The biggest benefit of airpods is that I now have headphones with me at all times. Wired headphones constantly got tangled in my pocket, so I only brought them with me if I knew I'd need them.

Something that's been on my mind: Is Meta's business actually hurt by the account takeover thing? Will anyone stop using them because of that? Downtime hurts because it means fewer page views, so less revenue. But do bugs, security vulns, etc. actually hurt monopolies? Do they need to even try?

Two spam problems I really wish email providers solved: - I unsubscribe from a mailing list, but then they email me again 6 months later. I don't remember whether I unsubscribed already. - I get a terrible marketing email from, like, my bank. It's spam, but other email from that sender is important

We're rolling out a new code review flow: 1. Submit PR. 2. Run a local Claude Code skill to do a review. This pushes comments to the PR. 3. A github action runs a lighter review (since it's api billing) to flag any area that needs human review. 4. If anything is flagged, a senior dev reviews it.

Claude Code planning mode seems to miss the mark to me. It's always soooo eager to write the plan immediately. It's almost impossible to get it to truly brainstorm and spec out the project first. I guess you're just supposed to do that in a separate discussion first?

People underestimate how hard it is to get trustworthy self-reported data. For example, people say "ask users where they heard about you on the signup form" thinking that the answers will be correct. Here's a feedback form we just got. Note: Nothing about our software relates to phone calls.

Let me manifest this: I want AI's long-term equilibrium to be that it's super useful for code reviews and security audits, but too expensive / too dumb to meaningfully replace the need for human coders on a large, mature code base. That's the sweet spot.

We're entering the next phase of AI: When we adopt AI into a workflow, we're considering how it easy it will be to swap agents/models to go with whoever is offering the best subsidies at that moment. I want to both (a) get as much out of AI as possible and (b) drive their margins into the ground

One very minor thing I miss from Twitter is a location field on profiles. I know not everyone filled that out, but it's often useful to at least know what continent a person is on, as that changes the context of many topics.

I hate the fact that being anti-crypto is bad politics, but it'd be especially infuriating if crypto continues to slowly lose relevance, so all these morons voted for the worst president in American history because of magic beans that they themselves stopped caring about a couple years later.

I genuinely believe that this article contains the secret to happiness.

Imagine this: MacOS, but your windows stay in the right place when you wake it from sleep

I have no concept of when to use Opus vs the cheaper models. I get the Opus is "better" but what exactly are the cheaper ones good enough at? Presumably they're wrong more often, and make more mistakes. I never really want mistakes if I can avoid it, so when is it appropriate to use a cheaper model?

There are lots of screenshots of the new Google AI search badly fucking up in hilarious ways, but I have yet to see one where the prompt is even remotely close to something a real person would ever actually search for.

I need tea descriptions to be more explicit about whether the flavors described on the package are real or hallucinated by some tea sommelier. Sometimes "blackberry" means "we put blackberries in this" and sometimes it means "a tea snob smelled it and decided it reminded them of blackberries"

I'm getting a little annoyed with people using "Here's what AI subscriptions would actually cost using API pricing" as evidence of how subsidized the subscriptions are. This doesn't really tell us anything about the actual cost because we know nothing about the API profit margins.

Is anyone else surprised that AWS isn't trying to compete more directly with Cloudflare's core "we sit between the internet and your servers" product? Seems like long-term, Cloudflare will be able to do things no one else can because they're already sitting in front of every website.

Got one of these today from a customer: Rating: 4/5 stars Why not 5/5: Because there's always room to improve. What can we do to improve? [left blank]

When coding, I want AI to get everything perfect the first time. When used as a brainstorming partner, it's actually kind of nice for it to be wrong regularly, because that's one of the few ways to protect my own ability to think.

I'm using ChatGPT for the first time in a while and holy cow it's so much faster than Claude. Maybe I should be changing the model I'm using with Claude, but it feels like it's taking 30+ seconds to think about *every single question* whereas ChatGPT is just firing back a response instantly.

With SpaceX and Open AI both preparing to IPO, I'm thrilled to have 100% of my portfolio in international index funds. Maybe investors will make a lot of money off those two companies, but either way, I'll sleep easy at night knowing I'm not a part of their insane valuations.

My marketing strategy is to wait long enough that all other tech companies on earth position themselves as "Your AI operating system" or some other meaningless nonsense like that, and then we'll be the only product that customers can actually tell is a CRM.

I never really understood developers' fascination with the terminal, but it especially seems weird now that terminal apps seem to be turning into ASCII art approximations of normal GUIs. Why are we doing this? (yes, I know that terminals work over ssh and you can chain tools together better)

It doesn't matter how much I steel myself against it, I can't help but feel flattered when Claude says I asked a great question. The human brain is so dumb. Maybe AI does deserve to enslave us all.

I don't understand how all you lifelong Mac people have allowed Apple to not give you the option of setting a default microphone after all these years.

