Executive Summary
The past 24 hours were dominated by the rapid evolution of open-source AI agent frameworks, with NousResearch's Hermes Agent seeing major updates (v0.2.0 LCM, v0.9.0 with Termux/Android and iMessage support) and growing community adoption as an open-source alternative to proprietary tools. Hugging Face launched "Kernels on the Hub," enabling GPU kernel distribution as easily as model pushes—a significant infrastructure shift. Meanwhile, enterprise AI governance emerged as a critical theme, with multiple voices highlighting that raw LLM coding tools lack audit, permissions, and visibility, fueling demand for platforms like Superblocks. The debate over Anthropic vs. OpenAI enterprise positioning intensified as Azure handcuffs on OpenAI loosen, and Google's internal AI adoption came under scrutiny.
Key Events
- Hugging Face launches Kernels Hub: GPU kernels can now be shipped as easily as pushing a model, with pre-compiled binaries and agent-driven skill deployment via
pip install kernels. → link - Hermes Agent v0.9.0 released: Adds local web dashboard, Termux/Android native support, iMessage (BlueBubbles) integration, eliminating manual config editing. → link
- hermes-lcm v0.2.0 released: Lossless context management with hierarchical DAG-based message persistence for Hermes Agent. → link
- Enterprise AI governance gap identified: Raw Claude Code lacks audit trails, permissions, and central visibility; Superblocks positions as governance layer already used by Instacart, SoFi, LinkedIn. → link
- OpenAI/Anthropic enterprise battle shifts: Anthropic's enterprise revenue partly driven by Azure deal constraining OpenAI; those handcuffs now coming off. → link
- Google AI adoption debate: Steve Yegge's claim of average adoption vs. DeepMind CEO's rebuttal; key insight: weekly active users is a red flag metric—"average concurrently running agents per employee" is more meaningful. → link
- Dockerhub under strain: ~30% build failure rate on pulls as vibe coding scales infrastructure demand. → link
- opencode 2.0 data model breakthrough: Team reports significant progress on embeddable data model for next version. → link
Analysis
Patterns observed: - Open-source agent frameworks are eating proprietary tooling: Hermes Agent's rapid community skill contributions, local-first architecture, and Termux support contrast with complaints about OpenClaw's "TypeScript bloat" and corporate telemetry. The open-source vs. proprietary agent war is intensifying. - Governance is the enterprise bottleneck: Multiple voices (LinusEkenstam, carlvellotti) independently flagged that vibe-coded software is accumulating inside enterprises with zero oversight. This gap will define the next enterprise SaaS wave. - Reasoning effort calibration matters more than token counting: GergelyOrosz and kunchenguid both pushed back on token-burn metrics; high reasoning upfront often saves tokens by avoiding rework—reframing how AI productivity should be measured.
Escalation signals: - Infrastructure strain (Dockerhub, Cloudflare billing opacity) suggests vibe coding is hitting real operational scale. - Model providers building app copies (Anthropic→Lovable, Google→ecosystem) signals platform encroachment on AI-native startups.
What to watch next: - Hermes Agent Jam (tomorrow) could produce significant new skills/plugins. - Whether Dockerhub degradation forces broader shift to alternative registries. - Anthropic's enterprise positioning as OpenAI's Azure exclusivity loosens. - Gemma 4's tool call reliability fix—could reshape local model agent suitability.
