← Tech / AI / IT Monitor Index Tech / AI Generated 2026-05-08 19:31 UTC

Tech / AI / IT Monitor

May 08, 2026 · Based on tweets from the last 24 hours · 159 tweets analyzed · model: ollama-cloud/minimax-m2.7:cloud

Daily Intelligence Briefing: Tech / AI / IT Monitor

Date: 2026-05-08 Period: Last 24 hours

Executive Summary

This period saw major activity in autonomous AI agent capabilities, with the Hermes agent (Nous Research) autonomously writing custom CUDA kernels to optimize its own inference pipeline on a DGX Spark — a concrete demonstration of self-improving AI systems. OpenAI made notable moves with GPT-5.5's capabilities, real-time voice-to-voice translation entering the API, and Codex gaining Chrome tab control on macOS and Windows. Meanwhile, Intel's market cap surged 6× under new CEO Lip-Bu Tan, signaling renewed confidence in hardware-software convergence. On the open-source side, Tinygrad's specification merged and Hermes Agent v0.13.0 shipped with multi-agent orchestration, reinforcing a trend toward locally run, self-hosted AI workflows replacing cloud API dependencies for power users.

Key Events

Analysis

Patterns: The dominant theme this period is the convergence of autonomous agents + local inference + self-optimization. Several power users are now running multi-step agent workflows overnight on local hardware (DGX Spark, M-series Macs), with agents autonomously writing and optimizing low-level kernels — a trend previously theoretical. This is accompanied by a clear deprecation of cloud dependency: llama.cpp from source is being recommended over abstraction-layered tools, and open-source agents (Hermes, OpenClaw) are gaining Windows and Mac desktop integrations.

Escalation trend: Agentic capabilities are escalating rapidly. The /goal command — where a user sets a single high-level objective and the agent executes, iterates, and reports back — is being adopted as a productivity paradigm. Combined with real-time voice integration and Chrome automation in Codex, the surface area of AI-accessible tasks is expanding from code completion to full-stack development and system administration.

De-escalation/caveats: Some friction signals persist: Claude Code reportedly took a user's site down and struggled to self-recover; GPT-5.5 in Codex CLI reportedly stopped reading files for some users, prompting regressions to 5.3-codex. The "frontier model regression" narrative (potentially from RLHF alignment tradeoffs) is referenced by at least one developer.

What to watch: The /goal autonomous workflow pattern is likely to be replicated across more agent frameworks. Tinygrad's merged spec and RDNA4 GPU box plans suggest hardware-level optimization is becoming more accessible. The gap between local and cloud inference quality continues to narrow, which may shift enterprise procurement patterns.

Tweet Feed

🤖 AI Agents & Autonomous Systems

@sudoingX · 2026-05-08T10:39

my dgx spark is writing custom CUDA kernels to make itself faster. let that sink in. hermes agent running qwen 3.6 27B Q8 autonomously decided to port its own triton kernel to native CUDA C++ for llama.cpp integration. it understood the dispatch chain. studied the mmq kernel structure. now it's writing the port itself. this machine is literally optimizing its own inference pipeline. no human in the loop. i set a /goal last night and woke up to a 12.91x speedup on SSM and 9.66x on Q8 matmul. now it wants another 2-3x through FP8 tensor cores. → tweet link

@sudoingX · 2026-05-08T10:50

do you understand what's happening here? if this doesn't excite you about local ai nothing will. my dgx spark is writing custom CUDA kernels to optimize its own inference. the agent studied the triton-proven algorithm, understood the dispatch chain, and is now writing a native CUDA kernel as a fast path for Q8 matmul decode. this is a machine improving itself. autonomously. powered by hermes agent /goal running qwen 27B locally. no human wrote this. → tweet link

@sudoingX · 2026-05-08T08:08

update: hermes agent with 27b dense has been running autonomously locally on my dgx spark since last night. here's what it did while i lived my life. built custom fused kernels for qwen 27B Q8. SSM kernel 12.91x speedup. triton Q8 kernel 9.66x faster than naive pytorch. now it's investigating FP8 tensor cores on the GB10 for another 2-3x. i didn't write a single line of this. i set a /goal. it executed. this is what co-evolving with your hermes agent looks like. → tweet link

