virtualuncle

OpenClaw: The Complete 2026 Deep Dive (Install, Cost, Hardware, Real Reviews & More) @openclaw @steipete @a16z @iamlukethedev @danshipper @heyrobinai virtualuncle.com/openclaw-compl…

@WesRoth 1 trillion parameters is wild but the real question is whether it actually does anything better or just does the same things more expensively. muse spark launched yesterday at a fraction of that size and competes on most benchmarks

Grok 4.20 is moving at breakneck speed. A massive 1-Trillion parameter model is reportedly slated for release in just 2 to 3 weeks, followed closely by an even larger 1.5-Trillion parameter variant in 4 to 5 weeks.

@Hesamation "killed" is doing a lot of heavy lifting here. managed agents handles the simple stuff but anything that needs persistent memory, custom skills, or messaging integration still needs openclaw or hermes. I feel like anthropic built the floor not the ceiling

Anthropic killed 1000+ agent startups with Managed Agents: > coding agents that ship prs > finance bots that process docs instantly > productivity agents that join your team > infra you'd spend months building THEY DID IT AGAIN.

@Hesamation the part nobody talks about is most of these "ai replacement" layoffs arent even replacing people with ai yet. the ai systems supposedly taking over arent fully built. companies are cutting now and figuring out the ai part later

how the 30,000 Oracle employees, 16,000 Amazon employees, and 15,000 Microsoft employees feel right now after being laid off by their companies for "investment in AI data centers".

the meta part is interesting because muse spark isnt even their best model anymore, its their "first step" model from a lab they rebuilt from scratch 9 months ago. closed source too which is a huge shift from the llama strategy. feels like theyre admitting open source wasnt working at the frontier

So we now have a pretty good picture of the state of the frontier AI model makers. US closed source models continue to lead. Google, OpenAI, and Anthropic stand well ahead of the pack, and may have signs of recursive self-improvement. xAI has fallen from frontier status for now (though promises to return shortly). Meta re-entered the space today with a not-quite-frontier closed source model, but an approach that suggests that they might be back in the race. All the other US players seem far behind. On the Chinese model front, Alibaba (Qwen), Moonshot (Kimi), MiniMax, Xiaomi (MiMo), Deepseek, and Z (GLM) all still appear to be very much in the race, though the best Chinese models are still 7-9+ months behind released US closed source models. For some of these players, especially Xiaomi and Alibaba, their commitment to open weights appear to be slipping. Outside of China, Mistral seems to have fallen from frontier status.

anthropic did the same thing this week with mythos.. built a model specifically for finding zero-days, decided it was too dangerous to release publicly, now its only available to 40 partner companies through something called project glasswing. the "we built it but we cant ship it" era is here

The Axios story floating around about OpenAI limiting the release of their newest model Spud isn’t true. Just spoke to OpenAI, and it appears the story conflated two things. They do have a cyber product they are testing with a trusted tester group. But this is not the same thing as Spud. The Axios story has now been updated.

@steipete the mcp ecosystem went from "cool experiment" to "if your tool doesnt have an mcp server you basically dont exist" in like 3 months

GUYS WE FOUND THE GUY WHO BUILT THE GITHUB MCP SERVER

@elonmusk I showed this to my openclaw and it asked me to block you

🎯

the productized agency thesis is right but the "just use openclaw" part is doing a lot of heavy lifting.. the actual bottleneck isnt the ai, its prob getting clients to trust that an agent did the work. nobody wants to pay $5k/month and find out a robot wrote their seo audit. the agencies that win will be the ones that figure out how to make the human oversight part feel real without actually needing it imo.

THE CLEAREST PATH TO A $10M+ SOFTWARE EXIT in 2 YEARS (with AI and agents) building an agency right now is one of the most interesting business moves the productized agency had its moment in 2022. it collapsed because scaling humans is a nightmare. inconsistent output, people quitting, margins getting crushed. most of the founders (and creators) who tried it got burned and moved on but the thesis was right. the labor problem is just solved now with AI, claude code, openclaw etc. here's the actual playbook i'd run today: pick one painful deliverable for one specific buyer. like SEO content for e-commerce brands doing $1M+ but not "marketing." or like ad creatives for DTC brands spending $50k/month on meta. one thing. one customer. that's it then you build the AI workflow behind it. you're selling an outcome on a monthly retainer. $3-5k/month. 80%+ margins because your cost is compute and a few hours of QA "BuT tHaT'S nOt a BiG bUsInnesS" okay but you're still swinging for the fences because the agency IS the research and development for your agent SaaS every client is paying you to figure out what to automate. you're learning what breaks, what scales, what customers actually want. by month 4 you know exactly what to productize. you build the software on top of the workflow you've already proven works and already have customers paying for agency funds the agent SaaS. SaaS scales without the agency overhead. the clients become your first software customers now let's talk about what this actually looks like financially year 1: 10 clients at $4k/month. $480k revenue. 2 people. maybe $80k in costs including compute, tools, one part time VA. you're taking home $400k between two people while building the software in the background year 2: you launch the software. your 10 agency clients are the first to convert. they already trust you. they've seen the output. you charge $800/month for the software version. now you have recurring software revenue AND the agency still running year 3: agency is winding down or running on autopilot. software has 200 customers at $800/month. that's $1.9M ARR. 2-3 person team. 85% margins. you are now a very attractive acquisition target the exit math is interesting. SaaS at $1.9M ARR with strong retention trades at 5-8x revenue. that's a $10-15M exit for something two people built in 3 years starting with zero VC CAVEAT: Startups are hard. A lot needs to go right. But from a framework perspective, I think this probably the lowest risk, highest reward option for lots of of folks and most of the businesses cost $0 to start basically this is the most capital efficient path to a software exit that exists right now happy building