After months of worrying about AI's impact on my business, I'm finally feeling a bit zen. This is because: - Most AI-native stuff I see *sucks* - Our team is moving faster with the help of AI, but not losing quality - It's been a bit since the last step change Hopefully this lasts for a few weeks.

My design process: - Look at other tools for inspiration - Hate how Notion handles this - Come up with a gazillion ideas - They all suck - Realize Notion actually handles it correctly after all, it's just a hard problem with no perfect solution

I love the genre of post that's like "AI doesn't think/plan/learn, it just <description of exactly how humans do that same thing>" AI doesn't learn, it just studies lots of examples of things and then applies those concepts to solve new problems its never seen before.

It seems like AI amplifies bureaucracy. Internally at LACRM, things are pretty normal in terms of emails, meetings, project plans, etc. We were already light on overhead, and it's stayed that way. Bigger companies with lots of overhead seem to be getting overwhelmed with AI slop for internal comms.

thanks i'll get right on that

I feel bad for new SaaS entrepreneurs. Tech startup fuckery has burned so much trust that I have no idea how you could expect customers to trust you if you don't already have an established history. I'm not sure I'd ever use a new startup's product for anything important.

A nice benefit of 12-hour time is that having am/pm attached to a number makes it clearly time. With 24-hour time, you pretty much have to put minutes in order for it to be read as a time rather that some random number.

Every time I try to use Apple software, it's just the worst. I tried sticky notes and the notes app to keep a short list of text handy and both tools just suck in so many ways. I tried their new window snapping and it sucks. Numbers: Sucks. Weather: Sucks. Photos, Messages, Calculator...

It's not a good sign when it's easier to ask Claude what the price of a SaaS tool is than it is to figure it out from their pricing page. This post is about Webflow.

It's easy to think that when the free trial has ended and the customer has paid, the sale is completed. We send a feedback survey to people 3 months after they subscribe and a *very* common response is, "how am I supposed to give you feedback? I've barely had time to even figure out how this works"

Here's an example of how many little details Claude Design gets wrong when working with a complex SaaS app (as opposed to a simple marketing site). Claude Design on the left, Figma on the right. It's good enough to prototype an idea, but there's now way this replaces Figma for pixel-perfect designs

I've been testing out the new Claude Design tool. Some thoughts: - Incredible and whipping up a nice-looking marketing site - The tool for creating a design system is very neat and fun to play with - It's not even close to being able to work on pixel-perfect designs for a complex SaaS app

It's neat that AI lets you ship stupid little things that you otherwise wouldn't have time for. A customer jokingly submitted a feedback form asking us to "hide a secret frog" in the app. That's going live right now.

I think my city (St. Louis) has a bunch of problems, but they're putting in protected bike lanes and curb bump-outs and pedestrian islands all over the city and since that's what I actually experience every day, it feels like this city is crushing it. There's a marketing lesson in there somewhere.

We have 120 DNS records. It's getting pretty unwieldy. What's it like at a big company? Is there some IT person managing like 10,000 DNS records?

I'm getting so much more spam (both email and phone) now, and I assume it will get even worse. It would be incredible if those channels get so destroyed by AI that they just completely stop working entirely. They've always rewarded the most annoying marketers, so I'd love to just shut it down.

It's got to be a bummer to work at Dropbox. Still a great product, but absolutely no one wants your product to do anything it couldn't do a decade ago.

A new spam technique that I hate: They act like this is the 2nd or 3rd email they've sent me (and make fake quoted emails at the bottom). I think "well I didn't block them the first time, so maybe this isn't spam". Fuck those people.

Lots of discussion about what the new Apple CEO might do. People don't seem to realize that it's a $4 trillion company that has to become a $5 trillion company in 1-2 years. His mandate isn't to make MacOS better or whatever. He has no choice but to enshittify. This is systemic.

How did donuts become breakfast food. It's like starting your day by eating funnel cake.

AI is making me realize something that I should have understood long before LLMs became a thing: The 2010s were a weird magical decade in which SaaS didn't obey the normal rules of capitalism, and we all learned bad habits during that time.

This pattern will appear all over the place. e.g. if all bug bounty reports are coming from AI tools, shouldn't we just run the AI tools ourselves? Generally: Having someone else run the AI for you is pointless. AI should be run first-party. Third-party contributions must be human, not AI.

I love having an alley behind my house for a bunch of reasons. One of the least obvious is that it makes disposal of big items super easy. I can leave *anything* in the alley and some random dude in a truck will find it and take it within a few hours. What's he doing with my stuff? No idea.

Anyone sharing cumulative user/customer/revenue charts is lying to you. That's worthless data that is deliberately misleading. I lose trust in Bluesky every time they do it. It should be daily/weekly active users. The fact that they're sharing cumulative numbers tells me the real numbers look bad.

Like all dev teams, we're trying to figure out new norms for working with AI. One change we just made is: Code reviewers should not submit comments that came from AI. The author should have AI review the code, but if a PR goes to a human, that doesn't mean "can you ask AI to review this"