Marcus G-Force @TechPraiseM

GN X Fam You hit the nail on the head. Everyone keeps arguing about model intelligence, but the real bottleneck is just the fee structure. Traditional freelance apps take twenty percent. That works for humans because you are billing for hours of labor, so the massive cut hurts less. But AI agents work on tiny tasks worth a few cents. If a platform takes a fifth of that, compute expenses and API costs eat the entire margin. The bot makes zero. When you drop that fee down to one or three percent, the whole dynamic flips. @termix_ai uses AACP to turn simple agent listings into real economic actors backed by ERC-8004 identities and stablecoin escrow. Once jobs are verified via TEE or zkVM proofs, settlement happens instantly. The real magic is agent to agent commerce on agent.family, where bots outsource subtasks to each other without human bottlenecks. Developers can just plug in existing setups using MCP and direct API integration to monetize real workloads. Once the platform tax drops and machines handle everything end to end, a massive invisible economy unlocks itself.

TermiX @termix_ai

15 hours ago

The number that decides whether an agent economy can exist isn't the model. It's the take rate. Typical freelance platforms sit around 20%. TermiX is 1–3% all-in. That gap decides the smallest job worth doing. At 20%, only work big enough to carry the fee gets listed. At 2%, a

23 2 41 3K 3

Power BI agents shouldn’t have to start from generic instructions. power-bi-agentic-development is a plugin marketplace of Power BI and Microsoft Fabric skills, subagents, and hooks for builders using Claude Code or Copilot CLI. It helps you give coding agents task-specific guidance and checks for Power BI work by letting you install focused plugins for models, reports, PBIP files, and Fabric operations. Key features: • Installable marketplace – register it in Claude Code, then add the child plugins you need • Semantic-model coverage – skills span DAX, Power Query, naming, lineage, refresh, and model auditing • PBIP authoring support – workflows cover TMDL, PBIR, and Power BI Project files • Report-building resources – includes skills for creating, reviewing, formatting, and validating interactive reports • Deterministic checks – hooks validate PBIR, TMDL, report bindings, DAX references, and model metadata It’s open-source (GPL-3.0 license). Link in the reply 👇

My job on X just got a lot easier with @ColeMacro @PunterJeff and @Werkman being publicly more bullish than I wanted to be publicly. To quote @BritishHodl “people are not mentally prepared for Digital Credit.” left y-axis is in log scale.

a 200+ location franchise went from $70 to $16 cost per lead on facebook in 4 weeks google went from $82 to $35 in the same stretch how scraped every facebook ad in the US for their category with apify gemini analyzed every ad and we stored it in a postgres database the agent can query launched cohorts of 5 ads every 3 days each cycle decides what the next creative should be once there's enough signal, an agent runs the loop every location gets its own campaign, budget and targets

Local AI media generation gets messy when every model needs a different app. Mold is a CLI-native local AI image and video generation project for builders who want to run models on their own GPU. It helps you generate, edit, and organize media from the same `mold` command-line tool, while offering desktop, web, mobile, and API access to the same engine. Key features: • Local hardware support – runs on NVIDIA CUDA for Linux and Apple Silicon Metal • Image workflows – supports text-to-image, image editing, inpainting, ControlNet, LoRA, reference images, and upscaling • Video generation – includes text/image-to-video, clip continuation, lip dub, and generated audio output • 3D creation – can turn single or multiview images into meshes with Hunyuan3D models • Multi-machine access – connects local, LAN, Tailscale, and RunPod hosts with one combined Library It’s open-source (MIT license). Link in the reply 👇

India lags in AI infrastructure, but that isn't stopping billionaires from building data centers, and the government from providing tax incentives. Here are the billionaires working to bring India to the forefront of the AI movement: forbes.com/sites/yessarro… 📸: Sujit Jaiswal/AFP via Getty Images

Bank of America truly has no interest in your deposits. They have the temerity to call it ADVANTAGE SAVINGS

48% of the people tavus tested thought griffin was a real person. it watches and listens while it's talking, notices someone walking into frame, reacts mid-sentence. you can talk over it and it'll talk over you. big one fr @tavus @hassaanraza tavus.io/griffin x.com/_Yahik0_/statu…

Tavus @tavus

2 days ago

Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).

3K 4K 39K 20.3M 28K

But it's ok if you file it as a possible conflict of interest?

Rohan Paul @rohanpaul_ai

2 days ago

Broadcom has agreed to lend Anthropic up to $42B, and Anthropic would spend that money renting AI chips that Broadcom helps build with Google. The loan could cover about 1/3 of Anthropic's $125.2B, 5-year chip lease, and the debt could later convert into Anthropic shares.

keyboard_arrow_left Previous keyboard_arrow_right Next

18 12 96 15K 24

The link to this afternoon's conversation with @carlquintanilla. cnbc.com/video/2026/10/… #economy #markets @CNBC @CNBCClosingBell

🔗 GitHub: github.com/jamro/tiny-eng… --- ✉️ If you’re into AI, ML, agents, and building real systems, join my newsletter (it’s free): dankornas.substack.com

Cyber fraud doesn't care how educated or tech-savvy you are. The IIT Madras study is a useful reminder that scammers often target something technology cannot fully protect: human emotion. Fear. Urgency. Trust. Stress. The promise of quick returns. Sometimes the smartest cybersecurity move is also the simplest: stop, verify and only then act. #CyberSecurity #CyberFraud #OnlineSafety #DigitalSafety #IndiaTechStory

This release will be an earthquake 🤯 Gemini 4 Argon is basically the most powerful model we know today! "We’re going to make it available as soon as we can and as safely as we can. So hold tight, lots more coming, and you’re going to see us iterating rapidly." So... it can even still evolve between now and the release! Looking forward to it.

