Welcome, humans.
So, Sen. Bernie Sanders said he planned to introduce a bill giving Americans a 50% public ownership stake in the largest AI companies, arguing that the wealth created by AI should flow back to the public whose data helped build it.
I mean, if it happens, I’m not going to say NO….
Here’s what happened in AI today:
🙀 Microsoft Build sketched the agent-first computer.
📰 Trump signed a frontier AI security order.
🍪 Cognition launched Devin Desktop for coding agents.
🍪 Perplexity moved agent compute onto your machine.
📖 AI forecasters moved their timelines forward again.
Hey: Want to reach 700,000+ AI-hungry readers? Advertise with us!
Microsoft Build had enough product names to make a developer wonder whether the keynote came with its own package manager.
So here’s the simple version: Microsoft is trying to build the world where AI agents do actual work.
At 1:50, Satya Nadella framed the new AI stack as compute, models, context, tools, runtime, and security. Translation: agents need more than a chat box. They need chips to run on, company data to use, permissions to follow, tools to control, and audit trails so IT can see what happened.
Here’s what happened:
Microsoft announced Windows AI API updates, including Aion 1.0 Instruct and Aion 1.0 Plan, two local Windows models for reasoning, planning, tool use, and on-device agent workflows.
It unveiled the Surface RTX Spark Dev Box, a local AI dev machine with 1 petaflop of AI compute, 128GB of unified memory, and support for running up to 120B parameter models locally.
Microsoft introduced Web IQ, Work IQ, Foundry IQ, and Fabric IQ, which are basically fresh web data, company context, model-building context, and data context for agents.
GitHub showed the new GitHub Copilot app, Microsoft’s answer to Codex and Claude Code, with access to OpenAI, Anthropic, and Google models.
Why this matters: Microsoft’s bet is that the agent race will be won by the company that owns the work environment, not the company with the most impressive demo at any given time.
That means Windows needs to become the local agent machine. GitHub must become the coding-agent control center. Foundry becomes the deployment layer. Microsoft 365 becomes the work memory. And Agent 365, MXC, Defender, Entra, and Purview become the guardrails.
OpenAI is moving in the same direction, now from the other side. Their most important app, Codex, now has more than 5M weekly users, and OpenAI says non-developers are already about 20% of usage and growing more than 3x faster than developers. Its new role-specific plugins package apps, workflows, instructions, and context for sales, analytics, creative production, product design, investing, and banking.
So a sales team can use Codex with tools like Salesforce, HubSpot, Outreach, Clay, and Slack to prioritize accounts, research buyers, draft follow-ups, update records, and review risky deals. Analysts can connect Snowflake, Databricks, Hex, and Tableau. Creative teams can use Figma, Canva, Shutterstock, Picsart, and fal. And Codex also just added Sites, which are shareable interactive websites or apps your team can open with a URL directly inside Codex.
All that might sound like a small leap from where we’ve been until you imagine the normal knowledge-work loop: spreadsheet, deck, Slack thread, meeting, revised deck, and then the ancient ritual of asking, “wait, which version is final?” Codex is trying to turn that hearty pile of tool slop into a single working surface.
Our take: Meanwhile, I couldn’t be more excited about Hermes Desktop from Nous Research, which has taken the baton from OpenClaw and started sprinting towards something like Open-Codex.
It’s an open-source desktop agent for Mac, Windows, and Linux that connects across Telegram, Discord, Slack, WhatsApp, Signal, email, and the command line with one memory. To install Hermes prior to this app, it was scary and confusing. Post-app, it’s easy-peasy. It can learn your projects, auto-generate skills, schedule tasks, browse the web, spin up isolated subagents with their own terminals, and run work inside sandboxes like Docker, SSH, Singularity, Modal, or local environments. Just watch this to check it out.
FROM OUR PARTNERS
Claude is currently the most powerful tool of 2026. It's been launching new features every week- Skills, Connectors, Cowork, vibe coding. Yet almost no one knows how to actually use them.
Our expert mentors have condensed 800+ hours of Claude research, articles, YouTube content and real-world practice into a focused 16-hour curriculum. Join the 2-Day Claude AI Mastery Workshop: a live, end-to-end deep dive into Claude plus 10+ AI tools, LLMs and workflows.
It would be silly not to SIGN UP
The professionals who master Claude right now will design carousels in seconds, ship code without being developers, and have AI working for them while they sleep.
You will learn how to:
master Claude's three modes : Chat, Cowork and Code.
Set up Skills, Connectors and Plug-ins to automate your desktop, Notion and files.
Vibe code apps and dashboards without writing code & 10+ AI tools and workflows that pair with Claude.
🧠 Saturday & Sunday
🕜 10 AM – 7 PM EST
Most “bad prompts” are really untested systems. Anthropic’s Prompting Playbook advice is simple: before you rewrite everything, build a tiny eval suite (a set of test cases that tells you whether the prompt improved).
Start with three test types: one control case the model should always pass, edge cases where it failed before, and capability-boundary cases where it should hand off to a human or refuse. Then fix one failure mode at a time.
