The future is polished slop

AI slop is everywhere. You probably already know that.

What may be less obvious is that professional workflows are polishing that slop to an increasingly high degree. Sufficiently polished slop may already be indistinguishable from artisanal output. And the amount of work required to reach that level of polish is falling quickly.

Not all slop is created equal.

An army of agents

hob is built by a couple of developers and an army of agents.

We have worked with coding agents for years, including when they were not very good. We have become skilled at corralling them, surrounding them with guardrails, testing their work, and putting in the elbow grease required to polish what they produce.

I am a backend guy. I have a vision for the system, a security- and privacy-first mindset, and strong opinions about ergonomics for both humans and agents. Modern teams are composed of both, and software increasingly needs to work well for the entire unit.

When I begin implementing a set of features, I already have a deep idea of what I want. I know roughly how I would build it, which security failures I want to prevent, how the finished system should feel, and which primitives it should rest on.

That is where I guard against slop.

I care less than I used to about the shape of the code at the character level. Character-by-character authorship is no longer the most valuable use of my time. But one level above that, I have more opinions than ever.

When the work fails to meet those standards, I polish it. I use more agents, automations, reviews, and deterministic checks wrapped around probabilistic workers. Each pass moves the system closer to something I am willing to ship.

Then I test it.

I smoke test while agents are working, before their changes land, after they land, and between other tasks. ABT: always be testing.

Automated tests prove properties of the system. Smoke testing builds a feel for it. That matters when you did not type every line yourself. You need to internalize how the system behaves, where it bends, and where it breaks.

That is development work now.

I am still an engineer. I simply spend more time polishing in post than I spend typing source code. That time is neither free nor effortless, but the velocity toward a polished result is still dramatically higher than it was when I coded everything by hand.

Expertise is where slop gets caught

The backend is only the part where I reliably recognize slop.

The frontend I produce to exercise those capabilities is often ugly. Sometimes I do not notice how ugly it is. Sometimes I even think it is good.

Then I hand it to my cofounder, who is much more of a frontend guy.

He smirks, tears it down, and starts again with his own army of agents. His process looks remarkably similar to mine: strong opinions, tests, iterations, automations, and an unreasonable amount of polish.

The result should spark joy. It should feel effortless, classy, minimal, and modern. He knows when it does not, even when I cannot see why.

The reverse is also true. You might be able to sneak backend slop past him more easily than you could sneak it past me. We are expert de-sloppers primarily in the domains where we were already experts.

Our loose division of labor lets us context-switch less, focus harder, and polish more aggressively.

The same pattern repeats across the company.

Our marketing lead starts with a vision and an army of agents, then produces videos, animations, and assets through iteration, automation, and taste. If I show him an “impressive AI video,” he identifies it as slop nine and a half times out of ten. He can pick apart problems my eyes cannot see until he points them out.

Slop blindness is real

We are an AI-native organization filled with passionate experts.

Those experts are transitioning from being the sole crafters of their output to directing, evaluating, and polishing a much larger volume of it. Their expertise has not become less important. It is the thing preventing raw generation from becoming the finished product.

But expertise is uneven.

Slop blindness is real, especially outside your own domain. Something can look finished because you lack the taste or experience required to see what is wrong with it.

That makes the people around you more important. You need passionate experts who notice different kinds of slop, hold different standards, and refuse to ship work that merely looks plausible.

Polish is the work

The velocity increase is real. Even after accounting for orchestration, review, testing, and rework, we produce polished results much faster than we could by hand.

The cost of reaching that polish will continue to fall. Agents will improve. Workflows will become more sophisticated. Automations will absorb more of the repetitive review and correction process.

Polished slop is the future.

At some point, sufficiently polished AI output becomes indistinguishable from artisanal work. When that happens, its provenance matters less than the taste, testing, judgment, and care embedded in the process that produced it.

The meaningful distinction will not be between human work and AI slop.

It will be between raw output and finished work.

Generation is cheap. Polish is getting cheaper. Expertise and taste remain scarce.