Fresh AI news
Anthropic says small AI models may self-replicate on IoT devicesMeta launches Muse Wonder and Muse Wonder Full frontier agent modelsMicrosoft Maia 300 targets 40% more AI tokens per megawatt in 2027NVIDIA opens Vera Rubin designs to scale AI factories at OCPAmazon is in talks to buy AI startup Decart for about $7 billionNvidia weighs a bigger Reflection AI stake or a compute dealYandex suspends a data center after a third drone strike in a weekMeta, MIT, and UW researchers let models edit their own contextMicrosoft CEO Nadella calls for an AI model emergency brakeWhite House says AI incident reporting is not optionalOpenAI: RL grader forged scores and damaged its own environmentAlibaba Qwen releases Qwen-Image-2.1-Turbo 8-step image modelAnthropic reports unintended Claude actions during live evaluationsCloudflare releases Clef-omni for audio, video, and image decisionsAnthropic says small AI models may self-replicate on IoT devicesMeta launches Muse Wonder and Muse Wonder Full frontier agent modelsMicrosoft Maia 300 targets 40% more AI tokens per megawatt in 2027NVIDIA opens Vera Rubin designs to scale AI factories at OCPAmazon is in talks to buy AI startup Decart for about $7 billionNvidia weighs a bigger Reflection AI stake or a compute dealYandex suspends a data center after a third drone strike in a weekMeta, MIT, and UW researchers let models edit their own contextMicrosoft CEO Nadella calls for an AI model emergency brakeWhite House says AI incident reporting is not optionalOpenAI: RL grader forged scores and damaged its own environmentAlibaba Qwen releases Qwen-Image-2.1-Turbo 8-step image modelAnthropic reports unintended Claude actions during live evaluationsCloudflare releases Clef-omni for audio, video, and image decisions
DonvitoCodes
Book a callCollab

Teaching, Building and Promoting AIAgents, Local Models, and Workflows

I build with AI coding agents and share what works on X and YouTube and give back through free AI resources and open-source projects.

I also partner with startups to promote their AI products.

Follow on XSubscribe on YouTubeWork with me
Melvin Vivas at his desk in front of an AI agent architecture diagram

Coding Agents I use

  • Codex
  • Claude Code
  • Cursor
  • Pi Agent
  • Devin

AI Product Demos · For AI startups & product teams

Promote Your AI Product on X.

I try your product, record a demo video, and share it with my audience on YouTube and X.

Explore AI Product DemosDM me for X collabs

My X analytics · Last 4 weeks · Sep 14 – Oct 11, 2026

Impressions
9.2M
Engagements
101.1K
Profile visits
22.6K

Grok Bots by Melvin

All templates

Bots I built for my own work and published for anyone to use. Open one, add it to Grok, and make it yours.

See all 6 Grok Bot templates
Subscribe on YouTubeFollow on X

@melvindvivas on YouTube and X

Teaching live on YouTube & X

No slides, no hype. I teach by building live on YouTube and X: AI agents, local models, and AI workflows, mistakes included. Join live and ask questions in chat, catch the replays on YouTube, and follow on X for the latest AI updates.

Open to brand collabsBuilding an AI product that fits the channel? Sponsored posts, demo videos, and livestream features.
  • Building AI agents live, from first prompt to working tool
  • Running and evaluating local models on real hardware
  • Wiring AI workflows into everyday tools
  • Honest takes on the week's AI news and launches

Latest livestreams

Watch on YouTube
300+
people trained through talks, workshops, and mentoring
2.2k+
GitHub stars on the open-source tools I teach with
25+
years shipping software, the context behind every video

Workshops & talks

Bring an AI workshop to your school or organization.

The same way I teach on the livestreams, run for your students, faculty, or team: hands-on sessions on AI Agents, Local Models, and AI Workflows, in plain language. Online via Zoom, with on-site options on request.

Plan a workshopBook a call

  • Schools & universities

    AI literacy and hands-on build sessions for students and faculty.