The best insight: instructions do not add capability. Telling a model “do the math correctly” does not make it good at math. Give it a calculator tool. And for agentic workflows, split big prompts into a generate → evaluate → repair loop instead of asking one prompt to do everything.
Act as a prompt debugger. Help me improve this prompt without rewriting blindly.
Prompt:
[paste prompt]
Task:
[describe what the AI should do]
Build a tiny eval suite with:
1. One control case that should always pass
2. Three edge cases where the prompt could fail
3. One capability-boundary case where the AI should escalate, ask for help, or refuse
Then diagnose each failure as:
- Prompt issue
- Missing tool or capability
- Harness / workflow issue
Finally, suggest the smallest change to test next.Have a specific skill you want to learn? Request it here.
Devin Desktop gives you one desktop surface for local and cloud coding agents, including Devin, Claude Code, Codex, and custom agents through the open Agent Client Protocol, free to download today.
Perplexity Computer will split agent tasks between a local model on your device and frontier models in the cloud, so private data can stay local while hard reasoning goes remote, no pricing details.
Odysseus from YouTuber PewDiePie gives you a self-hosted AI workspace for running your own agent setup instead of pushing everything through someone else’s cloud, free to try.
Factory launched Factory Router, an enterprise model router that cuts coding-agent token spend by 20-25% while keeping near-frontier performance.
MOSS-TTS is an open-source speech and sound generation model for long-form narration, multi-speaker dialogue, voice design, environmental sound effects, and real-time streaming, free to try.
AtomRetro lets you chat with molecules, sketch structures, generate analogs, and explore interactive retrosynthesis trees for chemistry workflows.
Invideo Agent One turns your script, treatment, references, and shot notes into a full video workflow, so you can direct scenes, casting, camera language, and edits like you’re talking to a film crew —limited free plan; paid plans use monthly credits.
Scott Hanselman joined Corey Noles live from Microsoft Build for a hands-on episode about engineering with AI, where the point was less “vibe coding” and more “what can a thoughtful developer actually build faster now?”
Scott showed practical examples from his own projects, including Baby Smash, Tiny Tool Town, and even worked on a personal blood sugar tracking app built from an open-source tool for his own diabetes management, live. Check it out here.
US President Trump signed an AI security order that creates voluntary pre-release government access for covered frontier models, classified cyber benchmarks, and new federal support for AI vulnerability detection without creating a mandatory model-licensing regime.
Black Forest Labs named Martin Scorsese as an advisor after a FLUX storyboarding session, framing visual AI as a pre-production tool for filmmakers.
Anthropic expanded Project Glasswing to roughly 150 organizations in 15+ countries to help critical-infrastructure teams find and patch software vulnerabilities.
SK Hynix plans to double memory-chip wafer capacity over five years as AI demand keeps squeezing the chip supply chain.
Adobe Research introduced SuperFit, a CVPR 2026 system that compresses 3D shapes into editable building blocks.
Stability AI researchers released Stable-Layers, a reinforcement-learning method for splitting images into editable layers using vision-model feedback.
FROM OUR PARTNERS
Want a computer you can actually train your own AI on? Dell Pro Max with GB10 brings NVIDIA Grace Blackwell architecture, 128GB memory, and NVIDIA DGX OS 7 to founders, builders, and AI enthusiasts testing agents, demos, and model workflows locally.
Demis Hassabis told Stanford we may be in “the foothills of the singularity,” with AGI possibly around 2030, but his more useful point was that today’s AI tools already have a capability overhang: they can do far more when paired with the right workflows, tools, and domain expertise, so students should lean in, preserve their agency, and use the tools to make more of what is in their heads.
Yoko Li argues the next frontier of visual AI is code, because SVG, HTML, React, Lottie, and Blender scripts are editable, versionable, and easier to turn into real products than flat pixels.
Ethan Mollick highlighted a new study showing AI coding tools produce far more code, but shipping only rises modestly because review, integration, and judgment become the bottleneck.
Rohin Shah shared what running AGI safety at Google DeepMind actually looks like, including why he sees safety work as practical engineering more than abstract doom debate.
Kate Deyneka resurfaced a great thread of ML visualizations, from Transformer Explainer to 3D LLM inference, because abstract AI concepts get much easier when you can see them move.
NVIDIA researchers argued video generation is now an infrastructure problem and released LongLive 2.0, a system for longer, more consistent video output.
Forecasting Research Institute said experts and superforecasters moved their AI timelines forward, with median AGI forecasts around 2050 for experts and 2047 for superforecasters.
 | That’s all for now.
|
Grant Harvey is the Lead Writer of The Neuron, where he continues to lead the publication's daily coverage of AI news, tools, and trends.
Don't fall behind on AI. Get the AI trends & tools you need to know. Join 700,000+ professionals from top companies like Microsoft, Apple, Salesforce and more.
Property of TechnologyAdvice. © 2026 TechnologyAdvice. All Rights Reserved
Advertiser Disclosure: Some of the products that appear on this site are from companies from which TechnologyAdvice receives compensation. This compensation may impact how and where products appear on this site including, for example, the order in which they appear. TechnologyAdvice does not include all companies or all types of products available in the marketplace.