Apr 23 • 🫖 AI News - The Tea On AI
What's tea in AI? 🫖You're Weekly AI News Update 4.23
This time we celebrate my new AI podcast.
I wresteled with doing this because I had a podcast ( several podcasts ) before and it's a struggle to be consistent.
But when you keep getting confirmations, you just need to challenge yourself and Go!
The news this week is brought to you by High Tech Tea.
📖 Episode Summary
Episode 1 covers everything that dropped in AI over the past 7–10 days. Two stories out of China, one from OpenAI, and one spicy, creative-world debate. The through-line: AI is getting sharper, not easier, and sharp tools require sharp operators.
Stories covered this episode:
- Kimi K2.6 by Moonshot AI — the open-source trillion-parameter model you need to know
- OpenAI Codex + Chronicle — screen-aware AI that watches, learns, and shows up ready
- Claude as a Design Tool — Figma killer or prettier slop? Jeneba's hot take
Hot Topics:
- Google Personal Intelligence — they already have your data. Now they want to make it useful.
- Canva AI 2.0 — full prompt-based overhaul, and it did not ask for permission
- Claude Opus 4.7 — Anthropic's sharpest model yet (but it'll drain your limits fast)
🍵 THE DROP: New Tools That Slapped This Week (Or Are Coming Soon)
🔧 Tool What It Does Kimi K2.6 — Moonshot AI Open-source, 1 trillion-parameter multimodal model. 300 sub-agents running 4,000 parallel workflows + 13 hours of autonomous coding.
#1 on Humanity's Last Exam.
- OpenAI Codex + Chronicle Screen-aware coding agent inside ChatGPT. Reads your screen in real time, learns your stack, and shows up already warmed up — no re-explaining required.
- Claude Design — Anthropic Generates UI, design systems, decks, and PowerPoints at the intent layer spacing, hierarchy, accessibility, IA. Not a Figma killer (yet), but a serious first-draft engine.
- Google Personal Intelligence Gemini + AI Mode in Search now plug into Gmail, Photos, and Google apps for hyper-personalized responses. Free in the US, personal accounts only ,Workspace is next.
- Canva AI 2.0 Full prompt-based platform overhaul. Persistent brand memory, object-based editing, HTML imports in Canva Code, and Slack/Gmail/Drive/Calendar connectors. Research preview. Claude Opus 4.7 Anthropic's most precise model yet. Self-verification, long-horizon coherence, 3× better vision, new "Extra High" effort tier. Stops fixing weak prompts for you.
🍵 AI AT WORK: Workplace Intelligence — What This Means for Your Work
Google Just Brought AI Into the Cubicle
What dropped: At Google Cloud Next this week, Google rolled out a wave of AI updates to Workspace, its productivity suite for professionals — designed to cut busy work across Gmail, Calendar, Chat, Drive, Docs, Sheets, and Slides.
The headline features:
- Workspace Intelligence — a new AI system embedded across Workspace that pulls from your Gmail, Calendar, Chat, and Drive to assist with tasks. Admins (and users) control which data sources it can access, and you can revoke access anytime. The tradeoff: more access = more useful assistance.
- Build + fill Google Sheets with Gemini — prompt Gemini to construct full sheets (formatting, data retrieval, the works). "Fill with Gemini" auto-populates rows up to 9× faster than manual entry, and a new feature converts unstructured text into clean, organized tables.
- AI writing in Google Docs — Gemini can now "generate, write, and refine" documents using context from your Drive, Chat, Gmail, and the web. Prompt it to "help me write" or "match my writing style" and it mimics your voice.
Why this matters for your work:
- The enterprise AI race just got real. Google is leveraging its biggest moat — Workspace is already inside most companies — to push AI into daily workflows before Microsoft, Apple, and the startups can catch up. The office tool you already pay for is becoming the AI tool you didn't have to onboard.
- "Less busy work" is the actual product. Drafting emails, building spreadsheets, formatting docs, and cleaning data — these are the tasks getting automated first. If your role lives in any of these, your job description is shifting this quarter, not next year.
- Admin controls are the story underneath the story. Google ships this with admin governance baked in — a direct response to the enterprise data conversations that have been holding Workspace AI back. Translation: IT and security teams now have a real seat at the table. If your org hasn't set its Workspace Intelligence policy yet, that's the next meeting.