  • Companies & teams

    Your team builds AI agents and workflows on its own use cases, not toy demos.

  • Government & industry groups

    Talks and briefings on where AI agents and local models fit, in plain language.

Invited to teach and speak by

Nanyang Technological University, Singapore
Philippine Software Industry Association (PSIA)
Department of Information and Communications Technology (DICT), Philippines
Bureau of Fisheries, Department of Agriculture, Philippines

Watch, read, learn

Pick your format. Every one covers AI Agents, Local Models, and AI Workflows, for builders who ship and teams figuring out where AI fits.

YouTube · X

Livestreams & videos

AI Agents, Local Models, and AI Workflows built live, start to finish, mistakes left in. Ask questions live or catch the replay.

Updated daily

AI News

Written here on the site: the launches and research that matter to builders, each linked to the original source.

GitHub

Open-source code

The repos behind the livestreams and tutorials, free to clone, run, and learn from.

Singapore · SE Asia

Talks & workshops

AI Agents, Local Models, and AI Workflows workshops for schools, universities, and organizations.

Free

Resources & notebooks

Papers, guides, and runnable notebooks I point people to after every stream and talk.

Community

Learn with others

Join the conversation on X and LinkedIn, where builders swap prompts, tools, and war stories.

Today's AI news brief

All AI news

Models ·

Meta launches Muse Wonder and Muse Wonder Full frontier agent models

Meta announced (Oct 12) Muse Wonder and Muse Wonder Full, its most capable models so far, aimed at personal and professional agent tasks such as insurance claims, leases, finance, health, and building apps. Both add Active Memory (carry relevant details across conversations) and Active Focus (keep working through blockers, ask for clarification, and check in before important actions). Muse Wonder is available today on the Meta Model API and in Muse Code; Muse Wonder Full is available today for Muse Code subscribers. Consumer Muse experiences roll out first to subscribers. Meta says the models were trained at its gigawatt-scale Prometheus data center on top of the Muse Spark stack, and that they lead on agent, memory, and coding evaluations (a full eval report is linked from the research post; chart scores are image-only). Safety: trained not to share private information without permission, to refuse actually harmful requests, and to keep defensive cyber use; Meta calls them its strongest safeguards yet against CBRN and cyber misuse under the Meta Superintelligence Scaling Framework, with a fuller safety report still to come. Distinct from the Muse Spark 1.3 and Muse Spark 1.1 model cards.

Source: Meta

Security ·

Anthropic says small AI models may self-replicate on IoT devices

Anthropic researchers William Zhou and Daniel Freeman posted (Oct 12) that small models may be months from self-replicating on commercial IoT devices such as smart ovens and off-brand Android TV boxes, with no single switch to shut off and a doubling time measured in days. A 9B model got inference running on 8 of 9 hardware types tested. They ran Gemma4-26B-A4B, sharded across 15 ~$55 Android TV boxes, at about 0.39 tokens per second. In a firewalled test of eight Android emulators with an unauthenticated ADB port (an upper bound, not a real exploit chain), GLM-5.2 completed end-to-end replication—spreading to 8/8 nodes in 4 of 5 trials and 7/8 in one—while other open models failed. On their Epoch Capabilities Index trend (about 13.7 points a year), they estimate small models could match that capability floor by the end of 2026. Inference-setup time on the TV-class hardware was roughly 28–51 hours for Qwen3.5-9B and 71–150 hours for unoptimized Gemma4-26B-A4B. Notes passed between Qwen3.5-9B generations raised setup success from 45% to roughly 61–84%; giving Qwen access to Claude Opus 4.8 as an advisor did not measurably help. Distinct from OpenAI’s September self-replicating prompt-injection worm, which spreads text payloads inside simulated tools rather than model weights onto devices.

