CTO of Softengi with 30 years of experience in software development, business applications implementation and digital strategy creation.getty​For decades, software companies hired engineers as the primary production force, paid salaries, provided tools and obtained software in return. In 2026, generative AI is unsettling that model. Modern-day developers rely more on Claude Code, Cursor or GitHub Copilot to generate meaningful software, tests and architectural solutions fit for production. Meanwhile, software development consumes a substantial amount of AI inference, making it a cost factor. As cognitive computing becomes an expense category, it raises the question of how software engineering economics can keep pace.The Incentive ProblemAI tool expenses may seem small relative to unlocked productivity, so any incentive that saves a few percent of AI input costs might destroy more output value than it saves. However, it is the magnitude of such an approach that rings the alarm, where the issue is not AI misuse but alignment.• The cost of employing a senior engineer is typically $94,000–$215,000 annually. • At companies like Cisco, a developer can spend $10,000 annually on token usage.Most organizations pay for AI instruments centrally. Developers get near-unlimited access to regenerating code, automating trivial tasks or calling on AI where a five-minute manual fix would suffice, while a company pays the bill. When a consumer of resources is not the one who incurs the costs, there is almost no reason to save someone else's money, which is a textbook case of a moral hazard. Developers As Operators Of Cognitive CapitalTraditionally, companies pay for human expertise, labor hours and tools, with cognitive computing emerging as the fourth line item that should be managed as a first-class production input. Positioning developers less as resource consumers and more as operators of cognitive capital increases accountability for outcomes and emphasizes deploying intelligence to drive delivery. Still, a proper compensation model that fuels this concept is to be found.An Obvious Compensation Model That Nevertheless FailsLet’s assume we hired an engineer at a total cost of $150,000 per year, allocating $120,000 to base salary and $30,000 to a personal AI budget, then converting any unspent funds into a bonus.At first glance, it encourages thoughtful use of AI resources while rewarding savings. In practice, once AI usage is tied to an engineer’s personal budget, they are no longer incentivized to pay for tokens. Usage decisions will no longer be based on productivity but on preserving the engineer's budget. That leads to underuse in cases where AI would have been cost-effective for the company. In the false-economy version, a developer spends three hours manually doing what $4 worth of tokens could accomplish in twenty minutes because the $4 is theirs and the three hours is the company's. Developers will "save" their $4 at the company's expense of $300, while feeling virtuous about it.A Lesson From The Sales DepartmentMature sales organizations solve a structural compensation problem by paying commission on margin or profit, not raw revenue. Once compensation tracks the net outcome (value produced minus cost incurred), a sales manager automatically internalizes costs because wasteful spending reduces their commission base. Cost follows value, so efficient spending is a result of creating value rather than a target to be policed. If we translate margin-based commission into software engineering compensation, a more sustainable structure emerges: • A base salary that absorbs variance.• Outcome-oriented team-level variable pay tied to a net measure that already accounts for cost—value shipped per total dollar or P&L-style metrics for product-owning teams.• AI budget as a tool, not a trophy, motivating developers to decide whether spending it will pay off or should be saved for bonuses.Cognitive FinOpsCloud computing gave rise to FinOps, as teams became accountable for infrastructure costs and learned to balance performance against budget. AI is producing the same discipline with visibility metrics: cost per delivered features, team AI spending and outcome per unit of cognitive spend.Today, wasteful AI usage is structural rather than behavioral: missing caching, oversized contexts that can be trimmed at the platform layer and expensive models doing low-level work. These are engineering problems that require engineering solutions, yielding far greater savings than any individual-level incentive ever could.Wrong Incentives Can Jeopardize Output Quality Rewarding cost reduction in isolation will force engineers to avoid using powerful models even when justified, dodge challenges that require more computing power and quietly substitute slower manual work for faster AI-assisted solutions. To streamline innovation and delivery, the focus should be on maximizing outcomes relative to total costs, rather than minimizing token use for its own sake.The Engineer Of TomorrowCompanies have long paid developers for output measured by closed tickets. Essentially, they are willing to pay for judgment and the ability to deploy intelligence where it matters, so the cost of tokens is not the issue.A forward-looking engineer cares less about how much code they write and more about how effectively they orchestrate human expertise, AI capabilities and automated workflows to optimize software and cognitive architecture.Rewarding leverage instead of output is one of the most consequential changes in software economics since the advent of the cloud. But the new paradigm won’t affect engineers' compensation—it will simply redefine how it’s calculated.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?