0.0 Context Setting

Thursday, 6 August 2026 in Portland, Oregon. It’s been 2 days and 29 years since Skynet was turned on, and we only have about three weeks or so until it gains sentience and all goes kaplooey.

0.1 Events: How People Work, Live!

I’m afraid I’ve had to postpone How People Work, Live! ... with Dylan Wilbanks this week. We’ll schedule it for another time. That means the next ones coming up are:

006 How People Work, Live! ... with weaver (register here!), on Friday 14 August. Weaver is a partner at Layer Aleph, and co-author with Marina Nitze and Mikey Dickerson of Crisis Engineering.

(Yes, show 007 is missing. I should have something to announce soon.)

008 How People Work, Live! ... with Margo Stern (register here!) on Friday 28 August. That’s Margo Stern, the author of Good Work: Managing Yourself and Managing Managing, a two-sided book that shares all the tips, tricks, hacks and cheat codes to working in tech. Drawing not only on her experience managing and being an IC at places like Google, Twitter, and Lyft, she shares the wisdom from interviews with PMs, Designers and Engineers who have been there too. Margo is also a Staff Content Designer at Lyft, and she lives in San Francisco, CA.

The show page has videos of previous shows and upcoming ones, too.

Honestly, I’m thinking about moving the show from a register-for-free-and-attend-on-zoom model to just... streaming them, and having people turn up in chat? For something like that, I’d use stream.place and take an ideologically open stance. But then I’d also have to figure out video archival and yadda yadda.

Anyway. Come along to one of the shows or watch one if you’d like to find out about how I approach my work.

1.0 Some Things That Caught My Attention

1.1 Tools are easy now, data is hard

I think this one takes a while to get there, sorry.

I was talking with Dorian the other day about Intertwingler, a piece of software he’s been working on for quite a while now. In Dorian’s words, Intertwingler is:

Intertwingler is an engine for creating dense hypermedia experiences. By dense hypermedia, I mean in contrast to sparse hypermedia, which is what the Web provides out of the box. Intertwingler helps you, the hypertext author, by managing an otherwise unmanageably large number of arbitrarily small, reusable, formally typed, structured data objects, connected together by an unbreakable thicket of typed, bidirectional links. It also maps these objects between a set of durable canonical identifiers—​designed to last forever—​and the friendlier, albeit more mutable kind of typical Web URLs we’re all used to seeing.¹

That’s.. a dense explanation, which I imagine might have totally airplaned over a number of heads. I’ll try, but any errors are mine:

Intertwingler is a way to store data in a way that you can create connections between that data, and then present that data to someone.

The way we were talking about it and Dorian was explaining it to me, I was trying to squish it down and explain it to a mid-level manager or explain to an executive what problem it might solve.

That was pretty difficult! People don’t really make decisions on data, they make decisions on vibes. Those vibes themselves are sure based on and influenced by data, but in my experience not the kind of data produced or displayed by a data dashboard. One problem with a data dashboard is that it kind of begs you to put more data on it, because a dashboard that’s just one big number feels slightly inadequate and unimportant. But that might be all you need, in many cases! It depends, of course, on the decisions you’re going to be making.

Anyway. Dorian’s tool makes it easy for you to record data, and for that data to be strongly typed. That means a date is a date, not just the words that you use to represent a date, which is great because once a computer knows that, it can do things with it. Part of Dorian’s thesis is that it would be great -- I think this is his thesis -- if people could actually document their decisions. If you could say: we made this decision or did this thing based on these facts and, crucially as a piece of hypermedia, link to that actual fact. There’s a whole bunch of other stuff going on here that I’m not going to write about that I can nerd on about in my head.

Anyway. I had thoughts. So much of my job involves squishing things down into, say, decisions that executives need to make. They are often not aware that those decisions need to be made. Often people are obfuscating, hiding, and sometimes that’s because goals aren’t even clear in the first place. But what I know is that most exec leaders don’t actually want all that data. They want to be able to trust that several layers down, through effective delegation, people are qualified to do the job of making sure the right data is collected and the right decisions are being made. Roughly. So no, execs don’t generally, I think want to use something that Intertwingler would do.

