The Role of Human Intuition in an Agent-Driven Engineering Workflow

By 2026, the daily workflow of a software engineer has shifted from writing code to managing a fleet of agents that handle everything from Slack messages to high-level architecture. This transition has turned high-performing engineers into '100x engineers,' but it has also highlighted a critical distinction: the human's role is moving away from execution and toward the nuanced judgment of problem definition and long-term viability.

The Shift to Agent-Led Workflows

The integration of advanced models, such as Opus 4.6, has moved debugging and architecture into the realm of agentic interaction. Engineers now frequently delegate writing tasks and even high-level thinking to these tools, often spending their time ensuring that different agents do not block one another.

Despite this automation, the output of engineers is not equalizing. Because great engineers still produce significantly better results than average ones, it suggests that the human's ability to steer the process remains a primary differentiator in the quality of the final product.

“But until then, our leverage isn't just in what we know, but in everything we've experienced that made us who we are.”

— On Being an Engineer in 2026

Tacit Knowledge and the 'Slop' Filter

A core differentiator for human engineers is 'tacit knowledge'—the idea that humans know more than they can tell. This manifests as an intuition for which problems are genuinely hard and which solutions will remain viable years into the future, rather than just solving a problem in the immediate moment.

This experience-based judgment also creates a collective 'allergy' to AI slop. Humans can often identify low-quality, generic AI output almost instantly, even without a specific definition of what makes a piece of content feel 'off' or poorly constructed.

The Trade-offs of Human Oversight

The evolving role of the engineer requires a human presence at both the beginning and the end of the development cycle. Humans are needed to define the intent of a project and to judge if the final result is sufficient for long-term use.

The trade-off for the modern engineer is that those who cannot steer an agent or distinguish between a hallucination and a useful fact are losing their value. The role is narrowing toward those who possess the product sense and diligent quality assurance to oversee the automated middle layer.

Uncertainties in Agent Evolution

It remains uncertain how long this human leverage will persist. As models continue to improve, they may eventually develop their own forms of 'taste' or experience-based intuition by processing the vast history of human experience.

Whether an agent can ever truly replicate the 'lived experience' that shapes human intuition—such as fears, desires, and mistakes—remains a central philosophical and technical question for the future of the profession.

Discussion

Commenters on Hacker News discussed the timeline of AI automation, with some expressing doubt that the current state of things will hold much longer. One commenter argued that a drop-in end-to-end replacement for a full software team could be less than a year away, leading to a 'bloodbath' once corporate leadership evaluates the necessity of human engineers.

Others emphasized the necessity of human oversight in the 'sandwich' model of development, where humans must define the goals and perform final quality assurance. One commenter noted that while AI can handle the middle execution, replacing the humans at the start leads to pursuing the wrong goals, while replacing them at the end results in 'slop.'

A technical counterpoint was raised regarding tacit knowledge, with one commenter suggesting that reinforcement learning post-training is essentially the process of training models on the very tacit knowledge humans use to differentiate quality. Another commenter argued that intuition is a product of human consciousness and biological systems like the limbic system, which agents cannot replicate without similar physical hardware.