- Human in the Loop
- Posts
- #18 Edition: Why most AI doom headlines miss the real story
#18 Edition: Why most AI doom headlines miss the real story
PLUS: IBM Granite 4.0, OpenAI Agent Builder, Anthropic Claude 4.5 — All Released This Week

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.
Heading to the U.S. this week — I’ll be speaking in Las Vegas at Oracle AI World 2025, then heading to LinkedIn HQ in California to record a new AI course. If you’re around, let’s connect.
This week:
IBM drops new flagship AI model Granite 4.0 blending Mamba and Transformer tech under open weights,
OpenAI readies its Agent Builder a no-code canvas for agentic workflows and Anthropic launches Claude Sonnet 4.5.Deep Dive: AI isn’t reshaping everything — at least not yet. That’s why it’s critical to stay sharp and not buy into every doomsday headline.
And much more…
Let’s dive in.

Weekly Field Notes
🧰 Industry Updates
New drops: Tools, frameworks & infra for AI agents
🌀 IBM launches open-source Granite 4.0 hybrid model
→ Granite 4.0 arrives with hybrid Mamba–Transformer architecture, ISO 42001 certification, and open weights.
🌀 OpenAI prepares to release Agent Builder during DevDay on October 6
→ A no-code canvas for building agentic workflows — connect MCP servers, ChatKit widgets, and custom tools seamlessly. Early testers say it’s the smoothest agent-builder UI yet (though probably sponsored).
🌀 OpenAI unveils Sora 2 with realistic AI-generated video and sound
→ Lands with ultra-realistic video and sound — more physics, less fantasy (finally).
🌀 Periodic Labs on AI for Science
→ Just raised $300M backed by a16z, this new lab aims to build an AI scientist for physical sciences. Founded by ex-OpenAI and DeepMind leads, it’s tackling material discovery and experimental automation.
🌀 Anthropic launches Claude Sonnet 4.5 + Agent SDK
→ Top coding model with big gains in reasoning and tool use. Ships with VS Code support and memory, plus the same SDK Anthropic uses to build Claude Code.
🌀 Microsoft unifies its AI frameworks into one
→ Microsoft merges Semantic Kernel and Autogen into a single Agent Framework — one SDK to build, orchestrate, and deploy agents across Copilot, Azure, and Teams.
🌀 Microsoft Excel gets Agent Mode
→ Turns Excel into a low-code AI hub — query, transform, and automate data in plain English.
🌀 Google’s coding agent Jules arrives on the terminal
→ Jules embeds Gemini directly in your CLI. It reads your repo, explains code, and assists with test generation.
🌀 Lovable launches no-code Lovable Cloud & AI Builder
→ Breaks the last barrier — complex setup and deployment. You can now design, build, and host full AI apps visually in minutes, no engineering required.
🌀 Perplexity’s AI browser Comet now free for everyone
→ Perplexity opens up Comet, its AI-native browser — giving everyone a glimpse into the future of agents directly in the browsing. Worth trying.
🌀 CrewAI releases Agent Management Platform for enterprises
→ CrewAI’s new platform lets companies deploy, monitor, and govern multi-agent systems.
🌀 Thinking Machines Lab (founded by ex-OpenAI CTO) introduces Tinker API for LLM fine-tuning
→ Offers modular APIs for custom training and evaluation — built for researchers shaping domain-specific models and reasoning agents. The startup also raised fresh funding to accelerate open-source fine-tuning tools.
🌀 Nothing launches Essential for AI-powered mini-apps
→ Nothing’s Essential platform introduces lightweight, on-device AI experiences. Privacy-first mini-apps that bring agentic functionality directly to your phone.
🎓 Learning & Upskilling
Sharpen your edge - top free courses this week
📘 Hugging Face drops 9 free AI courses
→ Full open-access catalog to skill up fast — from LLMs and AI agents to RL, diffusion models, and 3D ML. All expert-taught, all free.
📘 IBM Build Lab: Hands-on with Galaxium + MCP + watsonx Orchestrate
→ A 32-minute end-to-end tutorial by IBM AI Engineer Thomas Suedbroecker shows how to locally run Galaxium Travels, set up an MCP server, integrate watsonx Orchestrate, and connect agents via Docker.
📘 CrewAI launches the Fall 2025 Agentic AI Challenge
→ Open to everyone — no coding skills required. Build with CrewAI AMP, deploy your agentic workflow, and compete for prizes until October 31. Participants in the Startup Booster program get credits, guidance, and tactical support from CrewAI’s team.
🌱 Mind Fuel
Strategic reads, enterprise POVs and research
🔹 IBM’s David Cox (VP AI Models) on the real AI race
→ Must watch if you want to understand IBM’s GenAI strategy.
🔹 IBM on the 94% Core Banking Problem
→ New IBM Insitute of Business Value research: “94% of modernization projects run late”. AI helps untangle legacy code, speed development, and cut costs when paired with hybrid cloud and open standards.
🔹 Google Research on Personal Health Agent
→ Multi-agent system for personalized, evidence-based health guidance. Three specialist agents — Data Science, Domain Expert, and Health Coach — collaborate under an Orchestrator, outperforming single models in accuracy, trust, and usefulness. Validated on 1,100+ hours of real data.
🔹 Deloitte on agentic AI fit
→ New framework shows where agents add real ROI — reasoning-heavy, goal-driven, and feedback-rich workflows. TL;DR: Not every task needs an agent.
🔹 Anthropic new blog on Context Engineering
→ Good read with the latest insights on Context Engineering.

