0.0 Context Setting

Wednesday, June 24 2026 in Portland, Oregon where the high today is 86f/30c and the high in the U.K. was 97f/36.1c, and yesterday 40 drowning deaths were reported in France over the past week¹. So it’s time to re-read Kim Stanley Robinson’s The Ministry for the Future, or rather the book’s first chapter. Which I intend to never read again anyway, for the right reasons.

Events are on hiatus until I get scheduling sorted out.

In the meantime, let’s get on with it.

1.0 Things That Caught My Attention

1.1 Anything LLMs Can Do, I Can Do Better

I probably can’t do it backwards and in high heels though.

I have a vested interest in this one. Over the last year I’d put together a lot of how I work into the form of a workshop for a client. I’ve written about the workshop here before and how it helps people work together as teams. It’s great: the cohorts loved it, I love teaching it, and we spend a bunch of time together working on real problems as they’re experiencing them.

I think the workshops are valuable! (Trust me, this is not a shill. There is a point to this). They work through how you can develop better relationships, how you can increase your influence, how you can communicate more clearly, and how you can use strategy better to achieve your goals. Again, all boring business words that honestly only really make sense when they’re applied to something real and not abstract.

I felt quite pleased with myself in calling the set of workshops How People Work because, well, it’s in the name.

And then this whole AI thing happened? One way of reading this is that a while ago it might have been a good idea to spend some money on helping your teams work more effectively but now at least half the prevailing narrative is “fire as many people as you can and get the rest to use AI”. Or use AI as an excuse to correct for any over-hiring you’ve been doing.

The thing I’ve written about before here is that LLMs are great at generating writing that’s average. And most people have trouble writing to an average level -- writing is hard! And in thinking about what I do with teams (and with individuals, when I coach them too) is go really, really deep into how you develop and tend those relationships and how you pay a lot of attention into how you communicate and get your ideas across.

In other words, we spend a lot of time working on theory of mind and figuring out how to empathize with the people you’re working with or the people you need to communicate with. This may be my carbon-based chauvinism kicking in, but I really do think that generative AI and LLMs just aren’t good at this right now. The latest models -- the ones that you’re allowed to use, at any rate -- might have over 10 trillion parameters, but I’ll contend there’s a qualitative difference between a bunch of maths and statistics that is undeniably sufficiently-advanced-technology-that-looks-like-magic and setting out to understand your audience and think through what’s going to break through to them and their understanding.

Look: in one of the early sessions on relationship building I talk about what I do when I start working with a new client is a bunch of research. That’s not just the corporate entity or department or company or whatever, it’s also the person I’m going to be working with. To me, figuring out how to communicate well -- work well -- with them involves knowing as much about them as possible, which means looking up their professional history and their educational history. To me, what they studied in college is useful information! It helps me understand their background and the context of how they might see the world. All of which is to say that none of what I learn or figure out is always correct, but they always form hypotheses that can be disproven if I find out otherwise.

It can feel a bit like doing a cold reading. I can meet with someone for the first time and say: hey, as part of prepping for this, I’ve been thinking about the position we’re in here and I want to check my understanding, so stop me when I get anything wrong, OK? And more often than not I can be doing something like telling the story of how this particular client got to this particular situation and at least 3/4 of it is correct. Part of the reason for this is that there are a bunch of standard patterns: one being that if you happen to be good at something, there’s a chance you’re going to be promoted for it even if that has nothing to do with your job.

For example, I meet so many program managers or directors in government who have accidentally amassed responsibility for technology not because they have any formal education or training in it, not because it’s even been part of their job description, but simply because they care about getting their job done and they’ve understood that getting the job done involves technology, so they’re going to damned well track down who it is they need to work with and figure out how they can make whatever technology or software is critical to their job work better.

At a fundamental level, doing this type of research to understand someone before you start working with someone can be seen as being interested. (It might not be helpful for you to facetiously describe it as internet stalking because in general people don’t like being reminded about how much it’s possible to find out about them online).

