Paraphrasing William Gibson, the future of AI is here, but it's nowhere close to evenly distributed yet. This stark unevenness is the defining characteristic of enterprise AI adoption in 2026. Last week in London, two conversations highlighted this divide: one hedge fund's engineering team uses fleets of agents in production and all code is written by LLMs, while a retail bank's data division has no agents and sparse LLM use. These are not outlier examples; they reflect a broader pattern confirmed by multiple surveys and reports.
McKinsey's latest research shows 88% of organizations use AI in at least one business function, but only about one-third have begun scaling AI programs. For agentic AI, 23% report scaling such systems somewhere in the enterprise, while 39% are still experimenting. In any given function, no more than 10% say they're scaling agents. This gap between experimentation and production is the central challenge. As Deloitte's 2026 enterprise AI research reveals, only 25% of respondents moved 40% or more of their AI pilots into production. Just 34% say they're using AI to deeply transform their businesses, while 37% still use it at a surface level with little or no change to core processes.
Cue the engineering boom
Contrary to fears of AI wiping out software jobs, engineering openings are at their highest in over three years. Lenny Rachitsky highlights TrueUp data showing 67,665 open engineering jobs as of March 2026, up 78.2% from the recent low. Importantly, 44.6% of posted roles are entry and mid-level, versus 38.3% senior and 13.8% senior-plus. AI coding tools are not eliminating roles; they're changing what enterprises want from engineers. Box CEO Aaron Levie invokes Jevons paradox: as software production becomes cheaper, demand for software increases. Cloud computing didn't reduce compute needs; it expanded them. AI-assisted coding is driving a similar expansion in software demand.
Stack Overflow's 2025 survey found 84% of developers using or planning to use AI tools, with over half of professionals using them daily. McKinsey's software development research shows top AI-driven organizations see 16-30% improvements in productivity and 31-45% improvements in software quality. But these gains require reworking roles, workflows, and the full product development system. That's a much harder organizational challenge than buying copilot licenses.
Software engineering is alive and well
The hedge fund leader may represent an early glimpse of the future: less time hand-authoring code, more time specifying, reviewing, steering, and orchestrating systems. But the retail bank division is not irrationally lagging. In regulated environments, code generation is not the hard part—governance is. Deloitte reports only 21% of companies have a mature governance model for autonomous agents, and 73% cite data privacy and security as top risks. This is not bureaucracy for its own sake; it's recognition that plugging non-deterministic systems into deterministic, compliance-heavy environments gets messy fast.
Caution isn't free. Every quarter in pilot mode allows aggressive peers to build operational muscle. OpenAI's enterprise usage data shows frontier workers send six times more messages than median workers, and frontier firms send twice as many messages per seat. The primary constraints are organizational readiness and implementation, not model performance. The real divide is between teams that have integrated AI into repeatable work and those still treating it as a dangerous sideshow.
The distinction between task and job matters. Writing boilerplate code is a task; engineering is a job. Jobs bundle judgment, trade-offs, accountability, architecture, security, integration, testing, and the reality of operating systems. AI can automate more tasks, but hasn't eliminated the need for jobs, especially in environments with real consequences. McKinsey's broader survey finds high performers redesign workflows and treat AI as a catalyst for innovation and growth, not just efficiency. So no, AI isn't heading toward one uniform future where software engineers fade away. It's splitting enterprises into fast-learning and slow-learning teams, rewarding those that redesign work, govern risk, and turn lower software costs into more software. The code may get cheaper, but the ability to decide what to build, how to fit it together, and how to keep it from breaking the business keeps increasing in value. That's not the death of software engineering—it's the repricing of it, and every company and every team is paying different prices.
Source: InfoWorld News