Artem (@sl1ma4)

@sl1ma4Aug 5

My architecture flow is in public. You can see it in action here: uptimepage.dev/architecture

373

Artem · Nov 11, 2026 · 7:07 PM UTC

@sl1ma49h

stop letting AI edit uncommitted code. seriously — commit what works — ask for one change — read the diff — good? commit. bad? git reset --hard a bad AI edit should cost you 10 seconds, not the rest of your day

289

Artem · Nov 11, 2026 · 2:03 PM UTC

@sl1ma414h

2,948 followers 👀 52 away from 3k. can we get there today? 🚀 i post about AI coding tips, DevOps life and building uptimepage in public if you're a solo founder, DevOps, SRE, developer or AI builder, let's connect 🤝

ALT Screenshot of the @sl1ma4 X profile with the follower count, 2,948, outlined in red

767

Artem · Nov 11, 2026 · 11:28 AM UTC

@sl1ma417h

stop trusting your /health endpoint. seriously make it actually check things: — run one real query — ping the cache — check the disk isn't full a green check on a dead app is worse than no check at all

228

Artem · Nov 11, 2026 · 5:59 AM UTC

@sl1ma422h

hosted first, then self-host once a team depends on it. that's why uptimepage.dev ships both

Uptimepage: uptime monitoring, status pages and on-call

Hosted uptime monitoring for websites and APIs, with multi-region checks, on-call rotations, incidents, and public status pages. Start free; open source.

uptimepage.dev

Michiel V

@michielmvNov 10

What 12 days of self-hosting boxd taught us On 29 Oct we made @boxd_sh self-hostable: you can run it in your own AWS, GCP or any KVM server, free up to 16 vCPUs. The launch post was bookmarked a lot. Some insights: - There is strong demand for such offering, and it's different from other providers - Teams try boxd on our cloud first days to see how it works, then move to self-hosted - People value the ease of use: a single binary to deploy, cluster.toml -> apply and you're ready - Deal sizes are larger than on managed. Generally bigger companies tend to be interested in self-hosting. Next, the boxd runtime needs to be even more composable. That's what we're working on now. Get it at github.com/boxd-sh/self-host (a ⭐ would be appreciated)

GitHub - boxd-sh/self-host: boxd self-host installer

boxd self-host installer. Contribute to boxd-sh/self-host development by creating an account on GitHub.

github.com

112

Artem · Nov 11, 2026 · 4:23 AM UTC

@sl1ma4Nov 11

asked @FishAudio to build me a senior software engineer to review my code... and it actually made a voice agent for that 😄 now I can argue about my code with AI too 🧐

🤝 Paid partnership

Rissa Cao

@rissa_caoNov 10

Introducing @FishAudio Live 3: a full-duplex voice model that listens, says "mm-hmm," and knows when to shut up. Enterprise? Book a demo for your free custom agent: s.fish.audio/vxhgs6 Reply with your use case - we'll build it free for the first 1,000 people.

293

Artem · Nov 10, 2026 · 9:34 PM UTC

AI labs org charts

Hand-drawn org charts of six AI companies. OpenAI: a dollar sign above a classic top-down hierarchy. Anthropic: teams in locked boxes, with a separate locked safety team watching the top. Google DeepMind: a tangled network of nodes shouting "BENCHMARK!". Meta: separate trees for VR, Instagram, Facebook and WhatsApp with users below. xAI: one man in sunglasses on top of a single chain, with everyone else on one flat line underneath. Microsoft: divisions pointing guns at each other.

ALT Hand-drawn org charts of six AI companies. OpenAI: a dollar sign above a classic top-down hierarchy. Anthropic: teams in locked boxes, with a separate locked safety team watching the top. Google DeepMind: a tangled network of nodes shouting "BENCHMARK!". Meta: separate trees for VR, Instagram, Facebook and WhatsApp with users below. xAI: one man in sunglasses on top of a single chain, with everyone else on one flat line underneath. Microsoft: divisions pointing guns at each other.

Artem · Nov 10, 2026 · 7:06 PM UTC

drop your landing page i'll tell you in one line what i think you sell if i'm wrong, your headline is the problem

108

3,107

Artem · Nov 10, 2026 · 6:13 PM UTC

i set up a table-booking demo with Fish Audio's new Live 3 Duplex model 📞 it asks for the day, time, guest count and name, then reads the details back for confirmation. here's the call with a fictional restaurant.

🤝 Paid partnership

Rissa Cao

@rissa_caoNov 10

427

Artem · Nov 10, 2026 · 4:27 PM UTC

Fable 5.2 costs 2x Opus 5.5 per token $8/$40 vs $4/$20 but per task it can be cheaper 👀 Fable on high scores 45.3% on FrontierCode 2.0 Opus on max scores 45.5% Fable gets there at about half the cost stop comparing token prices. compare cost per finished task

FrontierCode 2.0 accuracy vs cost chart. Fable 5.2 at high effort scores 45.3% at about $6.7 per task. Opus 5.5 at max effort scores 45.5% at about $12.8 per task.

ALT FrontierCode 2.0 accuracy vs cost chart. Fable 5.2 at high effort scores 45.3% at about $6.7 per task. Opus 5.5 at max effort scores 45.5% at about $12.8 per task.

