The future of AI in the enterprise is already here, but its distribution is profoundly uneven. A simple glance at two conversations in London illustrates this disparity vividly. In one meeting, the head of engineering at a large hedge fund described teams running fleets of AI agents in full production, with nearly all code written by large language models (LLMs)—though interestingly, junior hires are not allowed to use them. In another meeting, a data engineer at a major retail bank reported the exact opposite: no agents in use and only sporadic use of LLMs. These two examples show that not only do different companies adopt AI at vastly different paces, but even within the same organization, adoption varies wildly across departments.

This is not a simple story of early adopters versus laggards. Rather, it reflects a fundamental reality: new technologies spread unevenly, and AI is widening the gap between teams that can operationally absorb it and those that cannot. Recent data supports this view. McKinsey's surveys indicate that while 88% of organizations use AI in at least one function, only about one-third have begun scaling AI programs. When it comes to agentic AI—autonomous systems that act on behalf of users—just 23% report scaling such systems somewhere in the enterprise, while 39% are still experimenting. In any given function, no more than 10% say they are scaling agents. This broad usage does not equate to deep institutional change.

This unevenness is a key reason why there is still time for companies to figure out AI. Being behind is not the same as being doomed. The real divide is between teams that have learned how to integrate AI into repeatable, governed workflows and teams that still treat AI as a promising but risky sideshow. Deloitte's 2026 enterprise AI research reinforces this: only 25% of respondents have moved 40% or more of their AI pilots into production, and just 34% say they are using AI to deeply transform their businesses. The remainder are using AI at a surface level, with little change to core processes. This is not a tidal wave; it is a messy, uneven organizational test, much like other technological shifts in history.

The engineering paradox: More jobs, not fewer

One of the most persistent narratives surrounding AI in enterprise is that it will decimate software engineering roles. The data, however, tells a different story. TrueUp data shows engineering openings at their highest levels in more than three years, with 67,665 open positions as of March 2026—up 78.2% from the recent low. Moreover, these openings are not concentrated solely at senior levels: 44.6% are entry and mid-level roles, compared to 38.3% senior and 13.8% senior-plus. This suggests that companies are not eliminating junior developers; they still want many engineers, even as AI tools spread throughout the profession.

The explanation lies in the Jevons paradox, invoked by Box CEO Aaron Levie: when a capability becomes cheaper and easier to consume, demand for it tends to rise rather than fall. Cloud computing did not lead to less compute; it led to more. AI-assisted coding appears to be doing the same for software itself. McKinsey's research on software development found that high-performing AI-driven organizations see 16% to 30% improvements in productivity, customer experience, and time to market, along with 31% to 45% improvements in software quality. But these gains do not come simply from adding copilot tools. They come from redesigning roles, workflows, and entire product development systems. That is a much harder organizational challenge than buying licenses.

Stack Overflow's 2025 survey found that 84% of developers are using or planning to use AI tools, and over half of professional developers use them daily. Yet this does not translate into fewer jobs. Instead, it is changing what enterprises want from engineers. Less time hand-authoring code, more time specifying, reviewing, steering, and orchestrating systems that generate code. This is particularly true in heavily regulated industries like finance, where governance is the hard part. Deloitte reports that only 21% of companies have a mature governance model for autonomous agents, and 73% cite data privacy and security as a top risk. Plugging non-deterministic AI systems into deterministic, compliance-heavy environments is messy and slow.

The real cost of caution

Caution is not without its risks. Every quarter a team spends in pilot mode is a quarter in which more aggressive peers are building operational muscle. OpenAI's enterprise usage data shows that frontier workers—the 95th percentile of adoption intensity—send six times more messages than the median worker. Frontier firms send twice as many messages per seat. The primary constraints, says OpenAI, are no longer model performance or tools, but organizational readiness and implementation. This rings true: the real divide is not between companies that have access to AI and those that do not, but between teams that have learned to integrate AI into repeatable work and those that treat it as a promising sideshow.

This also clarifies the distinction between tasks and jobs. Writing boilerplate code is a task; engineering is a job. Jobs bundle judgment, trade-offs, accountability, architecture, security, integration, testing, and the messy reality of operating systems. AI can automate more tasks, but it has not eliminated the need for jobs, especially where bad software decisions carry operational or regulatory consequences. McKinsey's broader survey found that high performers stand out by redesigning workflows and treating AI as a catalyst for innovation and growth, not just efficiency. That is a far cry from giving everyone a chatbot and reducing headcount.

The enterprise AI landscape is not heading toward one uniform future where software engineers quietly disappear. Instead, AI is splitting enterprises into fast-learning and slow-learning teams, rewarding those that redesign work, govern risk, and turn lower software costs into more software, not less. The code may be getting cheaper, but the ability to decide what should be built, how it fits together, and how to keep it from breaking the business is increasing in value. That is not the death of software engineering; it is a repricing, and every company and every team is paying different prices.

Source: InfoWorld News