joshstrange

And it doesn't work if you are signed into multiple Google accounts :/

Today I wanted to fix some mixed formatting in a Google Doc. Ok, I'll use Gemini, it's right there and surely it's the best. Nope, fails twice "Please try again" then suggests just trash formatting. Ok, fine, I connected this to ChatGPT, let's try that. Nope, multiple rounds of "I can't see that document" despite having the plugin configured. Ok, onto Claude... Nope, can't do it (even with a connector setup) but suggests I use this. Ok, sounds good, I'll install the plugin to Google Docs, login to Claude in the new sidebar, and.... "I can't read the document". I give up.

LLMs have flipped the build vs buy debate on its head for me. I also have been re-evaluating existing tools and building my own custom, just-for-me versions and it's been liberating. Using off-the-shelf software increasingly feels stifling and limiting. I do wonder if the future for many SaaS companies is to provide a solid core with a nice API/MCP/Hooks/etc to build on top of so you can add features only your company cares about instead of every SaaS trying to be a one-size-fits-all.

The cool part is building smaller tools to solve your problems also allows you link those all together (MCP/CLI/API) in ways that weren't possible in the past using LLMs as a glue layer.

I thought for sure you might be one of my coworkers based on this comment, we do pretty much the same thing with our “LLM Review” tool. It tries its best to determine what needs human eyes (mostly through file paths and the model’s own “discretion” for changes outside the “important” paths).

I've used Dots and Duet now (not planning to touch Muse/Grok) and while I want Duet to be better, Dots is the better product overall right now.

After setting up Duet it suggested things it could help me with. The things it suggested were "ok" but not great suggestions IMHO. I let it go to see what it would do and I hated the UI/UX within seconds.

It started "working" and dumping "Cards" in the chat of the tickets it was looking up in my ticketing system. Completely useless to me and distracting and without the context of "why" it was very annoying. Don't throw your tool use in my face, I don't care, do the work and come back with solutions. It was even worse than normal Claude chat where the tool calls are collapsed. I have to assume this was done to show it "working", well I hated it.

After churning for a bit with very confusing UI [0] it's findings were not even "lackluster", just completely useless. I have annual performance reviews I need to review and it offered to help. I didn't expect anything useful and didn't plan on using anything it did (I do performance reviews by hand for obvious reasons) but I wanted to see how it'd do. Well, it didn't have access to the most important parts of the data but instead of asking or realizing it had incomplete data it tried to write reviews from scratch using my Obsidian notes. Good idea I guess? But it ignored the employee-filled-out review and the layout I need to use. Again, it didn't have access but instead of asking "is there a form/rubric/etc" it just forged ahead in an effort to show its usefulness which only highlighted its incompetence.

The other 2 tasks it tried to help with included a failed "get ready for product launch" document that had very little to do with what we are planning to launch, complete throwaway, and it suggested scheduling a meeting with an email spammer. "XXXX Says they're free to meet but you haven't set a time yet" - and I never will. PS: HN spamming has gotten a little out of control "I saw your comment on <LLM generated> and I'd love to talk to you about how <insert product I'm shilling for> can help!".

So yeah, a pretty terrible start to Duet. To be fair, Dots was a trainwreck when I tried to use it to write code. I don't love codex but Dots+coding was like the dumbest codex with less control over what it's doing (or more often, not doing). Really, I cannot stress enough how much Dots sucks at writing/overseeing-writing code. But the sweet spot for Dots was monitoring communications and raising things I might have missed, like emails/slack messages that slipped through the cracks or were filtered out but actually were something I needed to look at.

[0] https://cs.joshstrange.com/W2VqC23v It did this 6 more times... completely useless.

I’ll probably be trialing it and Silo over the next little bit. Thanks!

Do you have external users (outside your network)? My understanding (and the last time I tried) is that Jellyfin lacks in this department quite a bit. That and ubiquitous clients.

Plex client is available almost everywhere and overall, Plex mostly "just works". The latest Apple TV client changes have me hot and there are things I don't love but that's what's kept me in the Plex ecosystem. I am looking around now though since I'm so disappointed with the new clients/UI.

I don't want to be a jerk and I know engineers writing the code don't always make the decisions but the new Apple TV Player was dog shit. Just horrible. 10 steps back in UI/UX and frustrating to use. I've been using Plex for well over a decade and to throw that all away on that new player... Just damn.

Not sure why Plex laid off the team, was it the response to the new updates (doubt it), was it "AI", are they planning on going in a new direction?

Yeah, I can't believe how bad the UI/UX is for the new Apple TV Plex app. I switched to Plezy as a client (still Plex server) but don't love it either. I really did like the old Plex UI on ATV.