@sudoingX · 2026-05-08T07:16

hermes agent is hammering my dgx spark at 96% gpu utilization right now. 76°C in bangkok heat. it's experimenting with custom fused kernels optimized specifically for dgx spark + qwen 3.6 27B dense Q8. continuing last night's /goal autonomously. i woke up. it didn't stop. let's see where this goes. → tweet link

@sudoingX · 2026-05-08T04:49

i named my dgx spark "spark." it runs hermes agent /goal overnight. brain is qwen 3.6 27B Q8, 262K context, i set a goal before bed and wake up to results. no rate limits. no token costs. just local inference grinding while i sleep. this thing never stops. → tweet link

@sudoingX · 2026-05-08T11:37

it's now writing CUDA C++ dispatch hooks for llama.cpp. the agent is patching mmqcu to route Q8 matmul through its own optimized kernel. a 27B model on a dgx spark is rewriting its own inference engine. i'm just watching. → tweet link

@sudoingX · 2026-05-08T15:46

anyone interested in or getting started with local ai personal inference, pay attention. start with the right practice. compile llama.cpp from source. i know lm studio and ollama exist. they're great onramps. but they're mostly wrappers around llama.cpp with abstraction layers that hide the flags you actually need to tune. what compiling once gets you: the best inference engine for personal use, full stop. → tweet link

@sudoingX · 2026-05-08T16:57

after today's spark posts, lots of you asking how the hermes agent /goal flow actually works. here's how to write a goal that actually executes. what /goal does: hermes agent autonomous mode, you set the goal once, model executes without supervision, writes files, runs commands, builds, tests, iterates, closes the loop or tells you why it can't. → tweet link

@sudoingX · 2026-05-08T05:02

now the problem is not how many sessions you run. it's how many agents you forget are still running. this is why i love tmux. you can always come back. it's always there. → tweet link

@Teknium · 2026-05-07T21:05

Introducing Hermes Agent v 0.13.0 — Multi-Agent orchestration through the Kanban system, Enforced goal completion with /goal, Big optimizations for disk usage, Much more extensibility, custom LLM Providers, custom gateway channels, and much more. → tweet link

@steipete · 2026-05-07T22:23

/goal + GPT 5.5 is amazing. I can now plan really extensive refactors with e2e tests and it just works. → tweet link

@badlogicgames · 2026-05-08T11:24

set up an auto-research loop to have the agent build out a GPU-side vector graphics renderer, using Slug's methodology, as well as a SVG -> Slug-ish format. generated ground truth .pngs from SVGs, then let the loop render the same SVGs, use imagemagick to get RMSE for the ground truth/GPU rendered diff. it's the perfect task for auto-research. → tweet link

@kunchenguid · 2026-05-08T17:20

the new @OpenAI realtime voice model just released + gpt 5.5 fast mode brings us a new possibility — realtime speech to live presentation! i just talk, and the whiteboard would whiteboard itself. prototype is open sourced. → tweet link

@thdxr · 2026-05-08T17:21

more and more you can just ask the agent to do things, but it's pretty hand wavey to say that means products don't need any native features. this is a nice example where you can go agent first but then your tool also understands what was done. → tweet link

@nummanali · 2026-05-07T22:17

I'm so close to giving an agent access to my whole life ie gmail, browser, phone etc. To get my long todo list out the way. But, alas, even with all my experience I do not feel comfortable. → tweet link

@steipete · 2026-05-08T06:02

Our claws talk to each other, Molty learns how to delegate cron jobs. → tweet link

@Teknium · 2026-05-07T22:17

RT @DODOREACH: Fully supported in Hermes Desktop — The safest, cleanest way to manage your Hermes agent on Mac. No cloud. No middleman. → tweet link

🧠 AI Models & Research

@gdb · 2026-05-08T16:12

GPT-5.5 is both very capable and very succinct. → tweet link

@gdb · 2026-05-08T02:56

GPT-5.5-Cyber is now in limited preview for defenders for securing critical infrastructure. It's a very capable model. → tweet link

@badlogicgames · 2026-05-08T00:11

yeah, going back to gpt-5.3-codex. 5.5 doesn't read stuff anymore, neither in pi, nor in codex CLI. → tweet link