(Here’s another attempt. Imagine Wikipedia as much, much, much more of a database that you could get for free. Every single word could be a concept if needed, with meaning that would allow you to link it to something else)

Dorian wants to know why this doesn’t exist because he’d use it (he does), and so has also decided to build it. I think that it’s because people like Dorian are not, well, there’s not a lot of them. But Dorian also thinks that the reason why this tool doesn’t exist, or at least the way of working with data doesn’t exist is because it’s too damn hard to enter it in the first place. I think? Dorian will let me know if I’m wrong.

I was trying to think about why -- or whether -- Dorian’s thing is useful. Usefulness in this world leads to value leads to money, and it’s not like money doesn’t help. Anyway.

I think people want certainty. Or rather, I think you could persuade people that they want certainty.

Things run on vibes, sure. GenAI can make you anything, people want to know what’s real (apart from when they don’t).

So the framing for Intertwingler isn’t that it’s a sort of app that lets you put together concepts and connections and gives them durable URLs so you can always address them. I mean that’s cool and awesome and necessary from a technical point of view (well done, Dorian).

No, the framing I think is that it’s a store of facts. They’re strong facts because they’re typed, which sure, requires work. The question after that, after you’ve got the facts, is what you’re going to do with them, why you need them, how you’re going to use them.

In the olden days, you’d spend forever writing middleware for this kind of thing so your Fact and Concept Thing could talk to other tools like, I don’t know, Figma or PowerPoint or Word or Google Docs. Tech people dream about this kind of stuff; one example is the WinFS days of Microsoft Longhorn, when they thought the great follow-up act after Windows XP was going to be structured data so that you could do precisely this. Turns out it was super hard and you vomited schemas all over the place, anyone who’s spent time on this is waiting for me to mention something like RDF.

Anyway. You wouldn’t do that now. Or someone who’s more... voracious wouldn’t do that. What you’d do is you’d stick an MCP layer in front of your Fact and Concept Store and let all the GenAI tools that people are being built into products whether users like it or not, get access to all that strongly-typed data. Once that data is accessible in a reasonably proper way by GenAI tools -- whether they’re ones producing code or other business artifacts -- then you’re off.

Which led me to this off-the-cuff remark on the call that I also off-the-cuff-posted.

Tools are easy now, data is hard⁴

I’m proud of that flash of insight because it stuck in Erin Kissane’s head all day until she came back with:

Tools easier now, data even harder than it was before.⁵

I’m afraid you can’t put the genie back in the Amazon Prime Same Day Genie Delivery Box. Fine, I say can’t. I mean it would be very very difficult (Difficult? This is Mission Impossible, Mr. Hunt. Difficult should be a walk in the park for you), yes we stopped smoking in public places. But you can agentically code tools now. You can make them work. That isn’t going away.

But the data? I’m not talking about training data, I’m talking about the data that the tools operate on. That data, like Erin says, was always hard. Now it is getting harder not just because of the introduction of synthetic data (I mean, sure it’s a fact that that data was generated, but is it grounded?), but because of the overall environment. Kissane linked to an MIT Tech Review piece on weather forecasting⁶. Writing the code is easier. Processing the data is easier -- but kind of? The importance of high quality data upon which to compute has never been more important.

In that way, I thought Intertwingler has value. It’s a way to preserve data. It’s a way to make it accessible. Sure it all depends on what you put in there. But putting it in there in the first place was part of the problem, and maybe making data easier to use might make that part easier, too.