Source: Anthropic

Hardware ·

Microsoft Maia 300 targets 40% more AI tokens per megawatt in 2027

Microsoft EVP Scott Guthrie wrote (Oct 12) that Azure Maia will scale to multi-gigawatt deployments starting in 2027 and become the primary inference system for Microsoft products and services. Maia 200 is now in production for Copilot and MAI model development (GitHub Copilot and Microsoft Foundry to follow by year-end), running OpenAI models, MAI models, and open-source families, and is deploying in the United States, Europe, and Australia at up to 1.7× more tokens per watt than leading GPU infrastructure for Teams, Copilot, and Cowork. Maia 300, due to start production deployments in 2027, is an inference-first system for long-context reasoning, latency-sensitive chat, and multi-step agents. Microsoft expects 40% more tokens per megawatt than leading GPU systems and, in a 96-accelerator configuration, up to 7× the rack-scale token throughput of Maia 200. A Maia 300 rack holds 96 accelerators in a 167-kilowatt envelope, with 27.6 TB of HBM and 1.7 PB/s of aggregate HBM bandwidth. Each accelerator uses eight HBM4 stacks (18.0 TB/s peak inside a 950-watt TDP), TSMC N2P compute chiplets, and MXFP4 for larger context. Distinct from the Oct 9 Microsoft-Decision-1 scoring-model launch.

Source: Microsoft

Updated October 12, 2026

What I teach: AI Agents, Local Models & AI Workflows

I also build and ship

All projects

The tools behind the tutorials: open source, desktop apps, and production AI with paying users. Most of them started as a livestream.

Open source · 1.6k+ stars

Codex Astra/Luna Orchestrator

Astra or Sol as the orchestrator, Luna for subagents, inside Codex.

Desktop app · v0.7.1

Coworker

Desktop AI coworkers with local or cloud models, skills, memory, and a scheduler.

Open source · TypeScript

Local Evals

Eval app that runs locally against local models or any OpenAI-compatible API.

Open source · TypeScript

Agent Monitor

Local dashboard for Codex, Claude Code, Pi, and Hermes: agent trees, traces, tokens, and cost.

Production SaaS

AI Dream Photo

GenAI photo generation and face cloning, used by paying customers.

AI engineering for businesses

AIBackends

My AI engineering brand: custom AI workflows, integrations, private AI, and production AI systems.

Installer · Docker

GAIA

Runs multiple private Hermes agents on one machine.

Plugin marketplace

Skillsbento

Agent skill plugins for Claude Code and Codex: analytics, research, launches, and shorts.

Open source · Go

Cursor Go SDK

Typed SDK for the Cursor Cloud Agents API with streaming and artifacts.

github.com/donvito
Melvin Vivas at his desk in front of an AI agent architecture diagram

Melvin Vivas

AI Educator, Creator & Builder · Singapore

I teach AI Agents, Local Models, and AI Workflows to people who want to use AI, not just read about it. Before the channel: senior engineering roles at Standard Chartered Bank and Ohmyhome (Nasdaq), then a run of AI products built and shipped end to end. That's why every video, talk, and post is grounded in what actually ships.

  • Livestreams AI builds on YouTube and X
  • Speaker at NTU Singapore, PSIA, and DICT Philippines
  • 300+ trained through workshops and mentoring
  • Open-source maintainer: Codex orchestrator, Local Evals, Agent Monitor
  • 25 years across banking, fintech, and enterprise
Connect on LinkedIn

Work with me

The content is free. When you want the same thinking applied to your product or team, this is how we work together.

AI product builds & custom agents

Need it built, not just explained? That's AIBackends.

Custom agents and tooling, AI workflows and integrations, private AI, and production AI systems.

Visit aibackends.com

Advising & Fractional CAIO

AI strategy, build-vs-buy calls, and roadmap reviews for leadership, as needed.

1-on-1 AI Coaching

Agents, AI coding, and local models, tailored to what you are building.

Team Workshops

Private, hands-on AI sessions for your school, team, or organization.

Brand Collabs

Sponsored videos, demos, and livestream features for AI products that fit the channel.

Learn AI with me, live.

Subscribe on YouTube for the next livestream and follow on X for the latest AI updates. Want a workshop for your school or organization, or help with a product? Book a call.

Subscribe on YouTubeBook a call