I recently saw a post that said "AI content is cheap to create, and incredibly expensive to consume". I have thought about it multiple times since, as it really encapsulates a growing frustration with AI-first workflows.
Co-workers send Claude-generated documents that are so long that the receiving party can only efficiently digest it by putting it into an LLM. Design artifacts that are so riddled with extraneous copy or visual elements that it's hard to tell what you should actually be paying attention to versus not, and what is worth critiquing.
Then over coffee yesterday, a colleague made a suggestion: instead of a token budget, people should have an output budget. You can only send so many words to your coworkers every day, and it's your job to ensure that you refine things enough to stay under that limit.
Now, obviously, this was tongue-in-cheek. But as glib as it was, I think it plays into a feeling that I've observed in many corners of AI-enabled building today.
We are willingly abdicating ownership over the artifacts we're sharing with team members. People are prefacing them as "Claude says..." as if that is an excuse for poorly examined concepts or overly complicated writing.
More importantly, this is simply transferring the burden to the reviewer or recipient. At its worst, it's causing humans to completely check out of the process, at which point, what is even the point of sharing this?
In one of my favorite blog posts from the past couple weeks the author used the term "meat proxy", which is an excellent turn of phrase.
I think this is a fundamental challenge for a lot of teams that are trying to move faster and leverage AI.
You can't delegate critical thinking and conviction to the LLMs, because ultimately they have neither. They can be a useful thought partner and building tool, but in terms of really gaining long horizon user-centric conviction, they are pretty crap.
I have 100% been guilty of this over the past six months, and I am now realizing how detrimental it is becoming both to my own sense of conviction in what I'm sharing, as well as how our teams work with each other.
Without being reactionary, I think one key part of a solution here is to set the cultural norm and expectation that sending someone something that you haven't thoroughly edited, written, or at least could critically defend on a five-minute call is simply unacceptable. This should be true in both life and business.
And if you're sharing something that you don't have the expertise to thoroughly examine (ie. technical code review comments) then, at minimum, be up front about this and work to eliminate that gap in the future.
Let's stop sharing AI slop and stop turning each other into meat proxies.