@Teknium · 2026-05-08T15:28

Love how big our ecosystem has become. → tweet link

@Teknium · 2026-05-08T15:26

RT @DivyanshT91162: Crazy how fast this changed. Web search is basically free now for AI agents. I switched my OpenClaw and Hermes agents to use it. → tweet link

@badlogicgames · 2026-05-08T03:32

but it's cool that frontier models are now basically regressing. maybe all this madness will come to an end soon. → tweet link

@levelsio · 2026-05-07T20:35

Something going on with my Claude Code right now, it just took my site down (second time in 12 months), couldn't fix it by itself, I had to manually fix it, I did within 5 seconds, so that's OK but yeah it feels supremely dumb and mostly slow the last few hours. → tweet link

@levelsio · 2026-05-07T21:00

Maybe Claude is just tired? → tweet link

@alexocheema · 2026-05-08T18:14

RT @exolabs: Hello everyone, we're working on making local AI viable for real work. To do that we need to understand where it's falling short. → tweet link

🛠️ Developer Tools & Platforms

@sama · 2026-05-07T20:16

RT @OpenAI: Codex now works directly in Chrome on macOS and Windows. It's even better at working with apps and sites in Chrome, and now works on the web. → tweet link

@gdb · 2026-05-07T23:04

Codex can now drive Chrome tabs in the background. → tweet link

@gdb · 2026-05-07T20:09

have been excited for realtime voice-to-voice translation as an AI application since we started OpenAI. extremely cool to see it now available in the API for anyone to build with. → tweet link

@sama · 2026-05-07T20:25

way cooler to help software developers pokemon-evolve into superheroes than to try to replace them. it is insane what one really good person can do now. → tweet link

@sama · 2026-05-08T01:16

we'd like to help companies secure themselves and we think it's important to start work on this quickly. → tweet link

@jezell · 2026-05-08T18:40

Ported the codex command / tool parser from rust to dart for better agent summaries. Also to Python for our CLI. Really improves the raw tool summaries. → tweet link

@gdb · 2026-05-08T17:40

codex is for everyone — a transformative tool for all work done with a computer, not just coding. → tweet link

@steipete · 2026-05-08T00:59

RT @BenjaminBadejo: Here's OpenAI's latest realtime voice model, GPT-Realtime-2, wired up to OpenClaw. It's amazing. Realtime continuous chat. → tweet link

@badlogicgames · 2026-05-08T03:01

TIL about "caveman mode" to "save tokens". how many tokens in a session are actually model output? i think i'll become a gardener. → tweet link

@badlogicgames · 2026-05-08T20:17

wrote a little script to analyze my pi sessions in the pi repository. this is what i mean with i keep my session scope small. → tweet link

@badlogicgames · 2026-05-07T21:43

welp (codex cli) → tweet link

@iamdevloper · 2026-05-08T11:08

ok, now I believe we're entering the "AI took my job" era for software engineers... → tweet link

@iamdevloper · 2026-05-08T12:32

5 mins later > it's done → tweet link

@badlogicgames · 2026-05-07T21:17

RT @antirez: In case you have doubts about the q2 quants inference of DS4 (I noticed many don't trust the README claims), here is it analyzed. → tweet link

@badlogicgames · 2026-05-07T22:04

RT @mitsuhiko: I'm so in love with @antirez' ds4. Patched some slop on it to get better streaming, but I can just install a pi extension on it. → tweet link

@thdxr · 2026-05-07T19:22

RT @jlongster: new feature in opencode: warping — never worry about whether or not you should work in a worktree again! now you can move s... → tweet link

@levelsio · 2026-05-07T19:39

[Image: Claude Code output] → tweet link

🔬 Open Source & Frameworks

@tinygrad · 2026-05-08T02:27

The tinygrad spec is now merged in tinygrad/spec. Unlike every other ML compiler, all optimization is done in this IR all the way up to instruction selection. → tweet link

@tinygrad · 2026-05-08T02:21

32GB of RDNA4 on USB 3.2 Gen 2. Like the dock? → tweet link

@tinygrad · 2026-05-08T02:36

Any market for a 25k box with 6 32GB RDNA4 GPUs on full fabric PCIe 5? We'll build it if we get two orders. Also, we're going to put the 5090 boxes back in stock once we calculate what we are paying for the parts (+20% markup). Our current price for 5090s is $4200 afai. → tweet link

