You should be writing longer form tweets. Period. I'm not going to be one of those Twitter Bros with my 850-person account telling you how to grow to be massive, but here's what I've seen. Let me explain... @pauldm tagged me in a comment and asked what I'd seen, so I started writing out my thoughts and rationale as to why I started going down this path. First off, I hate threads. I think the UX is janky as both a reader and someone writing them. I find the constant breaks distracting and disjointed. Your entire writing style tends to change because you're trying to hook and lead the reader in EVERY individual tweet so that they keep reading. So, I don't like reading them and I don't like writing them, and that's the #1 reason I avoid them. Plus, I'm verbose, so I like writing longer form. Now, before I get into any of this, remember, my account is "petite" - it's at 860ish people and was at 325 when I started experimenting with longer-form tweets. Take that into consideration - small sample size. Back in April, I noticed that if I wrote more than 280 characters and people clicked the "Show More" link to read the entire tweet, I got more impressions. That's the genesis of it. I didn't need any advanced calculus to work it out - writing more, telling a longer story, people who were interested clicked "Show More" and Twitter showed the whole tweet to more people it seemed. My hypothesis was that it was what I called "Dwell Time" - the more people read my single tweets, the more "value" Twitter seemed to place on it. Sort of like a silent engagement signal. Obviously clicking on the "Show More" link was proper engagement signal to Twitter that people were interested in what I was writing, so no real need to explain that. This is where I think it gets more interesting. Then I put on my nerd hat and realized that from purely a "cost of compute" resources, it would be cheaper for Twitter to show long-form posts rather than threads in terms of calls to their data and presentation servers. Based on that, I had a hunch a month ago that we'd see threads be pushed down the "algorithmic ranking signals" in favour of longer-form tweets because it's cheaper for Twitter AND it means people dwell on a tweet long. Longer dwell time is more likely to result in some kind of engagement happening - a like, a retweet, or hopefully a reply. I think replies are the magic "GO" button - why do I think that? What is Twitter going to pay creators out for? Ads displayed in the comments on a post. They are financially rewarding you for having people comment on your stuff - I'm gonna go out on a limb and say that this is probably a strong signal that Twitter values this. So as a strategy to complement that, I started replying to almost everything on my tweets - again, being a small account makes that easy, but I actively did it and I told people I was doing it. That creates more conversation, etc... My account had tripled in size off a small base, my tweets will easily get over 1m impressions this month in total, and considering that I went from about 350 followers a month ago to 860something today, that's a fair amount of impressions. The content is the content - it obviously plays a big role, but the strategy for that was pretty simple: "Say the shit out loud that other people are thinking but are too shy or unwilling to say because they don't want the blowback." I decided to just be honest. People talk about "being authentic" and then write rubbish platitudes. I shit post and toss out contrarian views. I don't care if people like me or not. I don't care if people agree with me or not. I wanted to have fun and say what was on my mind while pointing out the absurdity of a lot of what I was seeing based on nearly 20 years of doing stuff like this. Seems to be resonating with other people, who knows. I'd like to say I sat down and mapped this all out, but I didn't - most of it came to me while I was standing in the sun, out for a walk, lifting weights, or sitting on the toilet - the places where I do my best thinking. And without being egotistical at all, I think I have a pretty decent grasp of human psychology, so I tend to know what buttons to press. In summary, why do I think that long-form content is working for me? It's a confluence of things. It's tactics. It's strategy. It's understanding people. It's being willing to call bullshit on things in a fun way without being a crass asshat. And most importantly, it's being willing to try things and see how they work. This is working, you should try it if you're interested. Or don't. I don't care. Keep writing threads.

Terrible week for Anthropic that could get even worse. The whole thing with Peter Steinberger going to OpenAI was just a bad look. If Hegseth deems them a “Supply Chain Risk” as is being threatened, that’s basically a death sentence for the company. Amodei should stop staring into his own navel on podcasts and waxing philosophically and start running his company better.

Been sorting through this Opus 4.6 token consumption issue for the last few days. I'm finding that it desperately wants to read your entire codebase before going into planning mode and then reads MORE of the codebase after. It's the single area where I'm seeing tokens burn - reading the codebase and looking at unnecessary elements.

I would be willing to test that theory if you could spot me a billion or so.

After a full day of using Claude Code with Opus 4.6, the difference feels minimal. Opus 4.5 was already great. But man, 4.6 chugs tokens. My Claude Max account copped a beating today.

I love Claude. I spend a lot of time every day in Claude Code for work. Yesterday I saw some IAC done using Clawdbot at a client. Environment variables hardcoded into the Terraform. No tests. None. The internet is going to be a zoo in the next 6 months.

Good to see the “let me help you build a service business” grift it still alive in the AI era. “I can show you how to build a six-figure per month SEO business in under 6 months!” My guy, you charge $150 for a 1-hr consult call - why don’t you just build that SEO business yourself and make the money? You literally have to consult calls 24 hours per day every day of the month, no sleep, to make $100k. Just build the business you are telling other people about.

