Executive Summary
This period saw significant developments in AI coding tools and local AI deployment. Hermes Agent emerged as a top-ranked general AI agent on OpenRouter, surpassing OpenClaw, with new LINE integration announced. Local AI continues gaining momentum with successful meetups and optimized models running on consumer hardware like the DGX Spark and Mac Studio. Developer tooling advances included no-mistakes v1.15.0 with intent-aware validation, DS4 optimizations, and new Cloudflare Email Sending capabilities. The AI infrastructure layer is maturing rapidly with abstraction libraries and multi-provider support becoming standard.
Key Events
- Hermes Agent reaches #1 on OpenRouter, surpassing OpenClaw in global rankings, with LINE integration now available as an official messaging channel. → tweet
- Qwen3.6-27B-PRISM-PRO weights released with dynamic quantization options (PRISM, NVFP4). → tweet
- Cloudflare Email Sending launched with pricing comparison showing 3-5x cost reduction vs Postmark/Resend. → tweet
- no-mistakes v1.15.0 released with agent session tracking for intent-aware code validation. → tweet
- Local AI Meetup held successfully in SF Presidio, highlighting growing community interest in on-device AI. → tweet
- step-3.5-flash-REAP-121B optimized for DGX Spark with full 256k context at 17-18 tok/s. → tweet
- OpenCode LLM provider abstraction library in development, targeting multi-provider quirks handling at scale. → tweet
Analysis
Patterns & Trends
Rise of Local AI on Consumer Hardware: The DGX Spark and Mac Studio (M-series chips) are enabling 121B parameter models to run locally with meaningful performance. The step-3.5 REAP model achieving 17-18 tok/s on 256k context demonstrates hybrid-SWA architectures are making large-context inference viable on consumer devices.
Agent Competition Intensifying: Hermes Agent's #1 ranking over OpenClaw signals shifting competitive dynamics in the general AI agent space. Multi-channel support (Discord, LINE) and OpenRouter integration are becoming differentiators.
AI Coding Tool Maturation: Tools like OpenCode, Codex, and BlackBar are shipping rapidly with features like subagent visualization, video proof generation, and streaming status indicators. The focus is shifting from basic code generation to developer experience and workflow integration.
Infrastructure Commoditization: Email sending and LLM inference are following similar commodity trajectories. Cloudflare's entry into transactional email and the emergence of provider abstraction layers indicate the infrastructure layer is becoming standardized.
Escalation/De-escalation
No significant escalation observed. Competition remains healthy with multiple viable options in agents, models, and infrastructure.
What to Watch
- Hermes Agent adoption trajectory following #1 ranking
- Local AI community growth post-meetups
- Cloudflare Email beta feedback and stability
- DS4 on DGX Spark optimization progress
- Flutter WebGPU implementation timeline
Tweet Feed
AI Agents & Tools
@sudoingX · 2026-05-10T14:18
i love my dgx spark. this thing just completed an entire codebase with a full test suite while i was arguing on twitter. thank you nvidia. i wouldn't mind another one. → tweet
@sudoingX · 2026-05-10T14:13
hermes agent just hit #1 globally on openrouter above openclaw. but sure, bots did that. → tweet
@Teknium · 2026-05-10T14:03
Hermes Agent now has a new gateway channel! LINE is now an officially supported way to interact with your agent! → tweet
@nummanali · 2026-05-09T22:22
Claude Managed Agents is really good But we need an open source solution Haven't found any options so might need to build myself! Anyone else under this dilemma? → tweet
@steipete · 2026-05-10T12:45
Slop one-shot websites, 2025 vs 2026. → tweet
@steipete · 2026-05-10T12:48
RT @reach_vb: codex → tweet
@steipete · 2026-05-10T11:40
Slaps. → tweet
@steipete · 2026-05-10T11:59
Built BlackBar, a menubar for @useblacksmith → tweet
@steipete · 2026-05-10T03:11
RT @OpenAI: Just gonna leave this here. → tweet
Local AI & Hardware Optimization
@sudoingX · 2026-05-10T08:40
update 1 hour later: pushed it to FULL 256k context on my dgx spark. zero speed loss. now holding 17-18 tok/s on long convos... biggest lesson for me: the conservative defaults from the qwen3 era don't apply to hybrid-SWA models. step-3.5 lets you pay full-context price once and use it forever. still criminally underrated. 121B params, full 256k context, wild. → tweet
@sudoingX · 2026-05-10T07:41
this is criminally underrated for all dgx spark owners. hermes agent is running my new fav a 121 billion parameter model on my dgx spark right now. just swapped to step-3.5-flash-REAP-121B at Q6_K... speed: 12.21 tokens per second on streaming completion. → tweet