Tweet Feed
AI Agent Frameworks & Tools (Hermes, Claude Code, Pi, OpenClaw)
@Teknium · 2026-04-14T18:31
RT @SteveSchoettler: hermes-lcm v0.2.0 is out! Lossless context management for Hermes Agent — every message persisted, hierarchical DAG s… → tweet link
@Teknium · 2026-04-14T18:26
Awesome video about Hermes Agent! https://t.co/xU3CUUQDcg → tweet link
@Teknium · 2026-04-14T17:09
RT @jumperz: honestly i'm enjoying hermes agent and what's cool about it is simply that nothing breaks while I'm layering it with more an… → tweet link
@Teknium · 2026-04-14T17:09
Nice quick tutorial on multi agent profiles with Hermes! → tweet link
@Teknium · 2026-04-14T17:09
RT @TencentAI_News: For anyone running @NousResearch Hermes Agent locally and wishing it just stayed online: there's now a one-click deploy… → tweet link
@Teknium · 2026-04-14T06:04
RT @shangdu2005: 今日Hermes Agent v0.9.0 发布！本地 Web 仪表板来了！不用再手改配置文件了。📱 Termux/Android 原生支持 – 手机上跑 Agent。💬 新平台 – iMessage (BlueBubbles) +… → tweet link
@Teknium · 2026-04-14T09:31
RT @lumao_2026: 看到一个老外在健身房用 Hermes 查健身动作和营养数据，干脆写了个 fitness skill，被 Teknium 直接合并进官方代码库了 💪 这就是 Hermes 生态的魅力——你写一个 skill 解决自己的需求，社区觉得好就直接收录。… → tweet link
@Teknium · 2026-04-14T09:28
RT @ddkydesign: don (my AI agent) now helps me setup design systems for my next saas project the flow: → i chat with him → he sets up stor… → tweet link
@Teknium · 2026-04-14T07:13
RT @macbethAI: Creative Direct with Hermes https://t.co/Pn0IuMgwu0 → tweet link
@Teknium · 2026-04-14T06:07
RT @AndeanViber: @Teknium I just ripped out OC and replaced it with Hermes, easier to sell to corporates imo (safer) → tweet link
@Teknium · 2026-04-14T06:16
RT @aijoey: Hermes Agent Jam is tomorrow, excited to see what the Nous team presents and answer any questions in the Q&A Building the Herme… → tweet link
@Teknium · 2026-04-14T05:59
If you're in South Korea, come check out the Hermes Agent meetup! → tweet link
@sudoingX · 2026-04-14T05:54
first local model running on the rog 5090 mobile. gemma 2 2b it q4_k_m through llama.cpp, hermes agent pointing at the local endpoint. sitting on the balcony watching the beach while the model answers prompts. 24gb vram in a portable machine running a full inference stack plus an agent harness. same class as the desktop 3090 i benchmark on, now outside with coffee. the hermes agent local model auto-detection i pushed a while back just resolved the endpoint on its own. good time seeing some my own code work in the wild, on my own hardware. → tweet link
@sudoingX · 2026-04-14T03:27
tek the goat. while openclaw ships more typescript bloat and phones home to corporate servers, hermes agent moves at god speed. open source in and out. no open ai salesmen pretending local ai doesn't work. anon, did you drop the bloat yet? or still waiting for permission? → tweet link
@gospaceport · 2026-04-13T19:33
Hoarders of old Android phones we are finally have our day 😝 Full Hermes support via Termux! I can replace my crazy usb extension (flakey) with just an old ass phone now. Stick your agent anywhere, including spider infested wells! Major W! → tweet link
@victormustar · 2026-04-14T13:39
NEW: Claude Code is now writing Python inside JS files... I recommend everyone to give a chance to GLM-5.1 + Pi https://t.co/vZpQFJguFz → tweet link
@nummanali · 2026-04-14T16:53
How are the offering literal cloud agent orchestration for free. You configure it from your computer and any remote connected context is available when it's run. Not a feature I need but definitely impressed by the dynamism of the Claude Code team → tweet link
@nummanali · 2026-04-14T16:06
I've been deep on cmux for the last two months. But this new Warp updates looks extremely well throughout and towards increasing effective collaboration with your coding agents. Will update and try today → tweet link
@carlvellotti · 2026-04-14T16:17
I literally sell a course teaching PMs to build with Claude Code. I'll tell you straight: raw Claude is not an enterprise build platform. It has no audit trail, no permissions model, no central visibility. It builds. It does not govern. Superblocks solves this better than anything I've seen. → tweet link
@carlvellotti · 2026-04-14T15:28
The CRITICAL question to improve AI outputs: "What tools do you need to check your own work?" [...] A completely separate validation pass. AI just doesn't have the ability to reflect on its work as it does the work. You MUST have it test its work. → tweet link
@badlogicgames · 2026-04-14T12:15
look at me shilling @AmpCode at the agentic coding workshop i give today → tweet link
@thdxr · 2026-04-14T02:58
