Every engineering abstraction moves complexity somewhere else. AI is no different. When someone tells me AI cut their engineering effort by 30%, the accountability question never changes. Who is on the hook when it ships? The answer is never the AI that gives the false confidence of efficiency. The answer is always human-centric, either a team or an individual is always responsible and held accountable for the work that is delivered.

The headlines of efficiency gain and reduction of human efforts sound ecstatic! Engineering staff cut down by 90%, or AI has improved productivity by 100%. The numbers look sleek on investor presentations and in board meetings.

Don’t get me wrong, I am not against AI, it has and will continue to impact and change the world in ways we can not imagine. The future (optimistically) is very bright and wondrous.

The Philosophy of Judgement

Here’s what I have seen up close. In its current state, AI is a really really powerful fool (powerful tool, nonetheless, but more so a fool, and I’m cautiously intentional about it). There are countless stories of historical or mythological lore, where an impressively strong (yet not so intelligent) being that can do extreme wonders, is always the subject of morality - what can be used for greatness can also be used for great evil, and the intentions behind who drives this being are amplified. The similarities of these folklore to AI is uncanny. With certain blunt work it can lift ten times the effort in a fraction of time. However if not correctly guided, it can also derail an entire feature and cost you hours in remediation. While nitrous oxide is great to have on a race car, and can deliver a final push to win, if the driver isn’t skilled enough to use it correctly, it usually puts you into the wall in a fireball.

The guiding principles and values when using any technology are extremely important. Modern human history is littered with stories of great technology, bad human judgement, and extreme mistakes that became irreversible. The Titanic didn’t sink due to bad technology, it sank due to bad human judgement. The Challenger disaster, if anything, taught everyone that you could have the best rocket scientists and most advanced technology in the world, but bad organizational judgement can cause the worst of accidents and immense human suffering. Technology amplifies capability, it also amplifies the consequences.

I think now more than ever, it is crucial to have good human judgement. AI is a force multiplier, and amplifies the user’s judgement. It’s a great time to keep your wits sharp! Don’t fall prey to the folly of outsourcing your natural thinking to the great grand artificial intelligence, lest you become an artificial human devoid of natural thinking capabilities (there seems to be so much of that going on in the world these days). I think the “Automation Paradox” applies here so elegantly - as automation improves, humans spend less time practising the underlying skill. Because of this, if intervention is ever needed, humans are less capable to intervene because the skill has atrophied away. This happens across fields of study, be it engineering, medicine, aviation, and so on. While there is no reason to sound the alarm, it is quite alarming that we are more prone (just by nature) to atrophy away the very human essence of critical thinking.

Estimation, Review, and everything in between

If you’ve been in Engineering long enough, you know that there is a process (or various processes) for getting from an idea to a completed implementation. Brainstorming, Ideation, PRDs, UX/UI design, estimations, specifications, architecture, system design, implementation, testing, more testing, review, delivery (and many many more steps). AI has improved the speed and efficiency of all of these steps. A lot of the blunt work that went into structuring documentation around various steps has come down. AI is able to produce well-structured documentation in each step.

The folly here is that the thinking does not get faster because the first draft arrived quicker. Writing PRDs, or writing code for that matter, was never about typing and sentence structure. It was about the idea, the logic, the thinking behind all this, which transformed something virtual into a concrete structure. While AI does good analysis work when all the information is there, there still needs to be the right amount of glue-work and guidance that steers the process in the intended direction.

It’s very easy to book a 30% reduction in estimates citing that AI has made work faster. But statistically, more AI output has also driven the need for more human review. Human review has not gotten faster, but the volume of output that needs review has gone up drastically. This wipes away the gains from quicker generation and iteration. AI has basically moved the work down the assembly line. What used to take more time during initial prototyping and analysis, has now moved to post-processing review and evaluation.

Accountability and Ownership

As an employee, as a developer, as a manager, as a human, I am accountable for the work I commit to deliver. An estimate is a number that I can put on that commitment. While not all estimates can hold true, the human can still be held accountable to commitments made. But if my estimates assume a percentage gain from AI usage, AI is not accountable to deliver that gain. Yet as a human, I am accountable, and on the hook to deliver. This puts undue pressure on the most critical aspect of the development process - the human in the loop. While it seems easy to say that AI usage is a net positive gain in efficiency, the real world relies on human accountability, and the emotional pressures that come with that. When something goes wrong, the human in the loop pays the accountability price, and when something goes right, the AI tool is praised as the wonderful tool that it is.

This fundamentally puts humans in a difficult place - one which disincentivises human ingenuity and critical thinking. Humans have always created tools and technology that can be held to standards and held accountable in the progress of the task, and this has progressed humans into a future our ancestors could only have dreamed about. As we work with AI in this next chapter of modern human advancement, it is crucial that we be intentional about how and where we use AI. The pressure to book aggressive AI gains is real, and the path of least resistance is to quietly absorb the risk ourselves. I’m not arguing against using AI — I’m arguing for being honest about where the risk actually lands. AI doesn’t sign off on the final commit, we humans do.