← Tech / AI / IT Monitor Index Tech / AI Generated 2026-05-18 19:31 UTC

Tech / AI / IT Monitor

May 18, 2026 · Based on tweets from the last 24 hours · 142 tweets analyzed · model: ollama-cloud/glm-5.1:cloud

Daily Intelligence Briefing — Tech / AI / IT Monitor

Reporting Period: 2026-05-17 19:00 UTC – 2026-05-18 19:00 UTC

Executive Summary

The past 24 hours saw significant activity in AI agent tooling, with OpenAI's Codex and Nous Research's Hermes Agent both shipping major feature updates that emphasize persistent task execution and multi-agent orchestration. On the open-source front, llama.cpp added MTP support for the Qwen 3.6 family—a notable performance milestone for local AI—and Supertonic 3 launched as a browser-native open-source TTS model. Developer infrastructure discussions surged around the "agentic setup stack" (Tailscale, tmux, private Git), with practitioners arguing that foundational infra matters more than model selection. Hardware and compute constraints remain acute: users with RTX 5090s and DGX Sparks still report being compute-constrained, while data-center water-cooling innovations from Jane Street and Hudson River Trading drew attention for their energy-efficiency implications.

Key Events

Analysis

Patterns & Trends:

  1. The Agentic Infrastructure Layer is crystallizing. The most-discussed technical topic was not any single model but the foundational stack for agentic workflows: Tailscale for connectivity, tmux for session persistence, private Git for agent-to-agent state transfer, and structured dev environments. Multiple high-engagement posts argue this infrastructure "underneath the model" is what actually determines productivity, and that model selection is the most swappable part.

  2. AI agent tooling is in a rapid feature-shipping cycle. Codex added /goal persistence and mobile execution; Hermes shipped v0.14.0 with multi-backend support (xAI, OpenAI Codex), new gateways, and automation for task decomposition. The velocity suggests competitive pressure between agent platforms is intensifying.

  3. Compute constraint is universal and self-renewing. Even developers with cutting-edge hardware (RTX 5090, DGX Spark) report being compute-constrained, with workloads outgrowing hardware faster than it arrives. This validates continued demand for both cloud compute and local inference optimization.

  4. Open-source AI momentum continues. PapersWithCode's revival, Supertonic 3's browser-native TTS, Qwen 3.6's local performance dominance, and Nous Research hitting 1,000 contributors all signal a healthy and growing OSS AI ecosystem—contrasted by concerns about OSS sustainability (rate-limiting, API compatibility breaks).

  5. Pricing-layer opacity is emerging as a friction point. The Claude Code speed debate highlights a broader concern: AI providers may be quietly degrading standard-tier performance to make premium tiers feel necessary, with the endgame being deprecation of the "cheap" option entirely.

What to Watch: - Whether Anthropic responds to Claude Code speed complaints publicly - Hermes Agent adoption metrics post-v0.14.0 with xAI integration - Qwen 3.6 local benchmark challenges—who can beat 27B dense at Q4 on a single 3090? - Microsoft ClawPilot enterprise rollout details - Git-as-agent-shared-memory pattern adoption in production agentic systems

Tweet Feed

AI Agent Platforms & Orchestration

@Teknium · 2026-05-17T20:35

Hermes v0.14.0 is now out. This brings the xAI Supergrok & Premium+ account access for Grok models, image gen, video gen, and x search. Codex as a runtime backend for openai models, LINE as a new gateway messenger, Some huge performance increases, Native video generation, Our Windows native beta and a lot more, read the full release notes below! → tweet

@Teknium · 2026-05-18T07:29

The Hermes Agent Kanban just got a big automation upgrade. Drop one prompt into the triage, and the orchestrator agent can take it from there - decomposing it into all the subtasks necessary and automatically assigning agent profiles that fit the specialization needed. You can also now add descriptions for each agent profile, to better assist the orchestrator in determining what tasks should go to which profile! → tweet

@Teknium · 2026-05-18T06:22

Just pushed an update to use the tool use enforcement prompting we had on gpt models on grok models - if you find its a bit unwilling to get moving on tasks - please hermes update! → tweet

@Teknium · 2026-05-18T05:54

RT @akshay_pachaar: Hermes meets SuperGrok! xAI just made every SuperGrok subscription work inside Hermes Agent. One browser login, no AP… → tweet