team had a pretty good breakthrough for the data model behind opencode 2.0 today. it's gonna be so good to built on top of and embed anywhere → tweet link
@steipete · 2026-04-14T16:42
RT @cherry_mx_reds: 👋 A few useful OpenClaw 2026.4.14 changes that didn't make the main tweet 🦞 ✨ GPT-5.4 runs are less likely to stall or… → tweet link
@steipete · 2026-04-14T16:40
RT @jlehman_: lossless-claw 0.9.0 — the "stop touching my cache" release. 🔧 engine id fix everyone's been waiting for. ⏳ compaction defers w… → tweet link
@steipete · 2026-04-14T13:36
This release makes me unreasonably happy since I wasn't involved at all - @vincent_koc and the maintainer team did a great job. I'm back soon to work on OpenClaw, today/tomorrow I'm prepping for @TEDTalks in Vancouver. → tweet link
Hugging Face / GPU Infrastructure
@victormustar · 2026-04-14T15:44
RT @ClementDelangue: Introducing Kernels on the Hugging Face Hub ✨ What if shipping a GPU kernel was as easy as pushing a model? - Pre-co… → tweet link
@victormustar · 2026-04-14T18:12
RT @leopardracer: Everyone said 16GB isn't enough for a 35B model. They were right. Until this one flag. https://t.co/uAcLty6LzE → tweet link
@victormustar · 2026-04-14T18:12
RT @ben_burtenshaw: les fucking go... for agents to push kernels to the hub, do: > pip install kernels > kernels skills add <start agent… → tweet link
@victormustar · 2026-04-14T09:26
RT @vllm_project: 🚀 Great to see vLLM powering OCR at this scale — Chandra-OCR-2 (5B) serving ~60 papers/hour per L40S across 16 parallel j… → tweet link
@sudoingX · 2026-04-14T15:48
RT @ClementDelangue: Introducing Kernels on the Hugging Face Hub ✨ What if shipping a GPU kernel was as easy as pushing a model? - Pre-co… → tweet link
AI Models & Performance
@TheAhmadOsman · 2026-04-14T17:50
I changed my mind. Gemma 3 31B is better and more capable than Qwen 3.5 27B. Requires better prompting but it's more capable, intelligent, and token efficient → tweet link
@TheAhmadOsman · 2026-04-13T22:17
Best overall opensource LLM is GLM-5.1 > Best openweight model to run at home is MiniMax M2.7 → tweet link
@TheAhmadOsman · 2026-04-14T00:10
people are sleeping on exllamav3. there's so much to do with, and learn from, that inference engine → tweet link
@TheAhmadOsman · 2026-04-14T06:01
Here is an example. I asked Hermes Agent running Qwen 3.5 9B Dense (Unsloth's UD-IQ3_XXS via llama.cpp) about me. Results were retrieved from a locally hosted SearXNG using an MCP I built on the fly for it → tweet link
@TheAhmadOsman · 2026-04-14T02:49
Using local LLMs? Make sure to setup web search for them. Tell your favorite agent to setup SearNg for you. Give that to your local LLMs. Watch them become way more intelligent and efficient → tweet link
@TheAhmadOsman · 2026-04-14T01:45
Seeing a massive jump in interest + a rising number of projects in the opensource AI / LLMs space. How far we've come since 2023. Open-source AI will win → tweet link
@gospaceport · 2026-04-13T19:38
Gemma4 is smart but sucked at tool call reliability. Might be fixed! Excited to check it out when it drops. → tweet link
Enterprise AI & Industry Dynamics
@LinusEkenstam · 2026-04-14T15:19
Everyone can vibe-code now. Which means enterprises are accumulating AI-built software with zero visibility, zero security controls, and zero governance. Superblocks is the only platform that lets enterprises build AND govern AI-built software at scale. → tweet link
@jezell · 2026-04-14T14:31
This has been the case I've been maintaining for a while: Much of Anthropic's enterprise revenue isn't because they are better so much as it is because the Azure deal tied OpenAI's hands and forced enterprises to seek an alternative. Those handcuffs are off now. → tweet link
@kunchenguid · 2026-04-14T04:53
on the drama of "internal AI adoption at Google" [...] anthropic is one of the only companies at meaningful scale that's evidently AI-first [...] reporting AI usage with weekly active users as the metric is a red flag. one of my favorite metrics to look at is - average concurrently running agents per employee → tweet link
@jezell · 2026-04-14T06:00
Dockerhub seems to be showing some signs of strain. Builds randomly failing on docker pulls for a few days now. Status page says 100%, but it's failing like 30% of our builds. → tweet link
@FinansowyUmysl · 2026-04-13T19:13
Anthropic chce stworzyć kopie aplikacji lovable. To kolejny, po Google, duży gracz, który rozbudowuje swój ekosystem i po kolei tworzy kopie produktów. [...] dzięki temu, że są dystrybutorami i właścicielami modeli AI to ich wersję produktu mogą być tańsze i lepsze → tweet link