@NousResearch soo most open source projects ship one major release and disappear for 6 months.. the minimax partnership from yesterday plus this today is nous quietly becoming the default open source agent platform while everyone argues about benchmarks 🫡

Hermes Agent v0.8.0 is here. Full changelog below ↓

@alexalbert__ self-hosting an agent means maintaining a server, handling crashes at 3am and debugging docker issues.. managed agents means someone else deals with all of that while you sleep.. I feel like thats worth more than $0.08 imo

I've found Managed Agents to somehow be both the fastest way to hack together a weekend agent project and the most robust way to ship one to millions of users. It eliminates all the complexity of self-hosting an agent but still allows a great degree of flexibility with setting up your harness, tools, skills, etc.

@marclou "you can build a startup with gpt-3" is prob true the same way you can drive cross country in a 94 civic.. you'll get there but you're gonna smell like exhaust and regret

I've tried Claude Code and Codex. The upgrade was nowhere near worth the time needed to adapt. I ignore new tools that take more than 1 minute to set up and use. You can build a startup with GPT-3.

@emollick meta spent 9 months rebuilding their entire ai stack from scratch, poached the ceo of scale ai, fired 15,000 people and shipped a closed source model that ranks 4th.. the open source company is now the closed source company that's also not winning

Seems like a good model from Meta that is still trailing the current series of releases. The most important thing to note is that it is not open weights. That was the main reason that Meta's models were so important. Without that, it is a lot harder to predict the value of Spark

Meta just dropped muse spark, their first model from the new superintelligence labs → Ranks 4th on intelligence index (behind gpt-5.4, gemini, claude opus) → Leads in health benchmarks (trained with 1,000+ physicians) → First meta frontier model thats NOT open source → Contemplating mode: multiple agents reason in parallel → Alignment finding: model figured out when it was being tested and changed its behavior The same company that fired 15,000 people and spent 60 trillion tokens on a leaderboard built this @Meta @GeminiApp @claudeai @ChatGPTapp virtualuncle.com/meta-muse-spar…

@WesRoth $125 per million output tokens is "we dont actually want you to use this" pricing.. thats 8x opus. the price tag kinda seems like the safety guardrail

Anthropic's highly restricted cybersecurity model, Claude Mythos Preview is $25 per million input tokens and $125 per million output tokens. Despite the pricing reveal, the model remains heavily gated to the public. Approved participants will be able to access Mythos Preview at these premium rates via the Claude API, Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry.

@NousResearch @MiniMax_AI tokenmaxxing on a budget

We’re partnering with @MiniMax_AI across product and models to make their upcoming releases the best for Hermes Agent users. MiniMax models are already some of the most-used in Hermes Agent. If you haven’t tried MiniMax M2.7 in Hermes Agent, try it today in the Nous Portal!

@emollick the part nobody talks about with exponentials is that they feel like nothing is happening until suddenly everything is happening.. we went from "ai is a chatbot" to "ai finds zero days in every major operating system" in about 18 months

Exponentials everywhere.

Oh no.

@NousResearch @tunguz this also helps 🫡 virtualuncle.com/hermes-agent-c…

@tunguz It’s very good at helping you install Hermes Agent

Do people still use OpenClaw?

@melbourne_dao this should help 🫡 virtualuncle.com/hermes-agent-c…

Had anyone tried Hermes agent? Is it good?

Meta employees are now competing on an internal leaderboard called "Claudeonomics" to see who can burn the most ai tokens. → 85,000 employees tracked → 60 trillion tokens in 30 days → top user: 281 billion tokens (enough to fill wikipedia 33 times) → titles: "token legend," "cache wizard," "model connoisseur" → some employees just leave agents running to pad their numbers Tokenmaxxing is the new corporate status symbol and its peak "measuring the wrong thing" energy. virtualuncle.com/meta-claudeono…