1.2 Background Culture

Something Nitish Pahwa said about a Guardian piece, Demise of the monoculture: why so many have never heard UK’s record breaking No 1 single², caught my attention. Here’s what Nitish said:

another element of this that I think is relevant: people passively listen all the time and never register what it is that's playing—not even the basic melody or chord progressions, etc. I would not be surprised if there are quite a few folks who've heard this song but never "heard" it, as it were³

Background culture. That’s what I thought. Sure, the Guardian is talking about the demise of a shared culture, or the extreme version of that, a monoculture. But there’s something in the idea I think about continual, persistent culture. Living in a miasma of media. So. The replacement--not wholesale of course, people still talk about Things at the same time--but the encroachment of Background Culture on what we would normally just think of as culture.

Background culture isn’t active. It’s not, uh, foregrounded. But the floor of what it can be has been raised just because of ease of access and ease of production. People used to be -- are -- derisive of elevator music as a genre, but now you can talk about what turns something into elevator music, into that background. There is no cost to playing a video in the background, there is no cost to listening to music in the background, and yet because of data ha ha it’s all tracked. There’s no tracking (thank god) of intentionality, only (very) weak inference based on guessing from other signals.

Background culture!

1.3 Tingly is a decision I have to live with

There’s a whole thing where people can go their whole lives without realizing that they have an oral allergy or sensitivity. Like: oh no, I couldn’t possibly accept your banana, they make my mouth all tingly. And then: Wait, bananas aren’t supposed to do that.

So I was posting about it and then thought: you know what, someone should make a thing where you can vote on whether a fruit makes your mouth tingly and maybe that would help people discover sensitivities they might not otherwise. And I would normally leave it.

Long story short, I wrote a PRD⁷ and Claude Code on Medium one-shotted Spicy, which is a website that... does that? Three hours from Bluesky post⁸ to PRD to live site⁹.

I say one-shotted it. It made something that worked great/fine without any input APART from the fact that the entire copy every single place on the site was abysmal. Like, the worst slop. All the tropes. So I rewrote it all.

I like to think I wrote a pretty good PRD for it to work as a one-shot.

I’ve had to think a lot lately about my position on AI. Did a few job interviews. Like I said above, absent a lot of effort, agentic coding is here and it’s not going away. There are people who want to spend gobs of money on it; I have friends who have done genuinely useful work with it, producing things that are genuinely useful to other people. This isn’t... a defense? The stance I’ve come to is that this is a personal decision, and I could talk about it any which way. I mean, I could say: look, the amount I’m using it is tiny, it’s probably the equivalent of flying? I could say: the datacenters are going to be built anyway, if there were one less active user. I could say: we switched to an electric car, so that offset the whole thing [sic]. I could say the way to defeat this stuff is at the ballot box, and I sincerely believe that.

But while it’s here, I work on harm minimization and want to understand it.

I am not entirely joking when I make the offhand remark that in the way that land acknowledgements are made, AI products could/should have data acknowledgements before you use them. That, of course, people would just click past.

I know how to build the tingly thing. I know, in principle, how to knock it up with Django. I would never have done it. Yeah, sure it’s a power thing. But at the same time I’m careful to say I didn’t make it, because I didn’t.

I mean the thing runs on a home server. It’s not using a model or anything. You can just vote on whether you found something spicy or not. (The admin page does a ping when a vote is recorded, it’s ace).

1.4 Uniqueness Index

That said! A while back -- less than a year -- I gave some models access to all of my newsletter writing and tried to see if it could replicate my tone of voice. They cannot! I’ve decided to take that as proof that for now I cannot be replaced, at least on some axis.

(You could build a terrible thing out of this. Give a product a corpus of your writing, get a model to generate writing based on that corpus, get another model to rate the similarity of the generated text to the original human writing.)

That score? That’s your replacement score. Score above one and you’re safe! See, it’s terrible. I expect it to be launched next week. (God, one could one-shot it)

I like to think that my brain is sufficiently complex that it is indistinguishable from magic and thus cannot be simulated at this point in time.

Anyway. That’s nice and terrible.

--

That’s it for today. How have you been? I love getting notes from people and always try to reply, even when the notes are just “hi!”.

Think you know someone who needs my help? You can find out more about what I do and send them my way.

Best,

Dan