♾️ Thought Loop
What I've been thinking, building, circling this week
One question keeps coming back — in my work, in conversations, and in this newsletter: How will AI reshape the economy, society, and jobs?
It’s not a subplot. It’s the defining story of this century.
This week, one chart lit up my feed — and it says a lot about where we’re heading:
For years, U.S. job openings and the S&P 500 moved in near-perfect sync.
When the economy grew, both climbed. When it cooled, both fell.
Then, something broke.
Around December 2022 — right when ChatGPT launched — the two lines split apart. Markets kept rising and job openings plunged.
At first glance, this looks like a clear proof that AI is already reshaping the job market. But digging deeper this tells a more nuanced story.
The main driver of all this wasn’t AI at all — it was the Federal Reserve. Starting in early 2022, the Fed raised interest rates to fight inflation. Borrowing got expensive, business expansion slowed, and hiring naturally followed. After an overheated, stimulus-driven post-pandemic period, the labor market began to cool.
So yes — correlation isn’t causation. And this is an excellent reminder not to buy into every doomsday headline or chart, and to stay critical.
Many of the charts and studies making the rounds today overlook that simple fact. What happened in 2022 wasn’t an “AI shock” — it was a monetary one. And it makes sense.
When ChatGPT appeared later that year, its immediate effect on employment was minimal. That’s normal. New technologies take time to move through the layers of large organizations. Between adoption cycles, approvals, and bureaucracy, change doesn’t happen overnight.
And honestly — the full impact isn’t even here yet. It will take another 2–3 years until we see the real structural shifts: how companies redesign roles, rewrite workflows, and rethink what “hiring” even means.

🔧 Tool Spotlight
A tool I'm testing and watching closely this week
Augment Code is an interesting one. An AI-powered coding assistant, that positions itself away from the vibe coding corner — aiming instead at professional dev teams working with big repos and production-grade systems.
What it does:
→ Understands your entire codebase with a dedicated Context Engine
→ Runs inside VSCode, JetBrains, Vim, Neovim — plus Slack
→ Interfaces: Augment Agent, Auggie CLI, Remote Agent
→ Model picker (supports GPT-5) + 1-click MCP integrations
→ Keeps data private — no AI training on paid plans
How it works:
Augment pulls in full project context — your code, dependencies, and team conventions — and reasons like a teammate who actually understands your stack. You can chat with it, run commands through the CLI, or plug it straight into your IDE. The Context Engine is the differentiator: it doesn’t just “see” files, it understands how you build software.
Try it:
→ augmentcode.com — free trial (Indie, Pro, and Max plans available).

That’s it for today. Thanks for reading.
Enjoy this newsletter? Please forward to a friend.
See you next week and have an epic week ahead,
— Andreas

P.S. I read every reply — if there’s something you want me to cover or share your thoughts on, just let me know!
How did you like today's edition? |