@Teknium · 2026-05-18T05:49

RT @shannholmberg: how to go from prototype → production with Hermes Agents. here's the framework I use every time I spin up a new speciali… → tweet

@Teknium · 2026-05-18T19:02

👀👀 → tweet

@Teknium · 2026-05-18T17:56

RT @NousResearch: We have hit 1000 contributors on the repo, a nice milestone to start out the week. Thank you to all of the contributors… → tweet

@Teknium · 2026-05-18T05:58

We are currently hiring full stack engineers to work on managed services, nous portal, UX/UI for hermes agent and applications around it, and solving technical challenges cross-domain. If you're interested in applying, please email recruiting@nousresearch.com with the subject "Full Stack Engineer Role" and your CV and/or Portfolio of work. → tweet

@Teknium · 2026-05-17T19:04

Hermes Atlas is a really, really useful resource → tweet

@gdb · 2026-05-18T17:44

how to use /goal in codex — keep Codex working on a persistent objective until it's solved: → tweet

@gdb · 2026-05-18T18:55

Keep your Mac awake so you can build and work from your phone, with Codex in the ChatGPT app: → tweet

@gdb · 2026-05-18T04:01

Codex for unsubscribing from unwanted marketing emails → tweet

@gdb · 2026-05-18T01:33

codex for deeply personal insights → tweet

@gdb · 2026-05-17T22:26

so much joy in asking codex for random questions at work (such as finding some specific spreadsheet i'd been looking at a while ago), much more fun than searching around for context by hand → tweet

AI Models & Research

@victormustar · 2026-05-18T18:09

RT @ggerganov: llama.cpp adds MTP for the Qwen3.6 family. This is a significant milestone for the local AI ecosystem. The performance jump… → tweet

@victormustar · 2026-05-18T11:03

Supertonic 3 is incredible 🤯 Open source on-device TTS that runs in the browser, sounds human, handles a dozen+ languages, and finishes generating before you finish reading your prompt! ⬇️ demo available on Hugging Face → tweet

@victormustar · 2026-05-18T14:07

RT @NielsRogge: Introducing a revival of PapersWithCode! As @ilyasut said, we're back to the "age of research." Hence, it's important to… → tweet

@sudoingX · 2026-05-18T13:36

saying it out loud again. on a single 3090, the king is qwen 3.6 27b dense at q4, and nothing in that tier comes close. i've benchmarked the tier. happy to be wrong, so name the model that beats it. i just know you can't. → tweet

@nummanali · 2026-05-18T17:35

RT @leerob: @nummanali Based on K2.5 but it's here! → tweet

@louszbd · 2026-05-18T13:40

RT @xubinrencs: /goal build super mario game in @nanobot_project! End-to-end recording with glm-5.1 from @Zai_org. It only cost $0.1 in… → tweet

@tinygrad · 2026-05-18T06:42

The AI panic is really unbelievable today. The level of delusion and hype have grown to mythic proportions. Has AI beaten Pokemon Red yet? Like a normal 6 year old does, by looking at the screen? Oh it hasn't. But all jobs are over in 18 months? This website is full of idiots. → tweet

@sama · 2026-05-18T18:04

chatgpt has gotten soooo much better with the latest update. really proud of the team for this one. → tweet

@sama · 2026-05-18T00:11

ChatGPT Images 2.0 💚 India. Already more than 1 billion images created there; awesome to see. → tweet

@steipete · 2026-05-18T04:14

RT @RhysSullivan: the average person has only ever used ChatGPT 3.5 Instant and has no idea what the models can do → tweet

@levelsio · 2026-05-18T18:07

RT @elonmusk: Try it out! (Partially trained on Colossus 2) → tweet

AI Developer Workflows & Infrastructure

@sudoingX · 2026-05-18T11:20

six months ago i thought my problem was the model. i kept switching models, chasing the better one, certain the next one would fix the friction. it never did. the friction was never the model. it was that i had no foundation, context died every time i closed a window, long runs died on a disconnect, half my day went to reaching machines that would not stay reachable. the fix was a weekend, not a new model. set up the five and the friction is just gone. → tweet

@sudoingX · 2026-05-18T07:21

the model you run is the most swappable part of your setup. change it on a tuesday and nothing else has to move. the foundation underneath, how your machines reach each other, where state lives, how context survives a restart, that is the part that is expensive to change once you are deep in it. everyone pours their attention into picking the model and treats the foundation as an afterthought. you have it backwards. pick the model last. → tweet

