Intent doesn't flow in one direction. It loops.

Design informs systems, engineering shapes experience, and every artifact feeds back into the next. As AI absorbs the work, strict divisions between roles collapse, and the line between implementation and interpretation fades.

As the distance between idea and execution disappears, how we build splits in two.

How do we build when the loop includes us?

Contents

1. Loops, not pipelines
2. The collapse of boundaries
3. Structural taste
4. The second half of AI: the return of the manager
5. Comprehensive design
6. Conclusion

Loops, not pipelines

Feedback loops have always shaped good software. What’s changing is the speed, depth, and density of those loops, and who’s inside them.

In a loop, there is no "before" and "after." You don’t write the code and wait for the result. You shape the system and feel it respond. Each commit is a move in a conversation with it; each prompt is a bet.

Engineering stops being about specifying a system and stepping back. It becomes about staying in the loop long enough to steer it.

The collapse of boundaries

“You become a senior engineer when you stop waiting for specs to be right and start building the conversation instead.”— @graycoding

Design, engineering, and product were never as separate as their org charts suggested. Now the distinctions are visibly breaking down.

Models blur the roles. Specs become interfaces, engineers prompt, designers scaffold.

The product engineer is a much needed site of convergence. They work across domain models, UX flows, and infrastructure, not to own a layer but to ensure intent survives translation.

When boundaries collapse, coordination shifts from handoffs to coherence.

Structural taste

In this collapsed stack, 2025 brought "vibe coding": shaping whole systems in dialogue with the machine, leaning on tacit understanding rather than explicit logic.

But vibe is fragile and doesn't scale. When it works, it’s resting on something firmer: structural taste.

“You can’t vibe your way to a good product no matter how good your AI tooling is.” —@thenimmyjeutron

Structural taste is judgment in architecture: a specific, local conviction about how things should be, not a universal standard of high art.

It’s less about being smarter than about giving a shit when the machine makes it easy not to.

When AI responses are all 'good enough,' taste is the refusal to accept the median outcome.

It’s caring about the semantic precision of the domain model, the threshold of a loading spinner, the specific tone of a zero-state. These small things decide whether a system is habitable.

Holding a system coherent against entropy, choice by choice, is what intent actually means here.

“Prompting is thinking. The sharper your thoughts, the better the prompts, the better the outputs.” —@rauchg

The second half of AI: the return of the manager

As the friction of implementation dissolves, the engineer’s role expands in two directions.

We go deeper. Engineers use AI to take on complexity that was previously insurmountable. We become comprehensive designers of systems we couldn’t hold before.

We go wider. If product engineering is the collapse of the stack, this is the compression of execution.

“The second half of AI – starting now – will shift focus from solving problems to defining them. [...] To thrive, we’ll need a timely shift in mindset and skill set, ones perhaps closer to a product manager.” — Shunyu Yao, The Second Half

Here the engineer’s end state is a manager, but not of people. The cybernetic manager directs the compiler, the agent, and the business logic at once.

In this low-latency loop, the linear relationship dissolves. Implementation informs intent as much as intent causes implementation.

Comprehensive design

Whichever mode we operate in, strict specialization by role is dead. We are all systems thinkers now. What remains is specialization by taste, and a deeper question of responsibility.

That leverage comes with weight. As it scales, so does the impact of our design choices; every system we shape shapes something larger: economic, ecological, social.

Buckminster Fuller, the original comprehensive designer, framed "Spaceship Earth" as a design challenge: a closed system with limited inputs and global interdependence.

“We are all astronauts on a little spaceship called Earth. But it did not come with an instructions manual.” — Buckminster Fuller

As the loop tightens, the boundary between 'lab' and 'world' becomes porous. What we treat as local variables now triggers global side effects.

The challenge is to build systems that cohere, not just function. Systems that treat life as a design constraint, not an edge case.

Spaceship Earth is our shared platform. We’re constantly pushing updates.

Conclusion

Whether you are deepening the stack or expanding into management, the requirement is the same: hold the loop open. Interpret, structure, guide, and stay responsible for what the loop puts into the world.

Shipping the product isn’t the end of that. It’s where the next loop starts.