@nummanali · 2026-05-10T13:50
Mac Studio isn't the game It's the M3/4/5 in MacBook Pros with 32GB Ram+ With Qwen 3.6 and Gemma 4 you get private unlimited sonnet 4.6 perf at 70+ tok/s with MPS Millions already have this, the ecosystem will explode this year Dense intelligence is real now → tweet
@TheAhmadOsman · 2026-05-10T03:45
The first Local AI Get-Together was a massive success... Local AI is very real, very alive, and apparently willing to talk GPUs, open weights, inference engines, agents, and homelabs for hours → tweet
@TheAhmadOsman · 2026-05-09T21:08
Imagine seeing how capable Qwen 3.6 27B is when you give it web access and a proper harness and not yet getting that AGI will run locally Ngmi → tweet
Model Releases & Research
@Ex0byt · 2026-05-10T14:32
Qwen3.6-27B-PRISM-PRO 🧑🍳 Weights are ready. Which quant do you want? -> PRISM Dynamic Quant -> NVFP4 -> Full Weights? → tweet
@Ex0byt · 2026-05-10T18:42
Project LookingGlass update: Watching my (money-printing) local AGI Swarm Intelligence Engine debate a Polymarket prediction in real time is one of the coolest projects I'm currently working on. → tweet
@badlogicgames · 2026-05-10T15:34
RT @Dimillian: Well Code cooked, Doom in Swift is almost a 100% accurate rendering now. → tweet
@badlogicgames · 2026-05-09T23:33
RT @pradeep24: tested out @antirez' ds4.c this morning. so impressive and delivers. on a M3 max, 128GB, stock ds4 settings: 14–15 t/s → tweet
@badlogicgames · 2026-05-09T20:07
calling it slopex from now on so it can join its sibling slopus. → tweet
@sama · 2026-05-09T19:16
5.5 is an autistic genius with very strange taste in naming shocking that we would make such a thing → tweet
@sama · 2026-05-09T19:12
kicking off a bunch of codex tasks, running around with my kid in the sunshine, and then coming back at naptime to find them all completed makes me very optimistic for the future → tweet
Developer Tools & Frameworks
@thdxr · 2026-05-10T18:58
when you use a gpt model in opencode it swaps to using a more powerful patch tool for edits vs simple find and replace LLMs aren't like people they don't need to do things linearly i've seen gpt do many things in parallel in a single patch call → tweet
@thdxr · 2026-05-10T01:45
we're working on a library to abstract over all the llm providers there's very few teams that have dealt with the quirks between providers at the scale we have it's written in effect but will also have a vanilla api progress is in the opencode repo under packages/llm → tweet
@kunchenguid · 2026-05-10T17:04
no-mistakes v1.15.0 just released with a major update when no-mistakes starts to validate a code change, it will automatically find the agent session that made the change to understand your original intent this allows no-mistakes to not only see "what" changed, but also "why" → tweet
@kunchenguid · 2026-05-10T00:02
many people asked about the cost of running the realtime voice model here just released autopreso v0.1.5 with bug fixes and a new realtime cost tracker on the UI so you don't get surprised → tweet
@steipete · 2026-05-10T16:00
🧭 gogcli 0.16.0 is out Workspace admin grew up: - create/delete users, aliases, temp passwords - org units - Meet, Sites, YouTube, GA4/Search Console - Drive changes/activity A lot more Google API from one boring binary. → tweet
@steipete · 2026-05-10T10:21
We now have video proof generation for issues on OpenClaw as part of working on QA automation. Codex [or a GH workflow] generates before/afters (crabbox does the screen recording). → tweet
@steipete · 2026-05-10T07:22
Did teach codex to look for social signals when reviewing PRs. → tweet
@steipete · 2026-05-10T03:06
Latest spogo (Spotify cli) is much faster, codex is my dj now. → tweet
@badlogicgames · 2026-05-10T16:19
RT @bentlegen: New version of hunk is out (0.11.0) with: - support for jujutsu (see docs) - sidebar in pager mode - works w/ git log -p → tweet
@badlogicgames · 2026-05-10T16:19
RT @antirez: Appreciate Ivan tweet. To put this into context, to build DS4 I used: my MacBook M3 Max (mine, 8k euros), 1 M3 Ultra with 512... → tweet
@badlogicgames · 2026-05-09T23:52
RT @Dwsyyy: This release dramatically reduces disk and database I/O across the board, making PSM faster, lighter,... → tweet
@badlogicgames · 2026-05-09T11:43
RT @fu5ha: People of /tree, I present pi-treebase. I love /tree, but wanted even more control. /treebase lets you inter... → tweet
@ASalvadorini · 2026-05-10T17:23
flutter devs, if you're not following this account, you definitely should. I'm 🤯 by his micro-animations, they're something you rarely see around 🔥