@sudoingX · 2026-05-18T04:55

most people treat git as a chore. commit, push, don't forget, a discipline you impose on yourself. in an agentic setup it stops being that, git becomes the thing your agents talk through. one agent does work and commits. the next one pulls and now it has everything, the code, the docs, the context the last agent left behind [...] the agents never have to talk to each other directly. one leaves state in the repo, the next picks it up, async and durable [...] private git isn't a step you enforce, it's the medium. → tweet

@sudoingX · 2026-05-18T18:50

if that list feels like a lot, here is what nobody tells you. you do not set up all five at once. i tried that, all five in a weekend, bounced off every one. set up one. tailscale first, always, nothing else works until your machines can reach each other. run it a few days, then add the next. one at a time. a month out you have all five locked, and you stop losing your flow. → tweet

@sudoingX · 2026-05-18T13:43

if you are working with agentic systems and you quietly feel like you are faking it, like everyone else got the memo and you are just barely keeping up, i want to tell you what is actually happening. it is not a talent gap. it is a setup gap. [...] you are not behind. you are one weekend of setup away from feeling like you belong. → tweet

@sudoingX · 2026-05-18T05:20

the things that actually changed how i work are small and unglamorous, a tmux config, a control script, the git setup my agents live in, and i open every one of them daily. build something you come back to. → tweet

@sudoingX · 2026-05-18T03:59

i posted a list about tailscale and tmux, the most unsexy thing i could think of, and it's about to cross 100k views. that's not me, that's the signal, the agentic setup corner is way bigger and hungrier than the timeline lets on and almost nobody is posting into it. more of this coming. → tweet

@sudoingX · 2026-05-18T15:18

4 minutes 20 seconds. one prompt. this is the timer. i am not exaggerating, i am documenting. → tweet

@kunchenguid · 2026-05-17T22:43

i never use "remote control" in agent harnesses because they feel more like middle grounds. i tailscale to my mac + use terminus to do the real thing. this setup has no extra cost → tweet

@sudoingX · 2026-05-18T14:06

people keep asking why i don't paywall more of it. the honest answer: the default move is to charge for everything you can, and i think that is backwards. you give away everything that can be free, because that is what earns the right to charge for the one thing that can't. → tweet

AI Pricing & Performance Concerns

@sudoingX · 2026-05-18T15:11

i use claude code every day, opus 4.7 max used to be fast. now i wait 3 minutes for a single answer. that is not a one off, it is the new normal. here is my read. they slowed the standard model down so the /fast mode feels worth paying for, and /fast runs you 6x the cost. you are not paying for fast anymore, you are paying to undo the slow. and here is the part nobody says out loud. at some point dario deprecates the slow normal tier entirely, and 6x stops being the upgrade, it just becomes the price. → tweet

Hardware & Compute

@sudoingX · 2026-05-18T14:19

i have a 5090 on my desk and a dgx spark beside it. i am still compute constrained. i am always compute constrained. the work outgrows the hardware faster than the hardware shows up. it always will. → tweet

@TheAhmadOsman · 2026-05-18T17:35

My only regret is actually not telling people to Buy a GPU earlier than last summer. Given how many people called me crazy so close to the demand surge, I am not sure how people would've reacted to me saying that in 2024 for example. BTW, you should acquire some compute ASAP → tweet

@TrungTPhan · 2026-05-18T01:25

The arms race of quant funds making vids about water cooling AI data centres is unreal. Hudson River Trading uses one in Norway cooled by fjord seawater. It's piped into a former mountain-side mineral mine now hosting GPU racks. Heated water sent to a land-based salmon farm nearby and the farm produces 15,000 tonnes of salmon a year. → tweet

@TrungTPhan · 2026-05-17T22:22

Just watched Jane Street data centre tour and most entertaining part was team nerding out on air cooling vs. water cooling. Filters to ensure water perfectly uniform and flow is consistent to move heat. Ultrasonic monitors to measure the flow. The water has specific ratio of propylene glycol to prevent bacteria or algae growth. → tweet

@TrungTPhan · 2026-05-18T01:30

RT @bearlyai: Cerebras CEO Andrew Feldman talks about why there are so few young semiconductor chip founders: "The silicon industry is not… → tweet

@tinygrad · 2026-05-18T18:38

This is who we want as tiny corp customers. → tweet

Open Source & Dev Tools

@badlogicgames · 2026-05-18T09:50

People of https://t.co/oUoqqL9hAp. More fixes, because undici and Windows hate us! If pi update fails for you, try manually updating pi via npm/pnpm/whatever you use. Also, you must have Node >= 22.19.0! → tweet

