Hey, it’s Andreas.
Welcome back to Human in the Loop — your field guide to the latest in AI agents, emerging workflows, and how to stay ahead of what’s here today and what’s coming next.

This week:
→ OpenAI restructures into a for-profit organization
→ IBM releases a major study on AI-driven productivity across EMEA
→ NVIDIA hosts its flagship GTC event in Washington D.C., unveiling breakthroughs in chips, agents, and quantum computing

Plus: a deep dive into why four billion people - half the world - still lack the basics needed to use AI.

Let’s dive in.

Weekly Field Notes

🧰 Industry Updates
New drops: Tools, frameworks & infra for AI agents

🌀 OpenAI restructures into a for-profit PBC with Microsoft holding 27%
→ Major governance shift: OpenAI moves from capped-profit to full for-profit, cementing Microsoft’s influence as strategic shareholder.

🌀 OpenAI debuts Aardvark, an autonomous security researcher agent
→ Aardvark independently scans, tests, and exploits vulnerabilities to help secure digital systems.

🌀 Salesforce open-sources Enterprise Deep Research
→ A multi-agent framework enabling retrieval, reasoning, and synthesis for large organizations.

🌀 LangChain updates DeepAgents
→ Adds persistent memory, orchestration, and failure resilience to their DeepAgents.

🌀 LangSmith launches No-Code Agent Builder
→ Business users can now visually design and deploy AI agents without writing a single line of code.

🌀 Cursor 2.0 becomes an agentic IDE
→ Multi-agent coding workflows, context memory, and AI pair debugging - Cursor just became a co-developer, not a tool.

🌀 Google adds Vibe Coding to AI Studio
→ Build full apps from natural-language prompts. Instant scaffolding, UI generation, and API wiring inside Gemini Studio.

🌀 Windsurf unveils SWE-1.5, its fastest coding model yet
→ Claims 13× speed improvement over top competitors — optimized for real-time software generation.

🌀 Google unveils Pomelli, an AI marketing assistant for SMBs
→ Targeted at small businesses — Pomelli automates content, campaign setup, and ad optimization directly in Google Workspace.

🌀 Perplexity launches AI-powered Patent Search
→ Enables natural-language exploration of global patent databases — a big boost for R&D and innovation teams.

🌀 Canva launches Creative OS
→ AI-powered, fully editable design workflows for brand consistency and automation across teams.

🎓 Learning & Upskilling
Sharpen your edge - top free courses this week

📘 DeepLearning.AI x Google — Build Real-Time Voice Agents
→ Hands-on course using Google’s Audio Development Kit (ADK). Learn to build voice-interactive agents with live response pipelines.

📘 Hugging Face releases a 200-page guide on building LLMs
→ Comprehensive playbook for building, training, and scaling large language models — from tokenization to optimization.

🌱 Mind Fuel
Strategic reads, enterprise POVs and research

🔹 IBM on AI ROI in EMEA
→ IBM surveyed 3,500+ senior leaders — 66% say AI is already driving real productivity gains.

🔹 NVIDIA GTC 2025 — Jensen Huang keynote at Nvidia’s flagship event
→ If you want to see where AI is heading, watch Jensen Huang’s GTC keynote — NVIDIA’s equivalent of an Apple event for AI. The fall edition in Washington D.C. unveiled breakthroughs across chips, agents, and quantum computing.

🔹 Anthropic explores Claude’s ability to detect injected thoughts
→ Experiments reveal Claude can identify and resist manipulated reasoning paths — an early form of self-reflection in LLMs.

🔹 University of Rome Research shows LLMs are invertible
→ This new study finds it’s possible to recover training data from model outputs — raising deep questions on data privacy and AI security.

🔹 Microsoft 365 Copilot adds Researcher & App Builder
→ Copilot gains two major upgrades: a web-research “Researcher” with Computer Use, and a no-code App Builder for internal tools.

♾️ Thought Loop
What I've been thinking, building, circling this week

AI just hit 1.2 billion users in under 36 months.
Faster than the internet.
Faster than electricity itself.
Faster than the smartphone.

But here’s the real headline: AI adoption is splitting the world in two.

The Global North is adopting AI at roughly 2× the rate of the Global South.
And nearly four billion people - half the planet still lack the basics to even use it: reliable electricity, stable internet, and digital skills. Let that sink in.

So while AI is breaking records for speed, it’s also deepening the global divide. The story of AI’s rise isn’t just one of innovation - it’s also a story of uneven AI diffusion and unequal progress.

What “AI diffusion” actually means

Microsoft’s AI Economy Institute dropped an interesting new study this week on AI diffusion - a term they use to describe how far AI actually spreads and takes root in real life.

In simple terms, AI diffusion measures the real-world adoption of AI across populations, sectors, and economies — not who builds the models, but who uses them. It’s about how individuals, companies, and governments integrate AI into how they live, work, and learn.

To quantify that, Microsoft uses AI User Share — the percentage of a country’s working-age population that actively uses AI tools.

AI’s New World Order

• Leaders by usage: UAE, Singapore, Norway, Ireland — all >40–59% adoption. Policy + skills + broadband beat bragging rights about “frontier labs.” 

• Compute concentration: U.S. + China host ~86% of global data-center capacity — the rails of AI. Great for speed. Bad for everyone else’s latency, cost, and sovereignty. 

• Language as a new barrier: Low-resource language nations lag even after you control for GDP and internet. If your language isn’t well-modeled, your citizens are effectively tax-ed with worse AI.

Why some countries pulled ahead

When you zoom out, the winners aren’t the ones with the biggest models — they’re the ones that built the right foundations. Here’s what they did differently:

• They built the basics decades ago.
  Singapore’s 59 % AI adoption isn’t luck — it’s 40 years of broadband, STEM education, and national digitization.

• They invested in human capital early.
  Digital literacy, not PhDs, drives mass adoption. Everyone can use AI — if they’re taught how.

• They localize fast.
  Nations that train or integrate local-language models remove friction and make AI feel native.

• They coordinate policy and industry.
  The UAE didn’t wait for Silicon Valley — they made AI a national skill, not a spectator sport.

AI diffusion = AI impact

AI is spreading faster than any invention before it, yet half of humanity can’t join the race.

Nearly four billion people — half the world — still lack the basics to even use it: reliable electricity, stable internet, and digital skills.

I was genuinely surprised by this analysis. I always thought AI would finally be the technology that levels the field — giving people in developing countries access, speed, and opportunity.

Because that’s what makes AI powerful: it’s not just automation, it’s acceleration — of knowledge, education, and human potential.

But the reality tells a different story — and there’s still a long way to go.

🔧 Tool Spotlight
A tool I'm testing and watching closely this week

Project Bob is a new AI software development project from IBM that lives inside your IDE and helps you ship quality software, faster.

Why it stands out:
→ Understands your repo, intent, and security standards
→ Catches issues before they reach a pull request
→ Supports chat, code review, and vulnerability detection — all in-editor
→ Specializes in modernization: Java 8 → 17 upgrades, Struts → React migrations
→ Enterprise-ready with compliance alignment (FedRAMP, HIPAA, PCI)

How it works:
Bob runs directly inside your IDE. It reads your repo context, assists in debugging, reviews PRs, and suggests secure configurations in real time. Under the hood, it’s tuned for enterprise DevSecOps — integrating with Ansible, Terraform, and modern CI/CD pipelines to automate code and infra hygiene.

Why it matters:
Bob bridges the gap between developer productivity and enterprise compliance. Instead of context-switching between tools, teams can modernize, refactor, and ship inside one trusted assistant that truly understands their environment.

→ Join the waitlist for Project Bob