@MatejKnopp · 2026-05-10T16:49
It's time to admit: I'm just too stupid for sliver geometry. Spent weekend to get virtualized nested slivers working but failed miserably. Maybe it's time for a custom render object geometry that won't be frying my brain as much. → tweet
@jezell · 2026-05-09T19:08
Had to dig through some undocumented streaming events, but patch tool live status is wired up. Flutter web can just do things. → tweet
@gdb · 2026-05-09T21:11
Codex for expenses → tweet
Infrastructure & Cloud
@levelsio · 2026-05-10T14:29
If you wanna switch to @Cloudflare Email Sending today, here's my prompt for you... Prompt: Migrate transactional email to Cloudflare Email Service → tweet
@levelsio · 2026-05-10T13:25
✉️ Trying @Cloudflare's new Email Sending feature today If you send 1,000,000 emails per month: Postmark: $1,206/mo Resend: $650/mo SendGrid: $600/mo Cloudflare: $354/mo Amazon SES: $100/mo TL;DR email sending has become a commodity! → tweet
@levelsio · 2026-05-10T13:29
Resend is literally just SES with a 500% markup @grok fact check → tweet
@levelsio · 2026-05-10T14:57
Important tip before you migrate to Cloudflare Email! Migrate your suppression list → tweet
@thdxr · 2026-05-09T19:12
aws spent at least a decade acquiring a bunch of customers who pay by cpu time before they launched serverless options... with LLM inference the idea of serverless is here on day 1 - we all pretty much want to pay by token but the providers do not have enough scale to actually offer that product well and idle GPUs are a lot more expensive → tweet
@thdxr · 2026-05-09T19:38
some of these new ai infra startups have raised a lot of money - eg nebius raised $4B this feels like a lot but google is spending $180-$190B this year → tweet
@FrameworkPuter · 2026-05-10T16:07
RT @svpino: The Framework 13 Pro keyboard is better than both my MacBook Pro 16 and my MacBook Air keyboards. I honestly don't want to go… → tweet
Engineering Insights & Opinions
@hnasr · 2026-05-10T14:01
Trying to explain a brilliant data structure (such as bloom filter) often falls flat and unappreciated... You need to be in the bowels of engineering. You need to run into the inefficiency, the complexity of the system... Only then you will invent a novel way to solve it. Naturally... No AI, no senior engineers, no stack overflow. This is just you, the art of innovation, And perhaps a little perseverance. → tweet
@LinusEkenstam · 2026-05-10T09:16
I can understand why this does not work right now... market is not ready for it, most people are illiterate in the new paradigm... if you're a mass market appeal product you now face a very difficult task. how to keep staying innovative and fresh, while maintaining growth and profitability. → tweet
@levelsio · 2026-05-10T18:45
The most important people in tech you want to meet are almost all on here, and you can just tweet at them They're not at network events! → tweet
@levelsio · 2026-05-10T17:36
If I'd have to get a job I'd wanna work at @spacex @xai @x or @cloudflare or @anthropic → tweet
@badlogicgames · 2026-05-09T07:39
guess how fun it is having all of the openclaw user base beat up pi's llm provider abstraction. guess i'm "one of the very few teams that have dealt with the quirks between providers at scale" now ... → tweet
@badlogicgames · 2026-05-09T09:06
recommended viewing. in anders we trust. → tweet
@steipete · 2026-05-10T04:21
Crabbox now has great Windows terminal handling. So good that codex could E2E fix gifgrep to render animated gifs in the terminal. Just because it can. → tweet
Industry News
@steipete · 2026-05-10T04:46
RT @reach_vb: in the last ~15 days we shipped: - gpt image 2 - privacy filter - gpt 5.5 - gpt 5.5 pro - gpt 5.5 instant - gpt realtime 2 -… → tweet
@TrungTPhan · 2026-05-09T23:29
Still incredible that the DeepMind documentary has footage of exact moment Demis is told that AlphaFold can "easily" predict all known (1-2B) protein sequences "in a month" and he says to do it. → tweet
@TrungTPhan · 2026-05-09T20:02
Apple reached a $250m settlement for falsely advertising Apple Intelligence features on the iPhone in 2024. → tweet
@jezell · 2026-05-09T23:35
RT @mark_k: OpenAI has announced they will be winding down fine tuning. I got the email today. Existing active @OpenAI customers can keep r… → tweet
@jezell · 2026-05-09T23:34
Hey @Apple, we're gonna need more DRAM in the M6. → tweet
@thdxr · 2026-05-10T04:45
the most tiring thing about programmers is they can't just like something, they have to go come up with theories on why people who like something else are wrong i don't care man i'm getting as much done as you it clearly doesn't matter → tweet