The decline of Perplexity's quality should be studied. I basically used it as a replacement for Google for months. It became muscle memory. Now, the quality of everything it sends back to me is dubious. I've now found myself going to Google's AI Mode instead.

If the accountant who runs @Ryanair goads Musk into buying that horrific airline, then that guy deserves whatever massive payout he gets from the acquisition. RyanAir is seven types of hot garbage. If he can get Musk to pay more than $35B for it, he's a marketing genius.

It's incredibly challenging to explain to people how much better Markdown files are for just about every text-based use case. Nobody gets it... until they use it for awhile. And then they understand.

GPT Image 1.5 is much better at following prompts than Nano Banana Pro which seems to have progressively gotten stupider and stupider since its release.

I want Trump and Vance to go after the Eurocrats over their regulation of the internet for nothing other than GDPR and this cookie shit. Those people ruined the internet.

I see the UK and Australian governments are talking about banning Twitter because of Grok's bikini pic nonsense. I just saw a video of female "comedian" Kell Fire asking her audience if anyone wanted to go home with her. People laugh because that's apparently what passes as humour now. One guy says, "Sure." Everyone laughs some more because "comedy" apparently. In a follow-up, she posts that the "VID SOLD!" Yeah, much to the surprise of nobody, she also has an OF and so does he. Imagine that. Shocked Pikachu Face. She goes on to say that he did in fact "Come to my car, come to my place, and then he came on my face." Because, of course, "comedy". But yeah, Twitter is the problem. It's the REAL danger to young people and society at large.

I gave a rough draft of my book that I've written over the last 3-4 months to a female friend of mine to read. I trust her opinion, and I thought she might hate it. She read 110,000 words in 3 days. I'm pretty chuffed.

Most people who espouse vibe coding can't spell security, let alone understand controls and process. I use Claude every day. I'm a Max user. I also have 30 years of programming experience. There are going to be unimaginable failures happening when the Salesforce or Workday Admins with zero actual development background but full access to the API start ripping out integrations. There are "technical" people you ask about problems with Excel or to reset your password on an internal portal. Encouraging these people to "vibe code" solutions is madness, let alone letting Becky from Marketing go crazy with Claude Code.

One of the secret reasons I hopped back on this account days before 2026… I wanted to see all of the “experts” telling everyone how to use “latest thing” to create wealth in 2026. I’ve not been disappointed.

I do AI stuff all day at work. Building agents, designing agentic workflows, blah blah blah. I sat down starting in late October, and on nights and weekends, channeled my inner @BenSettle and wrote a book. Dystopian fiction. 120,000 words at the first draft.

It's genuinely hard to believe how much worse the special effects and CGI were in Star Trek: Picard with respect to the Enterprise-D versus ST:TNG decades ago. Picard looked like the entire show's space sequences were rendered in Unreal Engine 5.

It's called boredom.

Using Claude for refactoring code is pretty awesome. You connect it to your GitHub repo, mark the files you want it to be aware of, and then start interacting. I had this clunky code I wrote that awhile back that scrapes some RSS Feeds and then inserts the items into Notion. I literally cobbled it together piecemeal over the course of a few days a year ago to pull some useful articles out of Google Alerts. Like most C# projects that you just hack together, it's generally one long "program.cs" file and because I was using GitHub to run it with Actions, I had a YAML file as well. The more I added, the more brittle it became, I had hashing URLs in there, I had code that was querying Notion to make sure items were unique before injecting them, I had code that deleted older items or duplicates that snuck past. Literally, this thing had ballooned out to like 3500 LOCs and was routinely taking 14-15 minutes to run as a GitHub action. LOL. I created a Claude Project, linked the Repo, and said, "We're going to refactor this." Throughout two evenings after work (I used up all of my "quota" yesterday for 3.7 Sonnet as a Pro user) we've been able to completely refactor the code. I kind of knew how I wanted it to function but I decided to let Claude have a pass at it without too much prompting. It first told me I needed some "models". I gotta be honest, I'd not really cared or considered that but it was a good small starting point. Then it worked out seven different service classes I could create from the main program file. What was interesting was that it was very focused on "Separation of Concerns". I have to be honest, my refactoring plan was probably going to go down the microservices route so when I challenged Claude it replied, and rightfully IMHO, that the application didn't require that level of refactoring unless it was doing considerably more volume than it could estimate. So anyway, there were a couple of things it was a bit hamfisted with - GitHub Secrets for API Keys, missing Dependency Injections for the service classes it was creating, etc... - but for the most part, it was super impressive. The code is now 90% refactored, runs fully in under 3 minutes as a GitHub Action, so it's 5x faster, and the logging that's been added makes so much more sense. If you're a full-time developer and very meticulous about your architecture, then I imagine that something like GitHub CoPilot is probably pretty good for you, but for someone who doesn't program consistently and knows enough to be dangerous, Claude is exceptional for peer review and refactoring help.

I got around to watching that Randy Quaid video on here... I highly recommend you give it a pass unless you have a good sense of humour and a strong stomach. Never deleting this app.