@Teknium · 2026-05-07T23:21

Agents can make magic happen. → tweet link

@Teknium · 2026-05-07T22:19

RT @makenki_ai: ヘルメスにも/goalきた。最強やんえぐすぎる。seo運用のエージェント執筆フローがこれまた強化される → tweet link

@badlogicgames · 2026-05-07T20:56

my m1 max is falling apart. should i wait or buy some beefed out m5? i'm so torn. → tweet link

@ollama · 2026-05-08T02:51

RT @NVIDIAAI: Open source isn't just good for developers, it's one of America's strongest tools for AI security. More models means more de... → tweet link

@Teknium · 2026-05-08T15:52

RT @mr_r0b0t: HUGE thanks to @Teknium and the @NousResearch team for this update! Loving the quality of life upgrade. → tweet link

@steipete · 2026-05-08T03:45

RT @BenjaminBadejo: OpenClaw is snappier than ever. Good to have focused on stability in the last update. The next things I personally think about are... → tweet link

@Teknium · 2026-05-07T19:50

RT @ComfyUI: When your tool is open source and free, your creativity has no ceiling. The ComfyUI skill in @NousResearch Hermes Agent lets you... → tweet link

@badlogicgames · 2026-05-07T20:13

really neat project by @antirez. → tweet link

@badlogicgames · 2026-05-07T11:24

recommended reading by our junior developer @mitsuhiko. my name is pidalf, and i support this message. would love to join @antirez effort and use some of my GPU knowledge. → tweet link

🏢 Industry & Business

@TrungTPhan · 2026-05-08T17:41

how Lip-Bu Tan rolling up to Intel's Friday Happy Hour after INTC market cap has spiked 6x to $623B since he became CEO in March 2025. → tweet link

@TrungTPhan · 2026-05-08T17:57

Pat Gelsinger seeing Lip-Bu Tan increase Intel's market cap 6x to $626B by implementing most of Gelsinger's strategic roadmap. → tweet link

@TrungTPhan · 2026-05-08T17:55

RT @bearlyai: Lip-Bu Tan increased Intel's market cap by 6x to $626B since becoming CEO in March 2025 boosted by series of deals: August... → tweet link

@RealGeneKim · 2026-05-08T14:47

RT @alexalbert__: With the help of Claude Mythos Preview, the Firefox team fixed more security bugs in April than in the past 15 months combined. → tweet link

@TheAhmadOsman · 2026-05-08T12:23

In the DMs: hey I see you're in SF [talks about himself] wondering if you have time to meet & mentor me. Response: I have a dinner plans with GF, can meet you 2 hours after that. Seriously? → tweet link

@TheAhmadOsman · 2026-05-07T21:06

Yup, just start and you'll learn everything you need about local AI as you go. It's fun, addicting, and bad for your wallet so be warned. → tweet link

@TheAhmadOsman · 2026-05-08T18:28

Could you imagine that AGI might run on GPUs from 10+ years ago? Because it just might. There's a reason I haven't sold a single one of my GPUs despite the massive jump in price. Compute is the MOAT + you don't know what the next paradigm might allow you to do with that hardware. → tweet link

@TheAhmadOsman · 2026-05-08T02:36

RT @philipkiely: Talked inference with @TheAhmadOsman. Our conclusion: buy a LOT of GPUs. → tweet link

@TheAhmadOsman · 2026-05-07T19:19

I like the Tenstorrent office, there's always cool hardware on display. → tweet link

@TrungTPhan · 2026-05-07T23:58

Lloyd Blankfein has a great story of Mike Bloomberg going extra mile on client work. When Lloyd started on Wall Street, his firm had a Bloomberg terminal but no one bothered to figure out how to use it. → tweet link

@steipete · 2026-05-07T20:29

Had the honor of mentoring some of the folks in the ChatGPT Future Class of 2026 this year. Shoutout to @arhan_menta @nayelr_ @rushilkukreja who built Wi-Find, a system that detects disaster survivors through walls and debris using AI. → tweet link

@jack · 2026-05-08T15:08

RT @blocks: At the @tech_coalition's From Collective Action to Impact event, Block joined leaders across sectors to discuss how to scale so... → tweet link

