Executive Summary
The past 24 hours in tech/AI were dominated by the rapid evolution of AI coding agents and local model deployment. OpenAI's Codex saw heavy adoption and infrastructure investment, with teams running dozens of parallel agent instances for code review, security scanning, and issue management. The Hermes Agent ecosystem expanded significantly with official Grok OAuth integration and growing community-built skills/plugins. On the model front, Qwen 3.6 27B emerged as the leading choice for consumer 24GB GPUs, while NVIDIA's DGX Spark enabled serious local inference of Nemotron and DeepSeek models. Meanwhile, developer sentiment showed growing frustration with commercial AI tool reliability, even as the industry doubled down on agentic workflows.
Key Events
- Isomorphic Labs (DeepMind spin-off) raises $2.1B to apply AI protein-structure prediction to full-scale drug discovery and clinical trials — one of the largest AI-biotech bets to date. → link
- Hermes Agent gains official Grok OAuth integration via xAI, replacing community-built fingerprint-spoofing workarounds with clean subscription-based access. → link
- Teknium confirms SuperGrok subscription support in Hermes Agent, expanding the agent's model backend options. → link
- Steipete reveals massive Codex deployment at OpenClaw: ~100 parallel Codex instances handling PR review, security scanning, issue de-duplication, spam blocking, benchmark regression alerts, and meeting-driven PR creation. → link
- Qwen 3.6 27B dense at Q4 declared "undisputed king of 24GB tier" — fits in VRAM with context headroom, hits ~41 tok/s on RTX 3090, one-shots playable games. → link
- Nemotron 3 Nano Omni-30B running locally on DGX Spark at 58 tok/s with 1M context and multimodal support, largely unnoticed by the community. → link
- LLM APIs lack cost-in-response metadata, forcing developers to approximate billing via third-party data — @thdxr calls for native cost reporting. → link
- VS Code Copilot extension sends extensive telemetry for next-edit prediction; developer instrumented the JS to inspect it. → link
- Triangle releases "Zero", a new systems programming language designed "for agents" — early mach-o emitter issues noted. → link
- James Cameron and Peter Jackson frame AI as VFX cost-reduction tool, doubling throughput rather than replacing artists; Cameron joined Stability AI's board. → link
Analysis
Patterns observed: The shift from "AI as assistant" to "AI as autonomous infrastructure" is accelerating. The OpenClaw Codex deployment pattern — agents running on every commit, every issue, every meeting — represents a new operational paradigm where codebases are continuously monitored and modified by swarms of agents. This is mirrored in the Hermes ecosystem's evolution from a single model to a full agent OS with community-built plugins, memory systems, and enterprise adapters.
Tension escalating: Developer frustration with commercial AI tools is rising alongside adoption. Claude Code is called "buggy," Anthropic is criticized for fearmongering while shipping broken tools, and the broader community notes a disconnect between "replace all engineers" narratives and the reality of fragile tooling. Simultaneously, local/open-source AI is being championed as both a practical and ideological alternative — continual learning requiring local weights, vendor lock-in concerns, and cost control all favor on-prem deployment.
What to watch next: OpenCode's upcoming worktree feature (paired with an unannounced companion feature) could reshape multi-PR parallel workflows. The DGX Spark + local model combination (Nemotron, Qwen, DeepSeek) is reaching a tipping point where consumer hardware supports production-grade autonomous agents — if reliability catches up. Expect continued friction between open-source agent frameworks and closed-platform providers as OAuth/API access becomes the new battleground.
