mood“- A term invented by the co-founder of Openai Andrej Karpathy. The practice consists in giving high -level instructions AI rather than precise instructions, then refining what it produces.
There is a new type of coding that I call “Vibe Coding”, where you fully give in to vibrations, kiss the exponentials and forget that the code even exists. It is possible because the LLM (for example, cursor composer, W Sonnet) become too good. I also speak to composer with Super-Whisper…
– Andrej Karpathy (@Karpathy) February 2, 2025
“You just give him a kind of high level direction and let AI use his creativity to generate a first project,” said Govindarajan. “It will not work exactly as you wish, but it gives you something to play with. You refine games by saying:” It looks good, do more of that “or” these buttons are Janky, I don’t need it “.” “”
He compares the process to musical collaboration: “AI establishes the rhythm while the developer refines the melody.”
While AI excels in generating simple commercial applications, Govindarajan admits that it has limits. “Are you going to build the new generation database with an ambient coding? It is unlikely. But could you create a really cool user interface that makes database calls and creates a fantastic commercial application?” Absolutely. “
The new quality imperative: why test strategies must evolve because AI generates more production code
AI does not only write the code differently – it requires different quality control. Salesforce has developed its Agentforce test center After discovering that the code generated by the machine required new verification approaches.
“These are stochastic systems,” said Govindarajan. “Even with great precision, scenarios exist where they could fail. Perhaps it fails in step 3, or step 4, or step 17 on 17 stages that it works. Without appropriate test tools, you will not know.”
The non-deterministic nature of IA outputs means that developers must become experts in boundary tests and the railing parameter. They need to know not only how to write code, but how to assess it.
Beyond the generation of code: how the AI compresses the entire life cycle of software development
The transformation extends beyond the initial coding to include the life cycle of the complete software.
“In the construction phase, the tools include the existing code and intelligently extend it, which accelerates everything,” said Govindarajan. “Then the tests come – generate regression tests, creating test cases for a new code, all which can manage.”
This complete automation creates what Govindarajan calls “a much tighter loop” between the idea and the implementation. Faster developers can test and refine, the more ambitious they can become.
Algorithmic thinking is always important: why the fundamentals of computer science remain essential in the AI era
Govindarajan often poses anxious questions about the future of software engineering.
“I am constantly asked if people should still study computer science,” he said. “The answer is absolutely yes, because algorithmic thought remains essential. Decompose major problems with manageable parts, understanding what software can solve what problems, model user needs – these skills become more precious, no less. ”
What changes is how these skills are manifested. Instead of typing each character’s character character, the developers guide the tools to optimal results. Human makes judgment; The machine offers a speed.
“You always need a good intuition to give the right instructions and assess the outing,” said Govindarajan. “It takes an authentic taste to look at what AI produces and recognize what works and what does not work.”
Strategic elevation: how developers become business partners rather than technical performers
As coding itself becomes a merchant, the roles of developers are more directly connected to the commercial strategy.
“The developers occupy supervision roles, guiding the agents working on their behalf,” said Govindarajan. “But they remain responsible for what is deployed. The male always stops with them.”
This elevation places developers closer to decision -makers and further from the details of the implementation – a promotion rather than an elimination.
Salesforce supports this transition with tools designed for each step: AgentForce for developers Gande generation of code, Active Customation Builder and AgentForce Testing Center guarantees reliability. Together, they form a platform for developers to develop in these widened roles.
The vision of the company presents a striking contrast with the account of “developers is condemned”. Rather than coding in obsolescence, software engineers that adapt can be more essential than ever.
In an area where reinvention is routine, AI represents the most powerful compiler to date – transforming not only how the code is written, but which writes it and why. For developers arranged to upgrade their own mental models, the future is less like termination and more transcendence.
