AI;DR
What started happening to software engineering one year ago is now unfolding across the rest of knowledge work. And just like engineering teams in 2025, the rest of the modern organization is not equipped to deal with this monumental change. In fact, outside of the minority of companies that have fully embraced agents across every single function, I don’t think anyone is even aware of what’s about to hit them in three to six months, as agents become embedded in more of the workplace.
Documents are the currency of knowledge work. From the most banal worksheet to the most intellectually sublime strategy memo, their essential function is to capture and disperse thought so groups of people can work together with shared understanding and purpose. And as long as white-collar work has existed, creating a document, whether that’s a memo, slide deck, or spreadsheet, has required thinking about a topic and then deciding how to articulate it.
Documents used to be “proof of work.” They demonstrated that you put in the time to understand something. That is no longer the case. You cannot take documents at face value anymore.
As code has become cheap to generate, software engineering practices have had to evolve rapidly to use the benefits of coding agents while mitigating their downsides. Engineering teams on the frontier don’t read code anymore and are now even replacing code review with more scalable solutions.
Now that docs are cheap to generate, marketing, sales, finance, operations, and every other function will need to do the same to ensure that in the process of delegating writing to agents, teams are not also delegating their understanding.
Writing culture (as we know it) is dead
In 2004, Jeff Bezos outlawed PowerPoint presentations at Amazon. He required that all decisions be facilitated by written documents, most famously the six-page memo and the PR/FAQ. This ‘writing culture’ was eagerly adopted by startups seeking to emulate Amazon’s success, most notably (in my experience) Twilio and Stripe. At Twilio (where I worked from 2015–2020), we imported the full ritual — prereads with comments, quiet reading in the first ten minutes of the meeting, followed by a discussion closely following the outline of the doc. It was surprisingly effective!
Today, this writing culture is dead. Apparently, there’s a new term at Amazon now — AI;DR:
Amazon’s writing culture survived the pandemic. The printouts may have moved online, and study hall may have transformed into a shared doc with comments down the margin, but that was fine. People still wrote the documents, and people still read them.
But it did not survive the advent of AI.
From what I hear, the moment a model could produce something that passed as a human-written six-pager, they stopped being written by humans. A senior engineer in my community told me that he has started tagging documents as “AI;DR.” AI; didn’t read. In his view, “If a model wrote it, a model can summarize it.” He had better things to do with his time.
The L7 PM from above may have produced a six-pager, but because he had skipped the critical thinking step, he didn’t have a credible plan. I don’t know how the situation ended, but there are only two ways it could have gone. Either the plan got reworked, and a room full of senior engineers lost time to a meeting that should never have been scheduled, or it didn’t get reworked, and that team executed a plan nobody had thought through.
— Steve Huynh, “On the Death of Amazon’s Writing Culture”, A Life Engineered, October 14, 2026.
Writing a document used to force the author to think through many aspects of their argument (I routinely spent more than twelve hours on a single product strategy doc pre-AI), and shared reading gave everyone a common starting point, leading to a more productive discussion.
Once slopdocs proliferate, this cultural contract breaks down. AI;DR becomes the norm. People stop reading documents because they assume the author did not put much thought into them.
How to avoid slopdocs?
Steve suggests not allowing the use of AI in certain situations, like in documents used to make high-stakes decisions (one-way doors in Amazonspeak). I don’t think that’s a viable solution because it would be impossible to enforce well. Also, while there are telltale signs of AI writing today — “not X but Y,” “the interesting part is…” and so on — this will soon be solved too. AI writing will become harder to distinguish, so any potential solutions must work even when the output looks perfectly human.
The goal ultimately should be to preserve the original purpose of writing: encouraging people to think through a problem, develop a thorough understanding of it, and communicate that understanding to others. We need cultural expectations that hold regardless of how the document was generated, and ways to verify that those expectations were followed.
I don’t really have good answers yet. I haven’t seen any practices being used today that look promising (if you have, please drop me a note or comment). But here are two things that I think could be worth exploring:
1. Share the agent conversation(s)
Steve suggests sharing the prompt used to create the document. I agree with him, but don’t think he goes far enough.
I think you should also share the most relevant bits of your conversation with the agent along with the document. The sequence of prompts, how you evaluated what the agent generated, what sources you asked it to use, and how you drew your conclusions all demonstrate your thinking and earn your colleagues’ attention.
It could also create an environment where you are forced to stand behind what you put in a document, and set cultural expectations for more thoughtful use of AI.
Most importantly, it discourages meat proxy behavior on both sides — the author using AI to cheaply produce a document, and their colleagues using it to cheaply summarize it. This is the most dangerous failure mode in my opinion, because it leads to the complete elimination of understanding and critical thinking across your company.
2. Create higher-fidelity artifacts
As a marketer, it’s easier than ever for me to express my ideas in the medium my audience will experience them. I am no longer constrained by the written word! For example, instead of working on web copy in a Google doc, I can just start with a mock website. Instead of writing a brief to explore a concept for a social ad or product demo, I can just generate it in code, design, and even motion.
This forces me to think through decisions that a doc allows me to postpone. How do the words and visuals work together? Does the message still make sense when there’s only room for one sentence? It also makes discussions with my creative colleagues more productive and helps us move faster, because we have something concrete to start with.
This does require marketers to learn new skills: become more visually literate, understand the basics of motion and video, and learn enough of the language of design to describe what they want (I’m working on this myself). But if AI makes it easier to produce an artifact, we should raise our expectations for what that artifact helps us understand.
This may not work immediately for every function (and may never work for the less creative aspects of marketing, like planning), but perhaps it’s time we seriously considered killing the doc?
Fighting AI;DR at every turn
I expect there to be a lot more discussion about this in the coming months, and my own thinking to keep evolving. Perhaps collaboration tools and business software will grow to meet this need. But I’m convinced the most effective defense against slop grenades is ultimately cultural.
The best companies have always set a high bar for their work. That expectation should hold regardless of how the work gets done. The only lasting defense against AI;DR setting in is a culture where everyone stands behind what they produce and treats their colleagues’ attention as something to be earned.
All views and opinions are my own and do not represent those of my employer.