Appearing Productive in The Workplace

A meaningful share of my current role has become sitting across from account directors and go-to-market leads who arrive with AI-generated projects and argue them. What they are proposing, in most cases, is a dashboard or website that displays the status of a process that is not ready to be automated, built to track a workflow that does not yet warrant tracking. The tool has not solved a problem; it has driven its user to identify a problem worth solving, outlined an architecture for the solution, and produced enough material — diagrams, schemas, interface mockups — that the user arrives in the room convinced the work is real. My job is to explain, carefully, why the logic is flawed: why the process they have outlined is either a non-problem or a premature one, why the architecture the model produced is a plausible shape of a system rather than a system plan, why the confidence they feel is the model’s confidence and not their own. They do not take it well. The tool has given them the experience of building something, and the experience feels like expertise…

Requirements documents that were once a page are now twelve. Status updates that were once three sentences are now bulleted summaries of bulleted summaries. Retrospective notes, post-incident reports, design memos, kickoff decks: every artifact that can be elongated is, by people who do not read what they produce, for readers who do not read what they receive. The cost of producing a document has fallen to nearly zero; the cost of reading one has not, and is in fact rising, because the reader must now sift the synthetic context for whatever the document was originally about.

(I decided on just two big pull quotes this time, but seriously just read the entire thing.)

“Can I get the logo in cornflower blue?” is one of my favorite movie quotes to deploy in a work conversation. But other times, I’m in a meeting with so much ignorant bloviating that I have to subtoot:

Product management is the hardest job in software only because everyone thinks they are a f*n software expert even if they’ve never designed a single feature

Actual documentation used to be the safe-from-stupidity domain of actual professionals, people who had the expertise and humility to actually think through problems before proposing solutions. Now, “I read it on the Internet” has bled into our conference rooms and intranets. Take this hypothetical example of an AI-generated internal document at a software company:

URGENT: Platform Enhancement Requests
TO: [ Product Leader ]
FROM: [ Ops Leader ]
DATE: [ The height of the AI bubble ]
PRIORITY: 🔴 High – Immediate Roadmap Consideration Required

Executive Summary

[ Two paragraphs of slop ]

⚠️ These features must be prioritized in the next sprint cycle.

[ ~1000 more words of slop ]

There are other factors involved that would lead someone to generate such a document and immediately share it with their coworkers–their pain is real and should not be dismissed1–but the fact that the AI became the intermediary for this communication simply intensified all the pitfalls of less experienced product thinking:

I use AI to generate documents at work all the time. It often helps me put things into words I hope coworkers will better understand–the same motivation of the “author” of the above example, I assume. But I’ve first used the LLM to think through the problem: I’ve given it instructions to try to make it less sycophantic and get it to challenge me3, I’ve asked it to critique my original thinking, I’ve read its massive outputs, asked it more questions, instructed it very specifically about the document I would like, read that document, asked it to make edits, then made final edits myself. This results in a draft that I then socialize with the relevant and capable coworkers, and we all leave comments and make edits. The robot is then really good at reading all of that, plus the relevant documents and comments from last year we’ve all forgotten about4, and we start the entire process all over again. I use the robot to not just make assumptions. I use it to try to capture all the context. I preface the sharing of these documents with a disclaimer that I used an LLM to generate it.5

The Internet gave everyone a voice to complain (and complain about complaining…see, I’m self-aware), best described in one of my favorite posts of all time, Paul Ford’s 2011 The Web Is a Customer Service Medium. Now AI is making everyone feel and present like an expert without any expertise. Part of me wants to offer them my job. After all, Claude Code can develop their software too, can it not?

Postscript: no sooner had I posted this than I found On Humility in Product Development, another reference to Fight Club about almost the same topic.

  1. The acute pain, and potentially a feeling of helplessness, might have led to prompts that led a sycophantic and very helpful LLM to produce such a document. We have to get through to those Product people! I can hear the robot saying. ↩

  2. In Henry Gantt, I mention Hofstadter’s law, Goodhart’s law, and Campbell’s law. Lately I’ve been saying Conway’s law repeatedly. Conway’s law is essentially the entire point of this post, hidden in a footnote. It deserves it’s own, more thought through post. ↩

  3. I think the jury is still out on whether this is effective, or if it’s just placebo ↩

  4. I often have to ask the robot, “Where did you get this from?” It dutifully points me to the document in question. Well, about ⅓ of the time. Another ⅓ of the time, it admits it hallucinated the source. The final ⅓ of the time, it made a poor correlation and I have to correct it. ↩

  5. I mostly do this to protect my pride: if the robot exposes my ignorance, I want plausible deniability. ↩

Previous: The Output of Belonging