@badlogicgames · 2026-05-17T19:06

People of https://t.co/oUoqqL9hAp. The @matteocollina release. Node 22 is now the minimum node version. Needed to upgrade to Undici 8+ which requires Node 22+. I'm sorry, but also welcome to the glorious future. Plus a bunch of minor fixes. → tweet

@badlogicgames · 2026-05-18T00:01

People of https://t.co/oUoqqL9hAp. awkward. from my layman's perspective it seems node 26.0.00 has a few undici related booboos in it. if copilot or codex login didn't work for you, update and try again. Tested against Node 22 - 26... → tweet

@badlogicgames · 2026-05-17T22:08

fun debugging session. pi is accessing xiaomi endpoints via the anthropic messages API. they seem to have changed their endpoint to now require the non-standard openai-completions reasoning_content field, breaking their anthropic endpoint :/ → tweet

@badlogicgames · 2026-05-17T22:44

hey, @opencode it seems like the go provider load balances between different kimi 2.6 deployments, and some return reasoning in reasoning, while some do so in reasoning_content. however, all deployments only accept reasoning_content. that's a bit awkward. can have fix? → tweet

@badlogicgames · 2026-05-18T09:19

hey pi kids. if you use @opencode zen or go for their free models, expect them to rate limit you hard. pi will not add the special headers needed for that not to happen. gentlemen's agreement. → tweet

@badlogicgames · 2026-05-18T00:11

personally don't like auto-installing 350mb on Windows, but if you are a Windows user and like it, say the word, and Clankolas will fill your SSD. → tweet

@badlogicgames · 2026-05-17T20:15

my mom is now concerned about mythos, because dario and his possy got a prime time news show spot. cool. → tweet

@badlogicgames · 2026-05-17T21:05

recommended reading. also read part II. also read naur's original article. → tweet

@badlogicgames · 2026-05-17T20:59

recommended reading. still relevant. → tweet

@badlogicgames · 2026-05-18T12:31

recommended viewing. love the insights on distillation and edge inference. i want that. → tweet

@badlogicgames · 2026-05-18T10:39

recommended viewing. → tweet

@badlogicgames · 2026-05-18T16:32

RT @matteocollina: We made fetch in @nodejs v26 default to http2 if the server prefers… there are bugs. → tweet

@badlogicgames · 2026-05-18T10:09

RT @nateberkopec: I'm so sick of reading em dashes and "it's not x, it's y." I'm so sick of it, man. → tweet

@badlogicgames · 2026-05-18T17:32

RT @ryoppippi: omg so scary. end of oss… i'm serious → tweet

@TheAhmadOsman · 2026-05-17T20:24

Transformers library needs better quality control ngl. I am always anxious about what night break with each and every update that library gets → tweet

@TheAhmadOsman · 2026-05-17T22:08

RT @TheAhmadOsman: Opensource AI is going to win btw → tweet

@steipete · 2026-05-18T17:55

RT @bensen: "@Microsoft is piloting '#ClawPilot,' an always-on AI assistant built on the open-source @OpenClaw framework, with over 3,000 e… → tweet

@steipete · 2026-05-18T15:26

RT @cherry_mx_reds: ClawSweeper now turns OpenClaw PR review into a tiny crustacean RPG. Ranks: S 🦀 challenger crab, A 🦞 diamond lobster… → tweet

@thdxr · 2026-05-17T22:33

ok tried the same on cloudflare for our x-rank tracking thing. was just as easy and the built in sqlite is really nice, i was shoving stuff into blob storage on vercel → tweet

@thdxr · 2026-05-17T21:47

finally a vercel user. we have these internal apps we create 100% vibecoded and there really isn't a better place to throw those up, esp because opencode can do it all. one thing to make it better would be an IaC file the agent can create, thought i saw some mention of that → tweet

@levelsio · 2026-05-17T22:08

💾 Finally set up @Litestreamio for the first time on 🏡 Interior AI to test. It is open source and free and lets you add kind of a listener to your SQLite db which is constantly making a live copy of your databases to any S3-compatible bucket. In my case I use Cloudflare's R2. This is in addition to local provider backups, daily external backups, and cold storage off-site backups I have (e.g. the 3-2-1 Backup Rule!) → tweet