- Sheets just became a high-leverage skill again. "9× faster" data entry + prompt-built spreadsheets means the people who know how to structure a request will out-produce the people who know how to write a VLOOKUP. Prompt craft is the new spreadsheet skill.
"Mark my words again: Google is positioned to win the enterprise AI race. Workspace Intelligence is the proof. The data was already theirs — now the workflow is too." — Jeneba
🧪 Test It: Open one Google Sheet you maintain manually. Try "Fill with Gemini" on the next batch of rows. Time it against doing it yourself.
🔁 Try It: Draft your next client update in Google Docs using "help me write — match my style." Compare it to your usual first draft.
💳 Buy It? ✅ For Workspace orgs: yes, but with the governance conversation up front. Set Workspace Intelligence permissions before turning anyone loose.
☕ STORY 1 — Kimi K2.6 by Moonshot AI
📰 What Dropped
Moonshot AI — a Chinese AI lab — released Kimi K2.6, an open-source model with:
- 1 trillion parameters (an obscene amount — in the best way)
- Multimodal — processes text, images, video, and more
- 300 sub-agents running 4,000 workflows in parallel
- 13 consecutive hours of autonomous coding — no check-ins, no interruptions
- #1 on the Humanity's Last Exam benchmark — the Harvard/Yale of AI testing
- Claw Groups — a shared workspace where humans and AI agents collaborate side by side, like Google Workspace, but your coworkers are agents
"If you know, you know. K2 means something different where I'm from — but I'll let them have it." — Jeneba
🧪 What It Actually Means
13 hours of autonomous coding = you hand off a task and your AI assistant builds a full app, end to end, without checking in once. That's delegated, autonomous work — not autocomplete, not a suggestion engine.
Claw Groups = think of a live Google Doc where your AI agents are active collaborators alongside you, from any device.
Who this is for: Solo founders, solopreneurs, freelancers, and anyone building without a technical co-founder. This changes the math on what you can ship.
"Jeneba's note: I haven't tried Kimi yet — but I will be by the end of this week."
🛠️ Try It
- Technical? Pull it down and start experimenting
- Non-technical? Watch the space. Apps will be built on top of this fast.
- Start by reading the Kimi K2.6 blog post and learning what agentic AI means — that word is going to be everywhere
- YouTube University: search AI development and building — Jeneba will drop her favorite AI YouTubers + podcasts in the comments
💳 Buy It?
✅ Yes — full buy. The open-source angle means the ecosystem grows fast. Expect:
- Workflow automation wrappers
- A Notion integration (Jeneba's prediction: soon)
- A Canva plugin (also calling it: soon)
Buy the attention. This story has legs.
☕ STORY 2 — OpenAI Codex + Chronicle
📰 What Dropped
OpenAI Codex — the coding agent living inside ChatGPT — got a major upgrade with something called Chronicle.
One sentence: Chronicle lets Codex see exactly what's on your screen.
What it does:
- Reads your screen in real-time — error messages, open docs, half-built projects
- Fills in context automatically, no re-explaining required
- Learns your tools, your tech stack, and how you work
- Builds a mental model of you over time
"Google has had this tech for a while — which is exactly why I've been saying Google is going to come out as the #1 AI leader in the marketplace. Mark my words. Quote me." — Jeneba
🧪 What It Actually Means
The friction keeping non-coders away from AI coding tools isn't the code — it's context transfer. Explaining your whole project from scratch every time. Copy-pasting error messages. Describing your screen. Chronicle eliminates all of that.
This is also a preview of where all AI tools are headed: ambient awareness, where the AI understands your context without you narrating it. Writing tools. Project management. Email. It's all moving here.
"Jeneba's workflow now: copy errors into Gemini because Google knows everything and troubleshoots in real time. Chronicle is that — but built right into your coding agent."
🛠️ Try It
- Chronicle is not widely available yet — but if you have ChatGPT Plus or Pro, watch for Codex rollout updates in the coming weeks
- In the meantime: use Codex as it exists today
- Practice describing your problems less — let AI ask the clarifying questions. Train yourself to delegate, not just dictate.
💳 Buy It?
✅ Full buy. This is a foundational shift in how AI tools work. The AI that sees your context, remembers your patterns, and shows up already warmed up — that's the assistant you haven't hired yet.
If you're building a workflow around AI tools right now, Chronicle is your near future. Pay close attention.