Tweet Feed
AI Models & Local Inference
@sudoingX · 2026-05-16T15:57
if you run a single 24gb gpu, a 3090, a 4090, a 7900 xtx, whatever gets you the 24 gigs, the no brainer pick is qwen 3.6 27b dense at q4. not close. i have run the tier. it fits in 24gb with real context room to spare, it keeps the reasoning smaller models lose, it pushes around 41 tok/s on a single 3090, and i watched it one shot a playable game start to finish, zero iterations. nothing else in that vram class does what this model does. undisputed king of the 24gb tier, and there is nothing you can say to change my mind. → tweet
@sudoingX · 2026-05-16T08:11
nemotron 3 nano omni-30B reasoning at Q8 running autonomously on my dgx spark right now. 58 tok/s. 1 million context. multimodal. hermes agent is using it to research xAI's new grok algorithm that dropped yesterday. pulling repos. scanning code. breaking it down. all while i post this. 30B model. 58 tokens per second. 1M context. reads images and video. locally. for free. nobody is talking about this model and that's insane. → tweet
@Teknium · 2026-05-15T19:13
Two of these connected can run DeepSeek v4 Flash and one can run Nemotron 120B and Qwen 3.6 27B! → tweet
@TheAhmadOsman · 2026-05-16T17:28
Continual Learning has already been solved, it just requires the model weights to be running locally on your own hardware so big labs are avoiding the topic altogether. Local / Opensource AI will win. Inevitable → tweet
@Teknium · 2026-05-16T09:35
Welcome to the Hermes crew → tweet
@Teknium · 2026-05-16T09:35
So many great projects built around Hermes :) → tweet
AI Agent Ecosystem (Hermes / Grok Integration)
@Teknium · 2026-05-15T19:42
You can now use your SuperGrok subscription in Hermes Agent! Enjoy! → tweet
@Teknium · 2026-05-15T19:58
Welcome to Hermes Crew, Grok! → tweet
@sudoingX · 2026-05-16T06:43
grok just got official oauth inside hermes agent. a month ago i reverse engineered this myself. fingerprint spoofing. cookie workarounds. patchright CDP breakthrough. shipped it public on a fork because there was no official path. asked xAI to build the real one. today they did. the workaround became the product. this is what building in public does. benchmarking both now. → tweet
@sudoingX · 2026-05-16T05:47
a month ago, getting grok to run inside hermes agent meant fingerprint spoofing, cookie workarounds and a patchright CDP breakthrough. i know because i built that provider myself, on a fork, since there was no official path. i shipped it public, flagged the two things that still needed fixing, and openly asked xAI to ship the real one. today they did. official grok subscription inside hermes agent, OAuth through [url], no api key, no browser hacks. → tweet
@sudoingX · 2026-05-16T07:20
hermes agent is already the best on local models. but i'm working on more edges to make it fly even harder. before that, if your agent keeps crashing on local inference here's what to check: max_turns: bump from 30 to 50. gateway_timeout: raise from 600 to 1200. context accumulation: reset between major tasks. if you're running anything under 20 tok/s locally, these three settings are the difference between "broken" and "flying." → tweet
@Teknium · 2026-05-16T03:23
We just overhauled our built in Notion skill to take advantage of the new ntn CLI! Check out the docs: [link] → tweet
Codex & Agentic Coding Infrastructure
@gdb · 2026-05-16T18:25
using codex from the ChatGPT app is such a freeing experience. makes you realize how tethered you normally are to your computer. → tweet
@gdb · 2026-05-16T16:55
the Codex app is in a category of its own. "agentic excel on mac" is an interesting description. → tweet
@gdb · 2026-05-16T13:49
codex for improving computational complexity → tweet
@gdb · 2026-05-15T23:54
run codex on every commit → tweet
@steipete · 2026-05-15T21:48
People freaking out over my AI spend. What nobody sees: Part of what excites me so much about working on OpenClaw is that I'm trying to answer the question: How would we build software in the future if tokens don't matter? We constant run ~100 codex in the cloud, reviewing every PR, every issue. [... full list of agent deployments ...] All that automation allows us to run this project extremely lean. → tweet
@steipete · 2026-05-16T14:33
Try [url] on one of your repos and let codex work its magic. It's amazing at uncovering bugs you didn't know you had. → tweet
@steipete · 2026-05-16T14:35
Lossless is a really interesting concept for OpenClaw to have an "infinite" context window/memory. It compacts conversations in blocks that the model can refer to, building a tree to look up past messages. → tweet
OpenCode & Developer Tooling