I don't know what I'll do going forward. Jellyfin is rough (I've tried!) and especially for external users (family).

I found Dots to be both worse than I expected and better. Worse in every way when asking it to do anything coding related (just kept running into issue after issue) but very nice as a "Monitor my communication channels for me" tool.

It's surfaced a few buried/filtered email messages that actually were important and I like having a second set of eyes watching incoming messages.

That said, for only that task, I could probably make my own tool I just haven't had the time to play with it.

Looking forward to checking out Duet once it launches as I prefer Anthropic models over OpenAI and that's where I already have a bigger subscription.

> It is the maxim of every prudent master of a family, never to attempt to make at home what it will cost him more to make than to buy.

Ha! I had this exact quote on a slide I presented last week on this very subject (internally at my company).

I enjoyed finding a number of older references that I could apply to LLMs and this was one of the best ones.

So rich to have them post this when just this morning I had to switch to GLM 5.3 so I could mitigate the PHP CVEs (both Claude and Codex balked at the early investigation work). Once I was further along I was able to switch back to Claude to implement the fix but it refused to look at any POC's or test them against my dev servers, so freaking annoying.

The only things worse than the CVE getting dropped in our laps with little warning was having Claude run into its safeguards which slowed my progress in actually protecting our customers.

Last week Claude (and Codex) also refused to help write unit tests around existing code because it touched on authentication. I was able to ladder between them past everywhere they got stuck but today I went straight to GLM after hitting safeguards.

Yep, had to switch to GLM 5.3 to help protect our platform. So annoying. Claude and Codex crashed out.

This assumes way too much about your average consumer. Most of them from talking to them seem to like their smart TVs.

For myself, I buy the smart TV and then just never connect it to the internet. Mostly because commercial digital signage is either way more expensive or is worse technology (and also more expensive).

> They could always take the bold step of listening to what their customers want

Customers don't know what they want. Sorry but it's just true. They are anti-AI because of a multitude of reasons, only some of those reasons are based in reality. Most of them will give up their anti-AI stance once the dust settles and they will not reward the creators who "stuck to their guns".

It's the same way customers beg restaurants for healthy options then when the restaurant puts those items on the menu they don't sell. Right now mob mentality might sink a game that people think AI was involved in but give it a year and they won't care.

Thinking that not using AI will be a winning play (today) is just fighting the last war.

I'm sure this is a logical facility, so I apologize, I don't have an exact example but I imagine most developers are not eager to stick their hands up to say "Hey, I use AI!" given how consumers have treated games that they think use AI.

I beat Jev and only failed on the dart board, I have zero idea how to read a dart board.

I really like small indie games, sometimes there just aren't enough reviews or the reviews are inconclusive. Sometimes when I read reviews that are negative they're actually something that I _do_ want in a game, and so it's actually a sign that I do want to play that. So you can't just trust the review #/rating alone. And they're probably being gamed as well.

What Steam users want is a way to filter out crap but crap existed well before AI (trust me, I've spent a _lot_ of time scrolling through games on Steam both for "Fests" and just in general). And unfortunately, there isn't an easy way to filter "crap" out. There as good games that use AI, there are bad games that use AI, there are good games that don't use AI, there are bad games that don't use AI. I don't think "AI" is a good filter.

Thankfully Steam has easy refunds if you don't like a game or it doesn't live up to it's marketing page.

I don't plan to release it as it part of an overall "media health" tool I built but you should be able to vibe this pretty easily. I have mine using Whisper to check episodes at intervals throughout to compare against this subtitles and then shift them as needed.

One thing to watch for, make sure it never tries to 2 subtitles at the same time or they will stack in a weird way. It's easy to avoid that but my first pass on a few episodes screwed up the subtitles.

Also, you almost certainly want to extract the subtitles for this work, or at least that's what I did. A combo of pulling out the subtitles in the video file and using OpenSubtitles (VIP) to clean up my media.

Agreed, clay/modeling it is a good analogy. I keep trying to describe the level I work at now (if only to myself) and all I could conjure was shaping the product at a higher level with less control overall (unless you want to drop down and do some "detail work") but the ability to make sweeping changes much easier.

The Apple TV UI change drove me crazy. I switched to Plezy but it’s still not quite what I want. The good news is, it’s open source, so I can modify it if I want to, and it supports Plex and jellyfin, which would make switching to jellyfin in the future easier.

Here are some of the things I’ve build recently, all currently private but mostly because I’m scratching my own itch

- Macros: MyFitnessPal replacement, track my macros, meals/presets/custom food items, also pulls from OpenFoodFacts. MCP support so I can tell an agent “Log my normal breakfast” or “log the sweet potato beef meal for dinner with 2 tortillas”

- Foreman: My software factory (https://joshstrange.com/2026/10/12/there-are-many-like-it-bu...). Every one of these is highly custom/personal so it probably won’t be interesting for me to go into a lot of detail. But suffice to say, I am successfully shipping code through my factory for both my side business and my day job.

