For years, conversations about organizational efficiency have followed a familiar pattern. Reduce manual work. Streamline processes. Eliminate waste. Automate repetitive tasks. The underlying assumption was that people spent too much time executing work and that productivity improvements would come from making execution faster.
Artificial intelligence is beginning to challenge that assumption.
The first wave of automation focused primarily on predictable activities. Data entry, report generation, workflow routing, and other routine processes were natural candidates because they followed clear rules. AI expands the scope considerably. It can now assist with tasks that require interpretation, investigation, content creation, and even certain forms of problem solving.
What interests me most is not the technology itself but what happens to organizations when the cost of producing knowledge work starts to fall dramatically.
Anthropic recently shared observations from its own experience using AI to accelerate the development of AI systems. While their environment is far from typical, some of the patterns are worth paying attention to. Engineers were reportedly merging several times more code per day than they were only a couple of years ago. Much of that code was generated by Claude, while engineers focused on directing the work, reviewing outputs, and making decisions.
The obvious conclusion is that engineers are becoming more productive. The more interesting conclusion is that the nature of engineering work is changing.
Historically, a significant portion of software development involved translating ideas into implementation. Writing code was often the bottleneck. As AI systems become increasingly capable of generating working software, implementation becomes less scarce. The constraint begins to move elsewhere.
This becomes even more apparent when looking beyond simple coding tasks. Anthropic reported significant improvements in Claude’s ability to handle open-ended engineering problems. These are situations where there is no detailed specification, no predefined solution, and often no clear understanding of the problem itself. In one example, a routine upgrade caused tens of thousands of training jobs to fail. Rather than being given a carefully designed benchmark, Claude was pointed at a live incident with limited context and asked to investigate.
Whether these results generalize across all industries remains to be seen. But they suggest that AI is progressing beyond task execution and beginning to participate in forms of work that involve exploration, diagnosis, and reasoning under uncertainty.
At the same time, an important distinction remains. The comparative advantage of humans today is still in seeing the bigger picture and thinking beyond the confines of the immediate task.
Organizations are complex systems. They contain competing priorities, hidden constraints, cultural dynamics, regulatory requirements, customer expectations, technical dependencies, and long-term strategic objectives. Solving an isolated problem is often easier than understanding how that solution affects the rest of the system.
An AI model can propose an implementation. It can investigate an incident. It can generate a report. What it struggles with is understanding why a particular initiative matters more than another, whether a local optimization creates larger organizational problems, or whether a project should exist at all.
These questions are increasingly becoming the critical ones.
For decades, many organizations measured productivity primarily through output. How many features were delivered. How many reports were produced. How many tickets were closed. If AI continues to reduce the cost of generating output, those metrics become less meaningful. Producing more is no longer the difficult part.
The challenge shifts toward deciding what is worth producing.
This may explain why some organizations see dramatic gains from AI while others struggle to realize value. The technology itself is only part of the equation. Organizations that understand their objectives, maintain clear priorities, and have strong decision-making processes are often better positioned to benefit from increased execution capacity. Organizations that are already overwhelmed by competing initiatives may simply generate more activity without generating more value.
The conversation about AI often focuses on replacement. Will AI replace developers? Analysts? Designers? Managers? A more practical question may be how the distribution of work changes when implementation becomes significantly cheaper.
In software development, the engineer’s role increasingly includes framing problems, defining goals, reviewing outputs, evaluating trade-offs, and ensuring alignment with broader architectural and business objectives. Similar shifts may emerge across many knowledge-based professions.
The future of organizational efficiency may not be about doing the same work faster. It may be about redesigning organizations around a new reality: execution is becoming abundant, while judgment, context, prioritization, and systems thinking remain scarce.
That is not a technology challenge.
It is an organizational one.
And it may turn out to be the more difficult problem to solve.