@thdxr · 2026-05-16T04:08
OpenCode's worktree feature will ship next week out of experimental once we finish one feature that pairs really nicely with it. what do you think that feature is? → tweet
@thdxr · 2026-05-16T04:05
in OpenCode 1.50.1 you can pin sessions. i've been using this with the experimental /warp feature to run each PR i'm working on in its own git worktree and pinning it. when i start my day i pickup from what's on this list → tweet
@thdxr · 2026-05-16T03:11
one pattern we could do is the first time you run opencode it forks into the background as a server. then every other time you launch it or use the webapp or desktop app they all use that one instance so everything is synced. i'm worried this feels unexpected to people though → tweet
@steipete · 2026-05-16T16:23
BlackBar 0.2.0 is live for @useblacksmith 📈 24h vCPU + workflow graphs 🔔 opt-in status/job notifications 🧰 richer Blacksmith job rows 🟢 compact status badge. Tiny menu bar, less CI guesswork. → tweet
@steipete · 2026-05-15T21:38
Been using @sveltejs for a few projects lately, it's quite a nice alternative to React, fewer gotchas and complexity and Codex handles it really well. → tweet
@ASalvadorini · 2026-05-16T10:28
Dart. Next? #flutter #Flutterdev #dart 🎯 → tweet
AI Industry & Research Funding
@LinusEkenstam · 2026-05-15T23:49
Massive news from AI and biotech: Google DeepMind co-founder Demis Hassabis has just raised $2.1 billion for Isomorphic Labs. For years, Hassabis and his team have been building powerful AI that can predict protein structures, design brand new molecules, and dramatically speed up drug discovery. Now Isomorphic Labs is turning that technology into a full scale effort to tackle what traditional medicine has struggled with for decades. → tweet
@TrungTPhan · 2026-05-16T18:33
Peter Jackson echoing James Cameron take on AI being a tool. Cameron says an opportunity with AI-generated videos is to make more blockbuster films by cutting "in half" the cost of computer-generated graphics. Cost-saving isn't about "laying off VFX staff" but about "doubling their speed to completion on a given shot." → tweet
@sudoingX · 2026-05-16T09:48
anyone building solo, i think the hardest part was never the product, it was everything around the launch [...] i ran my real launch brief through Higgsfield Supercomputer and that gap is a lot smaller than i thought. you hand it one prompt, a product and a direction, and it plans the whole campaign itself, breaks the work into sub-tasks, sends each piece to whatever frontier model is strongest for it [...] → tweet
LLM APIs, Cost & Telemetry
@thdxr · 2026-05-16T05:26
LLM APIs need to return cost information in their response alongside tokens. literally everyone is using models[dot]dev data to approximate this - we see so many reqs to its api. but this is just sticker pricing, won't reflect discounts, etc so it doesn't really work → tweet
@badlogicgames · 2026-05-16T18:50
was trying to hunt down auto-complete lag issues on VS Code. turns out if you enable the GH Copilot extension, it will send a lot of funny telemetry. "Predict the next code edit based on user context, following Microsoft content policies and avoiding copyright violations." (agent just instrumented the js of the extension, it's fun!) → tweet
Programming Languages & Security
@badlogicgames · 2026-05-16T13:19
triangle company created a new "system" programming language "for agents" called zero. i love me some new PLs. it's very cute. looks like the mach-o emitter is broken tho. at least from source. → tweet
@gdb · 2026-05-16T17:04
using GPT for defensive security → tweet
@hnasr · 2026-05-15T19:32
Socket Management in Backend System Design [link] → tweet
Developer Sentiment & Critique
@sudoingX · 2026-05-16T02:45
Claude code is so buggy now. what happened ants? → tweet
@sudoingX · 2026-05-16T06:06
anthropic was the loudest about fear and wiping entire software engineering jobs and meanwhile the tools are broken lol. i honestly can't wait for the ai bubble to burst, hardware to get cheap, and open source to thrive → tweet
@sudoingX · 2026-05-16T03:53
bro how are we replacing entire software engineers with this if the replacement needs a replacement → tweet
@levelsio · 2026-05-16T13:27
How do I tokenmax my Claude Code? → tweet
Agent Lifestyle & Workflows
@sudoingX · 2026-05-16T17:56
1am now. agents running. laptop beside the bed. going to rest now while they cook. something will be ready when i wake up. i keep it beside me because when i get up for water at 3am i always end up prompting. goodnight anon. → tweet
@Ex0byt · 2026-05-16T18:18
prime Peloponnesian trap: If these two brains joined forces, what could we get? → tweet