💻 Hardware

@FrameworkPuter · 2026-05-07T21:32

RT @alexinexxx: new framework laptop 13 pro, what should i install? → tweet link

@alexinexxx · 2026-05-07T20:42

new framework laptop 13 pro, what should i install? → tweet link

@FrameworkPuter · 2026-05-08T16:52

RT @svpino: The trackpad of the Framework 13 Pro is one of its highlights. → tweet link

@badlogicgames · 2026-05-08T16:25

i swear they make these extra unrepairable. this would be a 2 minute soldering job if they didn't hide those damn screws in those narrow shafts. planned obsolescence in kids toys is the worst. → tweet link

🧪 Technical Deep-Dives & Tutorials

@badlogicgames · 2026-05-08T13:41

tired [image] → tweet link

@badlogicgames · 2026-05-07T21:01

living on the edge. → tweet link

@badlogicgames · 2026-05-07T20:33

RT @FredKSchott: Update your sites! Astro not affected (we don't use RSC) but always a good reminder to keep your deps up to date. → tweet link

@jezell · 2026-05-08T04:34

RT @Yuchenj_UW: A few OpenAI folks told me: "300M tokens/day is a rookie number." The biggest number I'm hearing now is 57B tokens/day! → tweet link

@jezell · 2026-05-07T22:42

RT @alexstauffer_: We post-trained a 3B model with RL to beat Opus on spreadsheet retrieval. Faster, cheaper, more accurate. → tweet link

@kunchenguid · 2026-05-08T04:13

when i added the meteor shower to gnhf, i totally forgot to update the screenshot in the readme file. but when i looked at it today, the screenshot does have meteors in it… somewhere one of my agents did it for me.. i don't know when and don't know how. → tweet link

@jezell · 2026-05-07T19:56

RT @diptanu: We have been very busy building sandbox infrastructure! If you are interested in KVM infrastructure, file systems, networking… → tweet link

@jezell · 2026-05-07T20:30

RT @juberti: Guess who's back, back again. Whisper, but now with realtime streaming. Check out the new gpt-realtime-whisper transcription model. → tweet link

@nummanali · 2026-05-08T10:28

"If I had asked people what they wanted, they would have said faster horses." – Henry Ford. → tweet link

@nummanali · 2026-05-07T22:28

RT @Seanfrank: two team styles crushing it right now: 1- young, no life, 12 hour days, VERY SMALL TEAM, in office, 6 days a week, hustle h... → tweet link

🏗️ Infrastructure & Cloud

@levelsio · 2026-05-08T15:36

Hetzner includes TERABYTES of free traffic! → tweet link

@levelsio · 2026-05-08T18:12

You get 20 TB of free bandwidth for $4.99/mo at @Hetzner_Online. Way way way way more than enough for most websites! I'm not sponsored by them, I don't even get discounts or anything, just a great deal! → tweet link

@levelsio · 2026-05-08T12:58

So @loaibassam asked me my stack recently, I replied: FREE: Nginx web server on Ubuntu, Auto upgrade with unattended-upgrade, Scheduled workers with Cron, Vanilla PHP, Vanilla CSS, Vanilla JS, Vanilla Node JS for game servers, SQLite for DB, Python for tool scripts, Cloudflare with Cloudflare tunnel for DNS/SSL, Tailscale for security, OpenFreeMap for maps. CHEAP: xAI for AI API, Stripe for payments, Cloudflare R2 for image storage. Hetzner VPS ($4/mo). So about ~$5/mo total costs with about ~5M unique visitors per month per site. → tweet link

@levelsio · 2026-05-07T22:14

RT @juanjovn: I actually enjoy my boring average days — 🏋️ gym + sauna, 👨🏻‍💻 morning dev lock in, 🐟 clean food, 📽️ afternoon marketing lock in… → tweet link

@levelsio · 2026-05-07T21:17

Man this makes sense, European evening now, American day, maybe that's why Claude is bad. → tweet link

@levelsio · 2026-05-08T14:41

You know your computer stores files on the same drive as your computer? How scary is that? → tweet link

End of briefing. All events relate to tech/AI/software/IT as instructed. Geopolitical, alien-themed, and off-topic tweets excluded.