When Anthropic CEO Dario Amodei declared that AI would write 90% of code inside six months, the coding world braced for mass extinction. However inside Salesforce, a distinct actuality has already taken form.
“About 20% of all APEX code written within the final 30 days got here from Agentforce,” Jayesh Govindarajan, Senior Vice President of Salesforce AI, advised me throughout a current interview. His crew tracks not simply code generated, however code really deployed into manufacturing. The numbers reveal an acceleration that’s inconceivable to disregard: 35,000 lively month-to-month customers, 10 million strains of accepted code, and inner instruments saving 30,000 developer hours each month.
But Salesforce’s builders aren’t disappearing. They’re evolving.
“The overwhelming majority of growth — at the very least what I name the primary draft of code — will probably be written by AI,” Govindarajan acknowledged. “However what builders do with that first draft has essentially modified.”
From strains of code to strategic management: How builders have gotten know-how pilots
Software program engineering has all the time blended creativity with tedium. Now AI handles the latter, pushing builders towards the previous.
“You progress from a purely technical position to a extra strategic one,” Govindarajan defined. “Not simply ‘I’ve one thing to construct, so I’ll construct it,’ however ‘What ought to we construct? What does the shopper really need?’”
This shift mirrors different technological disruptions. When calculators changed handbook computation, mathematicians didn’t vanish — they tackled extra complicated issues. When digital cameras killed darkrooms, images expanded relatively than contracted.
Salesforce believes code works the identical manner. As AI slashes the price of software program creation, builders achieve what they’ve all the time lacked: time.
“If making a working prototype as soon as took weeks, now it takes hours,” Govindarajan stated. “As a substitute of exhibiting clients a doc describing what you would possibly construct, you merely hand them working software program. Then you definately iterate primarily based on their response.”
‘Vibe coding’ is right here: Why software program engineers are actually orchestrating AI relatively than typing each command
Coders have begun adopting what’s referred to as “vibe coding” — a time period coined by OpenAI co-founder Andrej Karpathy. The follow includes giving AI high-level instructions relatively than exact directions, then refining what it produces.
There is a new form of coding I name “vibe coding”, the place you absolutely give in to the vibes, embrace exponentials, and neglect that the code even exists. It is attainable as a result of the LLMs (e.g. Cursor Composer w Sonnet) are getting too good. Additionally I simply discuss to Composer with SuperWhisper…
— Andrej Karpathy (@karpathy) February 2, 2025
“You simply give it a kind of high-level route and let the AI use its creativity to generate a primary draft,” Govindarajan stated. “It received’t work precisely as you need, nevertheless it offers you one thing to play with. You refine elements of it by saying, ‘This seems good, do extra of this,’ or ‘These buttons are janky, I don’t want them.’”
He compares the method to musical collaboration: “The AI units the rhythm whereas the developer fine-tunes the melody.”
Whereas AI excels at producing easy enterprise purposes, Govindarajan admits it has limits. “Are you going to construct the next-generation database with vibe coding? Unlikely. However might you construct a very cool UI that makes database calls and creates a implausible enterprise utility? Completely.”
The brand new high quality crucial: Why testing methods should evolve as AI generates extra manufacturing code
AI doesn’t simply write code in a different way — it requires totally different high quality management. Salesforce developed its Agentforce Testing Heart after discovering that machine-generated code demanded new verification approaches.
“These are stochastic programs,” Govindarajan defined. “Even with very excessive accuracy, eventualities exist the place they could fail. Possibly it fails at step 3, or step 4, or step 17 out of 17 steps it’s performing. With out correct testing instruments, you received’t know.”
The non-deterministic nature of AI outputs means builders should grow to be consultants at boundary testing and guardrail setting. They should know not simply learn how to write code, however learn how to consider it.
Past code era: How AI is compressing all the software program growth lifecycle
The transformation extends past preliminary coding to embody the complete software program lifecycle.
“Within the construct part, instruments perceive current code and lengthen it intelligently, which accelerates every thing,” Govindarajan stated. “Then comes testing—producing regression assessments, creating check instances for brand new code—all of which AI can deal with.”
This complete automation creates what Govindarajan calls “a considerably tighter loop” between concept and implementation. The sooner builders can check and refine, the extra formidable they’ll grow to be.
Algorithmic pondering nonetheless issues: Why laptop science fundamentals stay important within the AI period
Govindarajan steadily fields anxious questions on software program engineering’s future.
“I get requested continuously whether or not folks ought to nonetheless research laptop science,” he stated. “The reply is completely sure, as a result of algorithmic pondering stays important. Breaking down huge issues into manageable items, understanding what software program can remedy which issues, modeling person wants—these expertise grow to be extra beneficial, not much less.”
What modifications is how these expertise manifest. As a substitute of typing out every resolution character by character, builders information AI instruments towards optimum outcomes. The human offers judgment; the machine offers pace.
“You continue to want good instinct to provide the best directions and consider the output,” Govindarajan emphasised. “It takes real style to have a look at what AI produces and acknowledge what works and what doesn’t.”
Strategic elevation: How builders have gotten enterprise companions relatively than technical implementers
As coding itself turns into commoditized, developer roles join extra on to enterprise technique.
“Builders are taking supervisory roles, guiding brokers doing work on their behalf,” Govindarajan defined. “However they continue to be liable for what will get deployed. The buck nonetheless stops with them.”
This elevation locations builders nearer to decision-makers and farther from implementation particulars—a promotion relatively than an elimination.
Salesforce helps this transition with instruments designed for every stage: Agentforce for Builders handles code era, Agent Builder permits customization, and Agentforce Testing Heart ensures reliability. Collectively, they kind a platform for builders to develop into these expanded roles.
The corporate’s imaginative and prescient presents a stark distinction to the “builders are doomed” narrative. Quite than coding themselves into obsolescence, software program engineers who adapt might discover themselves extra important than ever.
In a subject the place reinvention is routine, AI represents probably the most highly effective compiler but—remodeling not simply how code is written, however who writes it and why. For builders prepared to improve their very own psychological fashions, the longer term seems much less like termination and extra like transcendence.
