AI Discourse, Writing, and Me

I’m tired of the discourse. The more I read about how AI disrupts the way we design and build software, the less I want to read and write about it. (Yes, ironically, this is about AI.)

It’s not that there aren’t interesting things happening—there are. Jev, Muse, Grok Bot, Siri AI, continuing revelations about rogue OpenAI swarms. There’s a lot.

However, I think we’ve turned the corner. AI is here to stay, regardless of whether the frontier gets paced or not, or whether data centers continue to proliferate. The models we have today have forever changed how software is made. At this point, I believe the sea changes that have been happening every two to three months have slowed. With each new model release, they might be getting faster, more intelligent, and able to perform longer-running tasks. But the major shifts have already happened. We have agents writing code, creating mockups and prototypes, doing research, and autonomously fixing bugs. Improvements will be more of the same.

For designers and product managers, it’ll be dispatching agents to take on work and then reviewing their work. I already see this on my company’s Slack using the Claude Tag. Everyone in EPD just @mentions Claude, “Fix this” and the agent dutifully does it. How fast a company can build depends on the token budget and human review bottleneck.

So the recent discourse gets more fine-grained, no longer about big shifts, but about tweaks and edges. The same topics seem to come up over and over again: design system + AI, prototyping + AI, should designers code, the bottleneck is the PR, etc. And I recently noticed that the LinkedIn posts, Medium articles, and Substacks just sound…the same.

It all has a sparkle-dusting of Opus or Sol with odd sentence constructions and overused phrasing. You recognize the phrases and words: worth, load-bearing, distinction, the X that does Y.

Mea culpa. I’m guilty too. I’ve been using AI to help me write for over a year. At first it was just a couple paragraphs here and there. But then I built a whole sophisticated writing harness to generate text that sounded like me. The flow was:

• I pick the links
• I add a directional note for my agent
• The agent reads the article
• The agent drafts the link commentary based on my direction
• I tweak
• The agent publishes (tags, slug, related posts, alt text)

I was fanatical about trying to catch anything that didn’t sound like me or sounded too much like AI. Here is a snippet from my very comprehensive review skill:

### Category 3: AI Writing Patterns (MUST-FIX)

10. **Structural tells.** Flag:
    - Antithesis: "It's not X, it's Y" / "The value isn't X; it's Y" — single strongest AI tell. Dan Shipper himself called this out in *The AI Sandwich* ("AI writing that reads like 'It's X not Y'"). Flag every occurrence.
    - **"The [noun] worth [verb]-ing" family** (pundit zoom-out). Flag: "the line worth keeping," "the argument worth having," "the part worth sitting with," "the question worth asking," "the distinction worth drawing." This construction zooms out from a concrete observation to a pundit-style abstraction. The reader expects a specific claim; instead they get a meta-statement about which claim deserves attention. Fix: state the concrete observation directly. "The line worth keeping is X" → "X." or "[Author] is right that X."
    - **"Lands"/"land"/"landing" as verb of arrival.** Flag: "[Author] lands the talk on craft." "That is where I want to land it too." "The piece lands on a different note." Distinct from the dramatic-impact "this lands hard" already flagged below — this one uses "land" to describe an argument's arrival point. Fix: "ends on craft" / "settles on craft" / "the piece closes with."
    - **Cute transition metaphors.** Flag: "from a different angle," "from a different door," "coming at it from another side," "approached from the other end." These present the same observation as a clever shift in vantage. Roger's voice doesn't perform vantage shifts; he just states the observation. Fix: cut the transition phrase; lead with the observation itself.
    - Fragmented rhetorical questions as dramatic punctuation: "The user? Secondary."
    - Dramatic impact phrases: "This lands hard." / "This hits different." / "That last line lands."
    - Clever zinger closers that sound like a copywriter, not Roger
    - Mid-sentence rhetorical question then enthusiastic answer: "And honestly? That's amazing." **Exception:** direct-address conversational turns at the start of a paragraph or before a plain restatement ("Interesting, right?", "You see...", "Wild, isn't it?", "Stop me if you've heard this one before...") are Roger's voice per the style guide (Section 5 — The Direct Address Turn) and should NOT be flagged. The flag targets enthusiastic self-answered questions ("And honestly? That's amazing."), not direct-address turns that invite the reader in.

Building this taught me a lot about agent skills and harnesses. The main harness orchestrated the writing and I was able to fire off 10 posts in a batch. It would dispatch sub-agents with specific jobs like researcher, writer, copy-editor, and reviewer. These were not one-shot drafts, but instead, went through an editorial process where I played the role of managing editor.

The whole workflow took about four hours to run each Sunday. I don’t think I saved much time, if any at all.

More recently I started to tire of the content treadmill. Traffic to this blog is modest at best. My overall goal with writing has never been to build a content business. It’s to have a creative outlet, share what I’m learning, and offer my POV. But it’s so easy to get sucked in and feel like I have to post something, especially when Google rewards regular posting.

I’ve decided to step back and raise the bar for myself and for you, dear readers. There will be fewer posts (still weekly), but hopefully, they’ll be of higher quality with more humanity.

Note: This post was drafted without AI. I had Claude give me feedback but asked it not to write any prose for me. I’m retiring the writing harness, except for the publishing script.