- Quartermaster: tool to scrape (via a browser extension) my purchase history at various retailers, including Amazon and Kroger (grocery store). Small web UI to view/filter the data, MCP to interact. I can see purchase price over time, quantity over time, and even things like “price per TB over time” for things like hard drives due to a tagging system that lets me slice and dice the data however I want.

- Vendors: a small web app for my mom to track the vendors she recommends (she’s a real estate agent) to clients so that when they ask for a painter, she can go filter down to the painters, select the people she wants to recommend, and copy all of their contact information into a text/email.

- Amdi: my catch-all app/platform for things small enough to not warrant (yet) their own project. This contains my own little chat interface that hooks into all the data sources above and provides rich UI for certain operations. It’s incredibly custom to me and how I think/work. My favorite features are all around my plex library, letting me cut a clip, find a clip (smart subtitle search), find out what else I’ve seen an actor in for the show I’m watching (using my watch history), and a bunch of metadata around my media exposed to the LLM.

100% agree, that’s the premise of a post I wrote recently [0]. I think for the foreseeable future these factories will be incredibly custom to your company’s workflow and it will take a while for a “one size fits all” to emerge if it ever does. Does an “open core” end up “winning” that every company augments (at the code, not just config) level to suit their needs?

[0] https://joshstrange.com/2026/10/12/there-are-many-like-it-bu...

I am successfully shipping code on side projects/business (real side business with real revenue) through my software factory all the way to production and shipping code all the way to PR within my day job. We aren’t ready to fully pull the trigger and go all-in, but my day job is mostly on the same page that this or something like it is the future.

Well, I’m building one for work and for my side-business (both real companies, making real money, with established processes, albeit different magnitudes/scales) and I disagree with this.

My prediction is that, at least for a while, software factories will be incredibly custom, and might always be that way. Perhaps there will be some sort of open core software factory that then companies augment as needed, but I doubt we’ll have a one-size-fits-all. I could be completely wrong. Nine months ago, I thought software factories were silly, so don’t take my word for it.

I wrote about this recently here: https://joshstrange.com/2026/10/12/there-are-many-like-it-bu...

Work provides a stipend/reimbursement of $100, the rest I pay for.

It’s paid for itself though, at least from my side business income.

Without this change I hated Sol and was burning tokens way faster due to all the forced compacting.

Since making this (obscure/hidden) change I enjoy Codex much more.

Not OP and while I’m not draining my accounts in a single day/hour, I’m hitting limits with a $200-Claude and $100-OpenAI account.

> Have you shipping anything?

Yes, hundreds of PRs and shipped new features to 3 different platforms (day job and 2 side-business apps/platforms I’ve been developing for many years, pre-LLM). Those new features have gone out to customers and been used.

> Do you have income from this?

Well my day job for one, and I’ve been able to more easily sell to new clients with my side business. Without the changes I’ve shipped I would have been able to land a couple of the clients I’ve signed on recently.

I don’t think OP’s message implied they are never finishing a project, maybe they edited it?

In addition to the 3 income-generating projects, I have a number of other “just for me” projects. Some are mature now and only need occasional development when I want a new feature. Some are in flux as I build them out (like my “software factory”).

At this point I’m limited by compute and tokens more than anything else.

I’ll admit to being a “fanboy” (though it’s waned in recent years). This is one of the very few “splurges” or “financially irresponsible” purchases I make and I do it because it’s fun. That’s it. I enjoy trying out the latest phone even if the changes are minor. Some people gamble or spend their money frivolously in other ways. This is mine.

I wasn’t trying to whine or otherwise pretend my day was ruined. It was 100% “aww shucks, the streak that really meant nothing to me was broken”. I just wanted to add a data point to the conversation and I apologize if it came off like I whining or felt entitled or anything like that. I don’t believe I’m “owed” a phone on release day and waiting 2 weeks isn’t a big deal. I do enjoy following the coverage and trying out things mentioned, that’s part of the “fun”, but I can wait 2 weeks and I’ll be just fine (obviously).

> out of touch and over-the-top fan-boyish

> privileged whining

Uh, are you sure we are talking about the same comment?

I just said I wished they had a better system and gave a data point.

I didn't pretend like it was the end of the world, I was just trying to provide information on the demand for the phone (8:02 and already deliveries were slipping) and I felt like adding the bit about "16 years" helped show that even someone who has been through this process and know how to do it well from lots of practice wasn't able to get the phone.