As machines take over most code generation, software engineers will shift from writing code to constructing the tools that produce it, according to a new analysis published by The New Stack. The workforce will split into two camps: product engineers who create customer-facing applications, and platform engineers who develop the infrastructure those product engineers rely on. Because automated systems will handle more code creation, organizations will need additional platform engineers to construct and refine these machines.

Platform engineering work now extends beyond traditional delivery infrastructure to include oversight of which AI models and frameworks receive approval, cost management for agent usage, authorization protocols that distinguish between engineers and the agents acting on their behalf, and the construction of feedback systems that redirect linter or security scanner findings back into agents rather than simply halting builds. Organizations are discovering that what initially appeared as two separate platform functions—one for continuous integration and deployment infrastructure, another for agentic developer tools—is merging into a unified platform responsible for both delivery systems and AI guardrails. This consolidated platform's influence now reaches earlier into the development lifecycle, starting before any code exists rather than at the first commit.

The analysis describes harness engineering as platform work tailored for AI, involving the creation of feedback mechanisms, protective boundaries, and contextual information agents require to operate safely within a codebase. According to the report, "a good harness does two things: it raises the odds the agent gets the task right on the first pass, and it gives the agent a way to catch and fix its own mistakes before a human ever sees them." The report argues that frontier models provide engineers with tools for faster code generation, but platform teams must still assemble, constrain, and harness these models to prevent them from becoming a quicker method of producing poor-quality output. The authors note that every organization shipping software effectively becomes a developer tools company, constructing the systems its own engineers need to work safely with agents using the same rigor applied to external products.

The report explains that productivity gains with AI stem from the strength of the underlying machinery rather than which model a company purchases or how strictly it enforces adoption mandates. When platform teams fail to deliberately build governance layers for AI agents, individual teams create fragmented, redundant solutions with no clear ownership when problems arise. The analysis states that raw agents from providers like Anthropic or OpenAI function as engines without knowledge of a specific codebase's conventions, risk tolerance, or budget constraints, and transforming them into systems that respect those parameters is work only platform teams can perform. The report concludes that organizational charts will shift as more engineers move to platform roles focused on building and maintaining AI harnesses, since that infrastructure becomes the more challenging aspect of the job compared to product development. Organizations treating platform engineers as a support function rather than central to productivity will struggle to ship quality software regardless of their talent, because broken tooling prevents delivery no matter how capable the people using it are. The industry faces a choice between deliberate investment in platform engineering or accepting the friction and waste that comes from every team solving the same infrastructure problems in isolation.