The deep work versus all the work

With AI tools there's more software shipping more than ever. The thread across my talks with other leaders are consistently being told to scale beyond adding headcount. The model of the trifecta and design being embedded in every piece of work on a single roadmap just isn't scaling for this new world. If we don't adapt we end up being either overwhelmed with work or only overseeing part of the product.

Watch enough work go out without design and you notice what's slipping: quality. Not some abstract version of it, the standard you've held for years, now leaving the building in your name without your hands ever having touched it. Hope and a design system alone cannot hold your design strategy together.

What quality actually is

The easy mistake is to think quality is how something looks or interacts. But looking good, and even working well, are mostly table stakes once you have a mature design system. The system carries a lot of this for you.

Real quality is fluffier than that. Whether the thing solves the actual problem. Whether it still holds together a year from now. Whether it stays accessible, fast, and steady under real use. Whether the experience feels like one thing, rather than a dozen good intentions bolted awkwardly together. And whether there's care in it, the sense that a person who made it gave a damn.

That kind of quality is hard to operationalise, because it's many things at once. Some of them a machine can now do. Some of them still need deep understanding and care.

The human in the loop

A surprising amount of quality now sits within reach of an LLM and a solid design system. Consistency, accessibility, performance, the patterns we already know to be good, this is best practice, and best practice is exactly what a machine is built to repeat. It gets us most of the way there in minutes, where it used to take weeks.

That doesn't remove us. It moves us up the chain. We become the human in the loop, making sure what comes out the other side clears the bar. Using an AI system like this is a skill in itself, there's a marked difference in one-shotting versus a honed outcome.

The AI in the loop

The flip side is more interesting. Once the machine is handling the repeatable making, our scarce attention is freed to move upstream, to the part an LLM can't do well. The empathy. The lived experience. The instinct for what a person feels when they land on a screen.

Part of this is the deep discovery: the messy, human work that happens before anything gets built. Finding the real problem and understanding who it affects. Bringing taste, a point of view, and the discipline to know what not to build. Sensing what people will tolerate and where the real pain points are. It's where the moments of magic come from, the reason someone remembers your product over anyone else's. Hard to measure, harder to describe, and usually the first thing to disappear when everyone's at capacity.

And the care isn't meant to be spread evenly. Every product has a handful of moments that carry it, the point where someone decides to trust you, the point where the thing finally pays off. Pour iteration into those and the return is wildly out of proportion to their size. Try to apply the same effort to every screen and it spreads thin. A small, eloquent interaction in the right place outweighs reasonable polish everywhere else.

It needs more of our focused attention now to make software feel human-made.

Not everything needs this

If we try to bring that level of attention and care to everything, and it spreads us too thin to achieve it.

The teams that get this right are deliberate about it. They separate the moments worth real intention from the parts they're happy to let best-practice generation handle, and they can control which is which, and why.

A settings screen. Another wizard flow. That work needs to be done well and done quickly, and an LLM thrives there. It rarely needs the deep dive. Spending your scarcest attention on it is how the work that actually deserves that attention ends up going out with nobody's hands on it.

The judgement is the job

So the skill now is knowing when to be the human in the loop, and when to be the loop yourself. Being the human in the loop on the routine work is exactly what buys back the time to bring empathy, taste, and judgement to the work that needs all three.

This isn't the lesser job or a loss. It's having greater control over greater scale.