☕ STORY 3 — Claude as a Design Tool
📰 What Dropped
The spicy question making rounds: Is Claude a Figma killer?
Claude Design (by Anthropic) can generate UI designs, design systems, visual components, decks, and PowerPoints — operating at the intent layer of design: spacing, hierarchy, accessibility, and information architecture.
"I used to be on a design system team. I already built a design system for everything I create in AI — websites, apps, landing pages, and email templates. It's a cheat code so you never have to keep re-explaining your colors and fonts." — Jeneba
🧪 Jeneba's Hot Take
No — Claude is NOT killing Figma. Not yet. Not this year.
Here's why:
- Seeing people in the comments saying Claude gave them the exact same design as everyone else
- "You can't put lipstick on a pig." If you don't know how to prompt contextually, you'll get basic output, regardless of how powerful the tool is
- We're going to see prettier slop for a while with Claude Design
- Figma already killed Adobe InDesign and Illustrator — the enterprise will be slow to replace it
- At the enterprise level, Figma is still the standard. It's staying.
What is shifting: the entry point to design. First drafts, wireframes, explorations — AI can do that now. The designer's job shifts to taste, refinement, discernment, and strategy.
For non-designers: The gap between "I have an idea" and "I can see that idea on screen" is closing fast.
🛠️ Try It
"Design a simple landing page layout for [your business / your role]. Use clean, modern design principles. Describe the sections, visual hierarchy, and color and font recommendations."
Then take that output to Framer, Webflow, Canva, Manness, or Lovable and see how fast you can go from description to something real. That's the 2026 design workflow. Available to you now.
💳 Buy It?
- Designers: Proceed with caution, not panic. Your skills are not obsolete — but your process is evolving whether you want it to or not.
- Business owners + creators: ✅ Full buy. Going from idea to visual without hiring a designer first is a competitive advantage.
"AI is not replacing creativity. It's removing the technical barrier between your idea and its execution. The ambitious ones collect $200 and pass go." — Jeneba
🔥 HOT TOPIC 1 — Google Personal Intelligence
"Google Wants to Know You Better Than You Know Yourself"
What dropped: Google expanded Personal Intelligence across AI Mode in Search, the Gemini app, and Gemini in Chrome — free in the US.
Connects to: Gmail, Google Photos, and all Google apps to deliver uniquely relevant responses.
What it actually does:
- Shopping recommendations based on your past purchases
- Travel itineraries built from hotel confirmations + past trips
- Tech support tailored to your exact device (from your receipts)
- Layover food suggestions factored around your gate + flight time
The fine print:
- Personal Google accounts only — NOT Workspace, enterprise, or education users
- Gemini doesn't train directly on your Gmail/Photos — but does train on your prompts and model responses
- You can toggle which apps connect — opt-in by default on free tier
"Google is saying: we already have your data. Let us make it useful. The question isn't whether it works — it's what you're trading for the convenience." — Jeneba
For corporate leaders: It's personal accounts only right now — which means Google is in beta. They are absolutely coming for Workspace. Have the data governance conversation with your IT team before it lands.
🧪 Test It: Connect Gmail to Gemini. Ask: "Find the last time I bought running shoes and suggest something that matches."
🔁 Try It: Plan a real upcoming trip using Personal Intelligence only. No manual research. Compare.
💳 Buy It?
- Personal productivity: ✅ Yes, low risk
- Business/enterprise: ⏸️ Hold — watch the Workspace rollout, data governance conversation must happen first
Jeneba's flag for orgs: This is a textbook use case for an AI Readiness Audit — decision-making authority, data governance, and staff readiness all surface here. Reach out if your org needs one.
🔥 HOT TOPIC 2 — Canva AI 2.0
"Canva Went Full AI and Did Not Ask Permission"
What dropped: Canva announced a complete overhaul — AI 2.0 — with a unified prompt-based interface across the entire platform.