@TheAhmadOsman · 2026-04-14T08:05
Anthropic will be the first to do an Uber-style surge prices for LLM subscriptions → tweet link
@RydMike · 2026-04-14T07:43
Already cancelled my Anthropic services. [...] The nerfed token limits, default thinking, the capabilities of 4.6 became dogshit and they blame customers for excessive token usage, regarding the cache expiry mess etc… bye bye → tweet link
@kunchenguid · 2026-04-13T23:51
live example for why Anthropic's "adaptive thinking" sometimes ends up nerfing the agent, and why estimating task complexity upfront is fundamentally hard even for us humans → tweet link
@kunchenguid · 2026-04-13T19:21
using high reasoning effort does NOT mean you will use more tokens compared to medium reasoning. [...] if you often use the agent for developing features, I would highly suggest setting reasoning effort to high as default → tweet link
@TrungTPhan · 2026-04-14T18:11
RT @bearlyai: Credo Technology has been interesting AI supply chain play: ▫️Stock is up 13x in the past 5 years to a market cap of $29B… → tweet link
@FinansowyUmysl · 2026-04-14T07:21
Niepopularna opinia: programiści, którzy nie nauczą się używać AI, będą zarabiali jak helpdesk za 3 lata. [...] programista z AI robi robotę za 3-4 programistów bez AI. Spotify ogłosiło, że ich programiści nie napisali ani jednej linijki kodu ręcznie od grudnia. → tweet link
OpenAI & Broader AI Vision
@gdb · 2026-04-13T23:17
The world is transitioning to a compute-powered economy. [...] AI is now on track to bring this same transformation to every other kind of work that people do with a computer. [...] Nearly a billion people are using these systems every week. → tweet link
@badlogicgames · 2026-04-14T14:26
RT @rachelnabors: Something heard more than once at @aiDotEngineer Europe: "learn to say no" (to your agents). → tweet link
@jsuarez · 2026-04-14T15:22
Reinforcement Learning dev with Joseph Suarez https://t.co/yn7lgKsd9Z → tweet link
Flutter & Mobile Dev
@RydMike · 2026-04-14T11:39
This is a surprise #FlutterDev package "get", the most liked package of all packages on pub, commonly known as GetX, is gone from GitHub? Or has it just moved to a new not yet disclosed location? → tweet link
@MatejKnopp · 2026-04-14T11:45
Hot take: Flutter Stable is placebo. Pick a revision on main, run it through QA, test thoroughly, ship it. Stick to it until you need something fixed or updated. Rinse and repeat. → tweet link
@ASalvadorini · 2026-04-14T17:29
Time to switch to Minimal 🔥🔥🔥 #Flutter #Flutterdev → tweet link
@ASalvadorini · 2026-04-14T08:46
I also feel AI fatigue, but it doesn't matter. AI is a tool, and like any tool embrace it if it improves your life and speeds up your product. Same for Flutter, keep it or leave it 🔥 → tweet link
Hardware & Developer Setup
@alexinexxx · 2026-04-14T15:20
so excited, my framework desktop just arrived → tweet link
@alexinexxx · 2026-04-13T20:02
this is my new nvim setup btw → tweet link
@sudoingX · 2026-04-14T15:50
Gpu → tweet link
@sudoingX · 2026-04-14T15:51
RT @yacineMTB: tailscale is probably one of the fastest growing companies right now → tweet link
AI Engineering Best Practices & Metrics
@badlogicgames · 2026-04-14T17:11
RT @GergelyOrosz: Anyone sensible inside the industry is rapidly coming to realize that tokens burned is the most silly thing to track. → tweet link
@TheAhmadOsman · 2026-04-14T03:24
Tip of the day: Always tell your agent to think through the plan comprehensively covering any and all relevant things, and using best practices and up-to-date information as of April 13th, 2026 → tweet link
@badlogicgames · 2026-04-14T13:35
RT @vboykis: New post: On mechanical sympathy https://t.co/PcsWuCJHOt → tweet link
@ASalvadorini · 2026-04-14T11:33
"Half of them just check a box: 'We enabled {Copilot/Cursor} for everyone!' " This hurts, you've no idea how much it hurts 😔😱😭 #AI #Copilot → tweet link
Community & Events
@thdxr · 2026-04-14T13:09
RT @opencode: We are doing our first buildathon in India. It's happening on Apr 19th in Bengaluru. 100 builders, $100k in cash & credits… → tweet link
@levelsio · 2026-04-13T22:11
🧑🚀 Day 11 of the @cursor_ai #vibejam. Proudly sponsored by @cursor_ai + @boltdotnew + @heyglif + @tripoai. Prizes to win (submit before May 1!) 🏆 $25,000 🥈 $10,000 🥉 $5,000 → tweet link
@ollama · 2026-04-14T00:50
RT @osanseviero: See you this Wednesday at the Ollama Gemma Meetup! 💎 → tweet link
@nummanali · 2026-04-13T18:42
Such an endearing write up of the AI Engineer London Conference. Warning: you will FOMO hard → tweet link