For years, the simplest way for a technology company to build more software was to hire more engineers. More products, more customers and greater complexity meant bigger development teams. But US-based enterprise software company Salesforce is now testing whether that equation still holds in the age of AI.
It has deployed AI coding agents, including Anthropic’s Claude Code, across its engineering organisation to see how much more existing teams can deliver without a corresponding increase in headcount. The company operates major engineering, product development, and R&D hubs in Hyderabad, Bengaluru and other cities.
The early results are striking. Salesforce CEO Marc Benioff has said the company has not hired additional engineers in recent months because AI coding agents have boosted productivity. “We’re not hiring more engineers,” he said, adding that the company was mostly expanding in sales. Salesforce says work items completed by each developer rose 50.8% year-on-year in April 2026, while pull requests – the packages of code submitted for review – increased 79% per developer.
Benioff’s comments come against a broader shift in Big Tech, where companies are beginning to rethink the size and role of software teams as AI becomes more capable. In recent months, Meta and Amazon have both announced significant workforce reductions while stepping up their focus on AI.
Meta has cut thousands of jobs even as it pours billions into AI, while Amazon has also announced major cuts and continued investing heavily in AI. The trend does not mean engineers are simply being replaced, but it does suggest that companies are looking at AI as a way to get more software output from smaller or slower-growing teams.
An even broader measure of Salesforce’s output rose 151.3%. The company calls it “Effective Output”, a measure intended to capture the value of software delivered rather than simply counting lines of code. In one example, a migration involving 33 application programming interfaces (API) that Salesforce estimated would require 231 person-days was completed in 13 days – 18 times faster.
These are Salesforce’s own internal measurements, but they illustrate the bigger change underway: the promise of coding agents is not simply that developers can type faster, but that one developer can potentially deliver much more useful software.
The first generation of AI coding tools offered autocomplete, suggested lines of code and helped developers find solutions. Coding agents operate at a different level. Given an objective, they can break it into steps, write code, run tests, identify errors and revise the result. “The shift to agentic AI isn’t a bet,” says Sriram Dinavahi, SVP, Engineering at Salesforce. “It is the logical next step in how we build software.”
That is also changing the engineer’s role. Developers are spending less time writing every line of code and more time reviewing AI-generated work, setting guardrails and solving harder problems. Salesforce is using AI for debugging, test generation, incident response and repetitive software scaffolding. “Everyone is a builder,” Dinavahi says, arguing that engineers can increasingly direct and review what AI agents produce. He says this gives them more time for “ideation, collaboration” and systems thinking.
The shift raises a larger question about how technology companies should measure engineering productivity. If an engineer once measured output by the amount of code written, the more useful question now is how much reliable software can be delivered by a person working with AI. That could change how companies plan their technology organisations, particularly if these productivity gains persist as coding agents become more capable.
But it would be premature to conclude that AI simply means fewer engineers. “Salesforce’s deployment of AI coding agents provides early evidence of a structural shift,” says Prabhu Ram, VP, Industry Research Group at CMR. He expects engineers to spend less time on implementation and more on orchestration, validation and system design. But new products, technologies and growing software complexity will continue to create demand for engineering talent.
Deloitte India’s partner Shraban Nayak makes a similar point. AI can automate code generation, documentation, testing and defect resolution, freeing developers to focus on architecture, security, business problems and design decisions. The paradox is that while AI may reduce the amount of routine coding, it could make human judgment more important.
The future engineer may therefore be less of a code producer and more of an architect, reviewer and decision-maker. Programming will remain important, but systems thinking, security, AI orchestration and business knowledge are likely to matter more.
For an industry accustomed to scaling through headcount, Salesforce’s experiment points to a different model: smaller teams tackling larger problems, with productivity becoming a more important measure of engineering capacity than the number of people writing code.