AI Made Code Cheap. The Rest of Engineering Just Got More Expensive.

For most of the last two decades, engineering organizations bent their roadmaps, their headcount plans, and their sprint math around a single piece of undisputed economics: writing software was the expensive part, so the scarce resource was a person who could turn an idea into working code. Everything downstream of that belief—how teams were sized and where the money went—inherited its logic.

Patrick D'Souza, Senior Director of AI Enablement and Data at Zeta Global, a data-driven marketing technology company, thinks that belief has come apart, and that a lot of teams are misreading what it takes with it. In a LinkedIn post titled "The Economics of Software Have Changed," he wrote that "For years, we optimized engineering around one assumption: Writing software was expensive." His argument isn't that the assumption got easier to satisfy. It just stopped being true.

The easy reading of the AI coding wave is that it's about tempo, that assistants let a developer produce the same work in less time. D'Souza wants to pull those two ideas apart. "I don't think the biggest shift is that developers can write code faster," he wrote. "I think the economics of software development have fundamentally changed." Faster typing is a feature of the tools, but a repriced cost structure changes what the job actually is.

Code is cheap. Alignment isn't.

The consequence he draws from that is specific. "When the cost of implementation drops, the bottlenecks move somewhere else. Suddenly, alignment, decision-making, context, architecture, and governance become a much bigger part of the equation." When the step that used to gate delivery stops being scarce, whatever was second in line becomes the new constraint, and none of the things he names are tasks a model does for you.

The size of that cost drop is easy to underrate. In Stack Overflow’s 2025 survey of more than 49,000 developers, nearly four in five respondents said they already used AI tools in their development process, including 47% who used them daily. Writing, debugging, testing, and documenting code are all increasingly being handed to AI. The expensive step in D’Souza’s story, producing the software itself, is already being compressed across much of the profession.

There's more to engineering than code

That step was never the whole job to begin with. Squeeze the time spent producing code, and the coordinating, reviewing, and deciding around it doesn’t shrink in tandem. It takes up more of the work that remains. That’s the change D’Souza is pointing at: as implementation gets cheaper, the judgment surrounding it becomes more expensive.

This is why he sees the purchasing question as the least interesting part of the conversation. "That's a very different conversation than simply asking, 'Which AI tools should we use?'" he wrote. A team that answers the tooling question and stops there has sped up the cheap part and left the expensive part untouched. Deciding what's worth building, holding an architecture coherent as the system grows, keeping enough shared context that a dozen people build the same thing, all of it rests on experienced technical judgment, and that's something a hiring plan produces slowly and a license key doesn't produce at all.

The next great engineering hire

For engineering leaders, cheaper implementation changes the hiring question. A team can generate plenty of code and still lose months to muddled priorities, architectural drift, and decisions nobody owns. The people who prevent that waste may produce less visible output themselves. They create the conditions under which everyone else, human or agent, produces work that holds together.

The next great engineering hire may be the person who saves the team from building the wrong thing at unprecedented speed.