117

Artem · Nov 10, 2026 · 2:04 PM UTC

🚨 i asked GPT-6.1 Sol to judge Fable 5.2 vs Opus 5.5 🤯 same uptimepage feature, 2 specs, both written at xhigh GPT's verdict: Fable's spec is "the better foundation for implementation" Opus's is shorter and cheaper to build, but "leaves more room for misleading advice" what Fable's spec got right: — hides setup tips once a monitor has passed a check. Opus's shows them forever — shows one shared hint only when every region fails the same way — says a firewall "can look like this". Opus's says "usually", with no proof GPT also caught Opus's spec contradicting the actual code and it still found 4 things to fix in the winner 😅 try it on your next spec: one model writes, a rival model reviews

Terminal screenshot of an AI review comparing two specs for an "error hints" feature. It says Spec 1 is the better foundation for implementation, while Spec 2 is shorter and cheaper to implement but leaves more room for misleading advice. A table compares them on scope justification, regional failures, error matching, diagnostic confidence, established outages, UI completeness and implementation cost. Below it, four numbered points to tighten in Spec 1, and a recommendation to keep Spec 1 and resolve those ambiguities.

ALT Terminal screenshot of an AI review comparing two specs for an "error hints" feature. It says Spec 1 is the better foundation for implementation, while Spec 2 is shorter and cheaper to implement but leaves more room for misleading advice. A table compares them on scope justification, regional failures, error matching, diagnostic confidence, established outages, UI completeness and implementation cost. Below it, four numbered points to tighten in Spec 1, and a recommendation to keep Spec 1 and resolve those ambiguities.

270

Artem · Nov 10, 2026 · 11:33 AM UTC

you can choose only one model for the next 30 days 1. Fable 5.2 2. Opus 5.5 3. GPT-6.1 Sol 4. Gemini 4 Argon which one?

468

Artem · Nov 10, 2026 · 4:18 AM UTC

🚨 Claude Fable 5.2 is out — up to 30% faster than 5.1 — 20% cheaper: $8/$40 per 1M tokens — about Opus 5.5-level coding at roughly half the cost per task — Fast Mode: up to 2.5x speed for 2x price — 1–3 Fable limit resets on paid plans the resets are the real headline

Anthropic's benchmark table "Evaluating Claude Fable 5.2 and Mythos 5.2", comparing Fable 5.2 with Fable 5.1, Opus 5.5, GPT-6 Astra and Gemini 4 Carbon. Fable 5.2 leads most rows, including FrontierCode 2.0 (49.1%), FrontierSWE v2 (69.9% vs 62.3% for Opus 5.5), OSWorld 2.1 (89.3%), Remote Labor Index (41.3% vs 17.9% for Fable 5.1) and Humanity's Last Exam with tools (74.2%). Gemini 4 Carbon leads AutomationBench (53.9%) and Opus 5.5 has the top knowledge-work Elo (1800).

ALT Anthropic's benchmark table "Evaluating Claude Fable 5.2 and Mythos 5.2", comparing Fable 5.2 with Fable 5.1, Opus 5.5, GPT-6 Astra and Gemini 4 Carbon. Fable 5.2 leads most rows, including FrontierCode 2.0 (49.1%), FrontierSWE v2 (69.9% vs 62.3% for Opus 5.5), OSWorld 2.1 (89.3%), Remote Labor Index (41.3% vs 17.9% for Fable 5.1) and Humanity's Last Exam with tools (74.2%). Gemini 4 Carbon leads AutomationBench (53.9%) and Opus 5.5 has the top knowledge-work Elo (1800).

373

Artem · Nov 10, 2026 · 2:41 AM UTC

most side projects don't fail they just stop getting opened

812

Artem · Nov 10, 2026 · 12:30 AM UTC

monday. what's the one thing you'll ship this week? 👀

205

Artem · Nov 9, 2026 · 10:21 PM UTC

hey @grok, rank by how much time they waste for a solo founder: meetings social media refactoring picking a tech stack my bet: tech stack first. yours?

309

Artem · Nov 9, 2026 · 8:20 PM UTC

they built their own simulation of X, trained a model on it, and somehow got 383 website visits for $200 vs Grok's 123 😄 the fact they're open-sourcing the model and benchmark makes this even more interesting. going to check this out 🧐

Ted

@notmiramuratiNov 9

Meet Yvligo, our open-weights model for running X ads. (3x better than Grok 4.7) At loglog.ai, we built an X simulation and post-trained DeepSeek V4.1 Flash, outperforming frontier models Live results: Yvligo: 383 relevant website visits for $200 Grok: 123 visits

316

Artem · Nov 9, 2026 · 5:46 PM UTC

Anthropic, ship new models whenever you want but the version we pick does NOT change if it says Opus 5.5, it better be Opus 5.5. every request we pay for a model, not a mystery box

ALT Cartoon unmasking meme. Top: a man pulls the hood off a tied-up ghost labeled "Opus 5.5". Bottom: under the hood is a scowling villain labeled "Opus 5.0".

357

Artem · Nov 9, 2026 · 3:27 PM UTC

hey @grok, if AI writes the code and AI reviews the PR who gets paged at 3am?

176

Artem · Nov 9, 2026 · 1:21 PM UTC