You need to understand someone. You could call it having a model of them, you could call it theory of mind. But you need to have enough to imagine what it would be like to be them (hopefully you are not Thomas Nagel and your coworker is not a bat) -- and that isn’t just their professional and educational history. That’s also their emotional state. That’s how stressful the current project is. That’s the network of relationships they’re enmeshed in in their work, who they report to, who reports to them, what the market is doing if there’s a relevant market. Imagine what it is like to be them so you have the best chance of being understood, and then pay attention to whether they actually understand you or not.

I won’t say LLMs can’t do that because one should never say never, and I also won’t say LLMs can’t do that because anything and anyone can do a job shittily. And I’m perfectly happy to admit that a shitty job done by an LLM can easily be better than no job done at all. Or that a shitty job done intentionally by an LLM can be better than a job done unconsciously or unintentionally.

What I will say is that those 10+ trillion parameter models and their 1 million token context windows (Anthropic reckons that’s about 2,500 pages of text) just don’t have enough space and nuance to do the job of understanding a particular person as well as a human can, never mind a human who’s been taught how to do it and is motivated to do it.

This is just a much longer way of saying: yes, you could get Claude Cowork to draft that presentation for you and then yes you could get Cowork to draft the email about the presentation for you and I’m sure they would be perfectly fine. But they’d be fine in the sense of being anodyne and just getting the job done. They would be average and they would work on an average audience.

Sometimes, though, you know that you are not working with an average audience. And other times, though, you will know that a particular job requires a little bit more effort. And it would be nice to know how to do that and what goes into it. Hell, I’m also perfectly happy to admit that taking the notes of the workshop and turning that into a system prompt for Claude would be better than not doing that and just one-shotting a presentation. Some structure is better than no structure! (In which case, I totally recommend signing up for a workshop or coaching just so you can take the notes and tell Claude to do what I’m telling you to do. Makes no difference to me, and I think you’ll learn something anyway?)

This whole newsletter episode came about because I was thinking about my workshops and coaching and feeling a) a bit sad about them because AI everywhere and b) is there anything I can possibly do about it. And I think there is?

At the very least I needed to be able to make the case to myself that there’s a point to what I’m teaching and that what I’m teaching is useful in an age where “just tell Claude to do it and I don’t even need to approve it” isn’t just an option but one that might be actively encouraged by management. I am here to persuade myself (and you) that great, now the rest of your work is done better and you can work with other people better, now you can concentrate more on this important piece of work and make that even better while Claude handles all of your dirty work writing email that other people will use Claude to summarize. Sorry, not sorry.

ANYWAY ANYWAY, you can find out more about the workshops here: How People Work, by Very Little Gravitas.

1.2 Some Other Smaller Things That Caught My Attention

Nailed It

Bullet-point list this time, because I’m remembering the whole point of this newsletter is that it’s me talking to myself and my stream of consciousness that, for some reason, all of you lot find interesting.

• I asked Claude (yes) to proofread a draft of this just to catch any typos. When I do that, I also ask it not to comment on the substance; that’s not what I’m looking for. Anyway, I got this back:

"The em-dash usage is inconsistent (some with spaces, some without), though this may be intentional voice"

In which case I thought I pretty much nailed it, and threaded the needle of “a human being who uses em-dashes and uses them inconsistently in an imperfect way, distinguishing themselves from horrific LLM em-dash usage”.

Trusted Computing Is Hard, Let’s Go Shopping

See, the whole point of app stores is that because computers are fantastically powerful you need to make sure that you don’t run software counter to your interests, and checking that is a Difficult Job You Totally Want To Outsource To Someone Else. I mean there’s other reasons too like if your TV is actually a computer then you want it to work like a TV, and if one of the apps you installed on it is doing something else like, uh, operating as a residential proxy² so a research or marketing firm can access the internet and pretend to be you, then that might slow down the TV-ness of your TV!

Fortunately app store gatekeepers will prevent stuff like that. You wouldn’t want your store full of low-quality slop, presumably you’d have to enforce some sort of rules³ to make sure that didn’t happen.

ANYWAY what I was going to say about this was that it could totally rule if someone used, say, Mythos to root every single internet connected consumer television and then hijack the signal (uh, the packets, I guess) for a Max Headroom-style intervention⁴.

That’s it for now. There’s a lot going on! How’ve you been doing?

Best,

Dan