What's new:
- Unified prompt interface: Type "create a multi-channel summer campaign" → Canva generates everything, ready to refine
- Persistent memory: Learns your brand, style, and aesthetic over time (no more repeating yourself)
- Object-Based Intelligence: Edit one button or font style without touching the rest of the design
- HTML imports in Canva Code — clutch for email templates
- Unified connector interface: Slack, Gmail, Google Drive, Calendar (more coming)
- PowerPoint-style generation for consultants, marketers, solopreneurs
"I stopped using Canva AI about 3–4 months ago. The old prompt box never understood what I was saying. 2.0 looks like the upgrade I've been waiting for. My first test: refining the logos for this show and Beehiiv ecosystem." — Jeneba
The catch:
- Rolling out as a research preview — first 1 million homepage visitors get access
- Adobe dropped a similar announcement the day before → the creative software industry is in a full sprint
"Adobe should just copy Microsoft's strategy: get integrated into enterprise and ride off into the sunset. Respectfully." — Jeneba
🧪 Test It: If you get early access — prompt a full newsletter cover + social campaign from a single brief. Does the brand output match without manual correction?
🔁 Try It: Use it for one full Beehiiv blog header + newsletter graphic + carousel. Measure time saved.
💳 Buy It?
- Solo creators + small teams: ✅ Yes
- Enterprise: ⏸️ Test brand consistency first — memory is only as good as what you give it
- Agencies: ✅ Yes, with brand documentation in place
🔥 HOT TOPIC 3 — Claude Opus 4.7
"Anthropic Built a Sharper Knife — But Now You Have to Use It"
What dropped: Anthropic released Claude Opus 4.7 — the most rigorous, precise model they've built yet.
What's actually different:
- Self-verification built in: Reviews its own output against your original request before responding. Catches logic errors mid-plan.
- Long-horizon coherence: Holds multi-hour tasks together — the kind of "press play, walk away" workflow 4.6 couldn't sustain
- Benchmark gains: +12 points on CursorBench (58→70%), 3× more resolved tasks vs. 4.6
- 3× better vision: Catches misaligned buttons, off-by-a-pixel layouts in front-end design
- Best PowerPoint generator any LLM has produced (reportedly)
- New effort level: "Extra High" — new default in Claude Code. Use MAX for complex architecture, Extra High for async handoffs, High/Medium for interactive iteration
"I have not noticed a major difference between 4.6 and 4.7 yet — I think this might be a patch for the Midas model leak. But the benchmarks are real." — Jeneba
The plot twist that matters:
4.7 stopped doing your prompt engineering for you. 4.6 quietly fixed vague prompts in the background. 4.7 doesn't. Give it a loose brief, you get a literal answer — nothing more, nothing inferred.
"This is exactly why I teach Think Before You Prompt™. Opus 4.7 just made that framework more important, not less. The model won't compensate for a weak brief anymore." — Jeneba
Usage limit warning: Opus 4.7 uses the max complex architecture. After 1–2 prompts, you're done on a standard plan. Token limits run out fast. This is a $200/month model in practice.
🧪 Test It: Take one prompt you tuned on 4.6 and run it on 4.7 without changes. Watch what happens. Then tighten the brief with explicit constraints and run it again.
🔁 Try It: Give it a complex multi-step task with full context, acceptance criteria, format instructions, and a PRD. See if self-verification catches errors you'd normally find in review.
💳 Buy It?
- Developers + power users with tight workflows: ✅ Yes, this is your model
- Marketers + writers: ⏸️ Not yet ,tune your prompts first, stay on 4.6 for writing tasks. The Through-Line: What This Week Is Actually Telling Us
Here's what connects all three: the bar for the human in the loop is going UP, not down. Google will surface your data in ways you didn't expect. Canva will generate content that misses your brand if you can't articulate it clearly. And Claude will give you exactly what you asked for, nothing more.
This week wasn't about AI getting easier. It was about AI getting sharper. And sharp tools require sharp operators. That's the work. That's what we do here."
🔗 Links from This Episode
- 🔗 Kimi K2.6 Blog Post
- 🔗 Qwen Studio — Try Qwen3.6-Max-Preview
- 🔗 OpenAI Chronicle Research Preview
- 🔗 Claude.ai
- 🔗 Claude as a Design Tool — UX Planet
- 🔗 Google Updates Workspace with AI for the Office — Yahoo Tech
🎙️ About High Tech Tea
High Tech Tea is the podcast where we spill all the AI news you actually need to know, break it down so it makes sense, and tell you what to do and what not to do with it.
Every episode uses the Test It → Try It → Buy It framework so you leave with a clear action, not just information.
3
2 comments
What's tea in AI? 🫖You're Weekly AI News Update 4.23
powered by
Suggested communities
Powered by