Anthropic CEO Dario Amodei recently predicted that AI will write 90% of code within six months and dominate nearly all coding within a year. While many worry this signals the replacement of developers, they may be missing the bigger picture. AI isn’t just automating work, it’s reshaping roles and unlocking new opportunities.

At Salesforce, we see something more nuanced‌ — ‌and far more promising. Developers aren’t being replaced wholesale. They are moving up the stack.

The conversation has been focused too narrowly on what developers will lose. It’s time to widen the lens to what they’ll gain. That’s because as AI agents become more capable, developers will be able to take on broader, more strategic responsibilities. Their roles will evolve to focus more on system design, orchestration, and long-term results.

Developers are feeling the vibe

The shift is already underway. One early sign? “Vibe coding,” a term coined by OpenAI co-founder Andrej Karpathy, is quickly gaining traction. Think of it as coding by intent‌ — ‌letting AI take a first pass at writing code based on natural language instructions or rough ideas, like a writer’s first stream-of-consciousness draft to get words on the page. Developers still shape the outcome, but the process is more collaborative. The AI sets the rhythm; the developer fine-tunes the melody.

This is just the beginning. What we’re seeing now offers only a glimpse of a much larger shift. As AI takes on repetitive aspects of programming — such as boilerplate code, test generation, and documentation — developers can turn to higher-order work like designing systems, solving novel problems, and shaping overall strategy. 

Maham Hassan, a Dubai-based Salesforce Architect at Cloud-1, a managed IT service provider, recently told us AI is already elevating her work.

AI agents are shifting my role from being purely technical into a more strategic side of the business.

Maham Hassan, Salesforce Architect at Cloud-1

“AI agents are shifting my role from being purely technical into a more strategic side of the business,” she said. “Instead of spending time on repetitive code analysis, code reviews, or ensuring we have a scale solution and architecture in place, I can focus on designing more scalable architectures, optimizing business processes and driving innovation.”

In this new reality, developers will take on a supervisory role, guiding agents, refining their outputs, and ensuring alignment with larger system goals. Success here requires a shift from syntax to systems thinking, context management, and long-term planning.

Tools for transition

Salesforce is already preparing for this future. We’re actively building tools that support this shift‌ — ‌proprietary tools like CodeGenie that we use ourselves to boost developer productivity. Powered by Salesforce’s own model, CodeGen, which was developed by its AI Research Group, CodeGenie has handled more than 7 million lines of code, answered 500,000 developer questions, and saved at least 30,000 hours per month — reducing labor costs and making developers more efficient.

For external customers, we’ve also added development tools to Agentforce, our digital labor platform, to help customers ‌experience similar benefits. They include:

• Agentforce for Developers: Helps automate routine coding tasks‌ like writing new code, explaining existing logic, and generating test cases‌.
• Agent Builder: Makes it easier for both developers and business users to create, customize, and deploy agents using low-code tools and AI assistance.
• Agentforce Testing Center: Provides a secure environment to simulate agent behavior, test performance, and refine decision-making.
• Agentforce Developer Edition: Opens the door for advanced users to build deeply integrated, highly customized agents by combining AI-generated code with their own expertise.

These agentic tools differ from traditional AI tools and copilots, which provide lightweight assistance by suggesting snippets or filling in boilerplate code. That can be helpful, but it’s not transformational.

Agentic systems go further. They understand intent, take action autonomously, and deliver meaningful results. For example, instead of manually programming a new component to enable capabilities for a given program, a developer can simply prompt, “create a new component that will accept these parameters, call this API, and return a message if it was successful or not.” The agent will then determine any libraries it needs to access, write the appropriate classes, test the component, and deliver it back to the developer. Like any effective manager, the developer reviews the output and adjusts it to improve the accuracy and reliability of the code.

Succeeding with this new model requires the right tools — and Salesforce has been building and delivering them through Agentforce.

What developers should do now 

We know that not all developers feel ready. Our latest State of IT survey of software development leaders found that more than 80% of developers believe AI will soon be a fundamental job requirement — but more than half say they still lack the necessary skills.

So, what should developers do about it? 

First, recognize that we’re in a new era, and software development will never be the same. Fully embracing this shift is essential for thriving in the future of the industry.

Next, developers should do their utmost to sharpen their AI knowledge. Hassan told us AI literacy is “the No. 1 skill gaining traction right now,” and we agree. But not in a traditional building and coding sense. Rather, it means “understanding and using AI tools to drive business outcomes,” Hassan said. Developers should also become familiar with prompt engineering, vibe coding, context management, and iterative design. Developers will need to think less about isolated functions and more about how to define tasks, guide intelligent systems, and continuously refine outputs. It’s a move toward systems design, product thinking, and long-term architectural planning.

Finally, developers should become experts at managing AI agents. That suggests knowing how to guide, instruct, monitor, and evaluate their work with a critical eye to ensure high-quality results and minimize mistakes.

Fortunately, developers don’t need to take this journey alone. Salesforce’s Agentblazer community brings developers together to learn, experiment, and share best practices. Whether someone is just beginning to explore AI agents or looking to advance their skills, access to a peer network can make all the difference.

A new chapter

To be clear: developer roles will change. But just as cloud computing didn’t end IT careers, and automation didn’t eliminate system administrators, agentic AI won’t make developers obsolete.

Instead, it will redefine the job. Developers are trading in their IDEs for control panels, shifting from writing every line to directing the overall flow — and having the expertise to step back in when needed. Early adopters are already stepping into these roles, setting standards, managing teams of agents, and solving higher-order problems at scale.

This is a promotion, not a pink slip.

The call to action

As software development evolves, fresh opportunities are opening up for developers ready to adapt. Creativity, strategy, and leadership define this next chapter. The path forward is clear: lean into the shift, build new skills, and step into the role of AI agent manager. Those who thrive will be the curious, the committed, and those who treat AI as a force multiplier.

For CIOs and IT leaders, this transition demands more than just new tools. It takes investment in training, support, and a culture that encourages experimentation. Teams need both the resources and the confidence to lead in the agentic era.

The developers who thrive will be the curious, the committed, and the ones who view AI not as a replacement, but as a creative partner.

Learn more: 

• Read how CodeGenie is enhancing developer productivity
• Learn how AgentExchange empowers developers, partners, and the Agentblazer community to build and monetize agentic AI components
• Find out why AI models are a dime a dozen (it’s the platform that matters)
• Discover why you’re probably doing enterprise AI wrong

Blog Article: Here