@gospaceport · 2026-05-18T12:45

This is GREAT NEWS! No more: > Windows Server licenses > Win CALs > o365 seats > Exchange Server > Copilot > Sharepoint SE. Pretty much the entire Enterprise sw suite, unneeded. → tweet

@ASalvadorini · 2026-05-18T09:58

RT @rseroter: Just built my first @FlutterDev app! That was fun. Used Flutter tools, a loaded up @antigravity, and @AndroidStudio (for the… → tweet

@FrameworkPuter · 2026-05-18T09:53

RT @cyber_v1: @DevBredda @i2cjak It's there and it kind of works (devices can be connected and are recognized). I didn't spend any more tim… → tweet

AI Industry, Robotics & Applications

@TrungTPhan · 2026-05-18T16:07

The Boston Dynamics Atlas demos are always impressive but its current most useful robot isn't humanoid. It's prob their truck-unloading robot called "Stretch", already operational with DHL (the logistics company has a deal for 1,000 of these to unload trailers). [...] Stretch is ~$120k a pop. → tweet

@TrungTPhan · 2026-05-18T19:00

Visualising Linus Torvalds typing up this e-mail from his legendary home office: a walking desk with single monitor while wearing flip flops on socks and doing a slow methodical "zombie walk" so he doesn't fall. → tweet

@TrungTPhan · 2026-05-18T17:29

RT @bearlyai: Linus Torvalds weekly update on state of Linux kernel went off on AI-powered bug detection tools. Many researchers are findi… → tweet

@nummanali · 2026-05-18T17:09

@xai and @cursor_ai are cooking up a beast → tweet

@nummanali · 2026-05-17T23:01

I wrote this on 8th December 2025. Look at where we all are now → tweet

@MengTo · 2026-05-18T12:18

Images 2.0 + Grok Imagine for landing pages is an insane workflow. I start with a DESIGN.md, select the landing page options and specify the app theme. After generating the page, I recreate all the detected images with Images 2.0. Then I turn them to videos. All in one tool. → tweet

@jsuarez · 2026-05-18T17:07

PufferLib RL dev with Joseph Suarez → tweet

@jsuarez · 2026-05-18T17:03

Starting now! Quick overview followed by the first full day of research, streamed live right here → tweet

@levelsio · 2026-05-18T16:52

In a way I think the top tech companies have just vacuumed up all the top talent worldwide for such great salaries + equity (for $500K to millions $ per year). And the top tech companies also have built such a great talent acquisition funnel that everyone else in the world who isn't working for top tech is either 1) already rich and retired, 2) a founder already or 3) just not good enough → tweet

@levelsio · 2026-05-18T16:48

The quality of people's work is so low that you just simply can't hire for most things anymore [...] hiring now is kinda like doing charity work [...] But then at the same time AI isn't good enough to do most of these human things well (yet) → tweet

@FinansowyUmysl · 2026-05-18T08:45

AI to nie tylko bity – to gigantyczny popyt na atomy i molekuły: Mag 7 + Oracle wyda w 2026 ponad 820 mld USD na inwestycje. Amazon zużywa więcej energii niż większość krajów OPEC. Potrzebne są: energia, miedź, woda, gal, stal, beton – wszystko, czego od 10–15 lat nikt nie inwestował. → tweet

@LinusEkenstam · 2026-05-18T06:19

I'm headed to San Francisco to this year's edition of Google I/O. Extremely excited to meet other creators and builders at the Shoreline. I'll be in SF for a full week, let's meet up. Thank you google for inviting me. #giftbygoogle → tweet

@LinusEkenstam · 2026-05-17T19:59

f.03 is now streaming on the big screen → tweet

@uwteam · 2026-05-18T06:12

Mikrus jest partnerem konferencji AI Miners. To wydarzenie dla osób, które pracują ze sztuczną inteligencją w praktyce - dla developerów, inżynierów, tech leadów i osób z biznesu, które wdrażają AI w realnych projektach. 📍 Katowice - 📅 21 maja → tweet

@sudoingX · 2026-05-18T05:39

Hermes agent from shower → tweet

@sudoingX · 2026-05-18T05:45

what's stopping nous research from building this? → tweet

@alexinexxx · 2026-05-18T17:50

God's plan for me doesn't involve AI replacing my job → tweet

@alexinexxx · 2026-05-17T21:55

God's plan for me doesn't involve a 6-round application process → tweet