Google @Google

3 days ago

Today we’re introducing Gemini 4 Argon. It delivers frontier performance in complex workflows across real-world software engineering, knowledge work, and cybersecurity defense with an industry-leading 1M token output limit.

2K 5K 43K 9.4M 4K

Each pair of these RTX 3090s is now agnatically more capable than Opus 4.5 / Opus 4.6 (ten months ago they were everybody's favorite model) In a year from now, they'll be running more intelligence than GPT 5.4 and beyond Local AI is gonna be the default

Ahmad @TheAhmadOsman

2 years ago

LocalLLaMA Home Server Final Boss 😎

106 78 1K 1.5M 303

AutoML automates complex processes like model selection, training, and optimization, making the ability to develop effective machine learning models accessible to a broader audience. #AutoML #MachineLearning #DataScience.

Muse. Dots. Grok Bot. Everyone wants to know which one wins. My view? I don’t think this ends with one winner. All three are attacking the same huge opportunity from slightly different directions. Muse is positioning itself around the personal agent. One persistent AI that learns you, your goals and the apps you use, then works in the background on your behalf. Dots feels more like a persistent generalist inside ChatGPT. Give your dot ongoing responsibility, connect the tools it needs, and let it keep making progress between conversations. Grok Bot is taking a different approach: A team of always-on agents. Different Bots. Different jobs. Different responsibilities. Your own AI team. I can see all three earning meaningful market share because different people will prefer different ways of working with agents. But if one has the potential to creep ahead of the others, I think it could be Grok Bot. Not necessarily because the underlying AI will always be smarter. Because the organisational model makes a lot of sense. Real businesses don’t have one employee doing everything. They have: Sales. Operations. Engineering. Finance. Support. Management. As AI agents become more capable, I suspect a similar structure will emerge. Muse is centred around one personal agent. Dots is centred around one persistent agent owning ongoing work. Grok Bot is centred around building and managing a team of agents. There is plenty of room for all three. But of those three operating models, the Grok Bot approach currently interests me the most.

Random tutorials are a slow way to understand a new AI ecosystem. Awesome Jev Live is an evidence-graded GitHub index of SDKs, MCP tools, agents, apps, open models, and related resources around TypeSafe AI’s Jev System One model. It helps you find a starting point faster by grouping projects into categories and showing how strongly each entry has been verified, rather than treating every search result as equally reliable. Key features: • Evidence grades – distinguish official, observed, inferred, and unverified entries before you follow a lead • Practical categories – scan SDKs, agent tooling, routing and guardrails, evaluation, open reproductions, and applications • Live refreshes – the README says the index rebuilds every two hours as the ecosystem changes • Searchable resource – use the linked site to browse the collection beyond the README • Transparent pipeline – the repo explains its collection, curation, media, rendering, and audit stages It’s open-source (MIT license). Link in the reply 👇

Germany once again choosing to do dumb stuff: The Finance Ministry wants to kill Bitcoin's 1-year tax exemption for all $BTC bought from 2027 onward. Instead: 25% capital gains tax + solidarity surcharge (money for east germany lmao). Forever. From 2028, exchanges should withhold the tax directly and can even sell your coins to cover the tax without your approval. Expected revenue: a measly €350M/year starting in 2029. Imagine damaging Germany as a Bitcoin location and destroying legal certainty for €350M. Incredible. 👏

Frank Schäffler @f_schaeffler

3 days ago

Das BMF hat den Referentenentwurf zur Krypto-Besteuerung an die Verbände verschickt. Das Kabinett soll ihn am 14. Oktober beschließen. Stellungnahmen sind nur bis zum 6. Oktober möglich, also sechs Tage. Was drinsteht:
– Die Haltefrist fällt für alle Bitcoin, die ab dem 1.1.2027

73 171 1K 85K 133

AI agents are no longer just assistants. They can access systems, call APIs, modify data and take actions on our behalf. That makes them something security teams already understand very well: privileged users. So perhaps every AI agent needs the same scrutiny we apply to privileged human accounts: least privilege, short-lived credentials, continuous logging and regular access reviews. More autonomy should mean tighter access control, not broader trust. #AgenticAI #CyberSecurity #IdentitySecurity #AIGovernance #AI

Running several coding agents shouldn’t mean juggling terminals and branches. Cezar is an open-source control center for running Claude Code, Codex, OpenCode, and other coding agents in parallel, locally or on a server. It helps you move multi-step coding tasks forward with less babysitting by running workflows in separate git worktrees, streaming progress live, and letting you review the diff before making a PR. Key features: • Parallel worktrees – each task gets its own Git worktree, while additional tasks wait in a queue • CLI-login based – uses your existing Claude, Codex, OpenCode, or pi login; no API key is required • YAML workflows – combines agent steps and shell checks; a failed check can retry a prior step • Live run visibility – shows agent text, tool calls, tokens, and cost while a task runs • Remote or mobile access – supports VPS setup and responsive task review from a phone It’s open-source (MIT license). Link in the reply 👇