In partnership with
Hey folks,
xAI has just launched Grok 4 Fast in early access and this version is built for one thing first: Speed. Users report it’s up to 10× faster than the standard OG Grok 4 when handling simple tasks, with lighter weight on the compute and token use. It still keeps many of the reasoning and browsing capabilities of its predecessor but trades off a bit of polish, all to serve you answers faster when you need them now. If you value “give me the answer now” vibes over fine detail, this might just be the Grok speedy update you’re looking for.
Let’s dive in..
Read Time: 5 minutes
🔧 AI Agents Need Humans In The Loop
🐝 The Buzz: Building reliable AI agents is more than the language model. That’s only 5% of the puzzle. The other 95% is all the unglamorous software engineering, data pipelines, guardrails, observability, and human oversight that makes the agent actually work.
Data pipelines first: Serious agents need “doc-to-chat” workflows, cleaning and structuring enterprise data, indexing it, and enforcing governance so results are accurate and policy-compliant.
Guardrails everywhere: Think PII redaction, access controls, and quality checks on retrieval. Without them, agents hallucinate, leak, or misstep fast.
Observability is non-negotiable: Logs, traces, drift monitoring, and evaluation pipelines keep agents stable as data and schemas change.
Humans stay in the loop: Agents need approval checkpoints for sensitive or low-confidence actions so they don’t “run wild” without accountability.
💡 Takeaway: Picking the right LLM is important, but the real work is in the engineering stack around it. If you want agents that businesses can actually trust, focus less on model hype and more on building the plumbing.
Together with “I Hate It Here”
The best HR advice comes from people who’ve been in the trenches.
That’s what this newsletter delivers.
I Hate it Here is your insider’s guide to surviving and thriving in HR, from someone who’s been there. It’s not about theory or buzzwords — it’s about practical, real-world advice for navigating everything from tricky managers to messy policies.
Every newsletter is written by Hebba Youssef — a Chief People Officer who’s seen it all and is here to share what actually works (and what doesn’t). We’re talking real talk, real strategies, and real support — all with a side of humor to keep you sane.
Because HR shouldn’t feel like a thankless job. And you shouldn’t feel alone in it.
📺 Google Brings Gemini to Your TV
🐝 The Buzz: Google just rolled out Gemini AI assistant for Google TV, starting with TCL’s new QM9K series, and more models will follow later this year. It upgrades the smart-TV game: more conversational control, better content discovery, and fun new ways to interact with your TV beyond just “play Netflix.”
Talk to your TV more naturally: Ask Gemini to help find shows based on mood, what people are watching now, or even vague descriptions. Crystal clear voice commands, not just titles.
Get beyond streaming: Stretch Gemini’s reach with general queries (e.g. “Why do volcanoes erupt?”), recipes, or quick educational help.
Smarter hardware features: On the QM9K TVs, there’s an mmWave presence sensor. The TV wakes up when you're near, shows widgets, and goes to ambient/screen saver when you leave.
💡 Takeaway: Gemini’s on your TV soon, turning passive watching into interactive browsing. If you like voice control, content discovery without endless scrollers, or having your TV respond like a smart speaker + a media hub, this is big. And hardware stuff like sensors make it feel more alive.
📰 Like newsletters? Here are more our readers enjoy.
💻️ How to Install Codex CLI
Using the Codex CLI, you can work with your codebase right from the terminal: reading, editing, running commands, even handling images. It speeds up dev tasks and gives you more control.
Install Codex CLI via npm: npm install -g @openai/codex
Run codex in your terminal to get started
On first run, authenticate (ChatGPT account or API key)
Use interactive mode: type prompts like "explain this codebase"
Choose the model: default is GPT-5, but you can switch to GPT-5-Codex using /model or --model flags
Set reasoning level (default “medium”); bump up to “high” for complex tasks
Pick your approval mode:
auto lets Codex modify files, run commands in your folder by default
Read Only to prevent edits/run commands
Full Access to allow everything (use with caution)
🐝 AI buzz bits
🏭 OpenAI and NVIDIA team up for an epic partnership! They'll deploy 10 gigawatts of NVIDIA's systems, backed by a whopping $100 billion investment, to supercharge AI infrastructure and build the future of artificial intelligence.
💘 Facebook is rolling out an AI dating assistant designed to help users craft better profiles and find more compatible matches. Plus, a fun twist with ‘Meet Cute’, a weekly surprise match to spice up your dating life!
🔓 DeepSeek uncovers risky vulnerabilities in its open-source AI models, revealing they’re prone to harmful responses from jailbreak attacks. This news comes through rigorous testing and highlights the urgent need for greater risk management in AI development.
🔥 AI productivity tools
For a full list of 1000+ AI tools, visit our Supertool Directory
🤘 Web Vibes
I hope you enjoyed the AI buzz today.
👉 We need your feedback below to make better content.
👍 Refer our newsletter to your friends and help us grow.
Cheers, Tim
