Hello, AI Explorer!
Happy Saturday. This week: Perplexity is now your personal finance AI, Amazon finally revealed $15B in AI revenue it never disclosed before, Oxford's model detects heart failure five years early from a scan you were already getting, and a 20-minute workflow that makes Claude stop writing like AI.
Welcome back to your weekly AI digest: human-curated, zero fluff.
Every Wednesday and Saturday, we bring you:
- The week's most important AI developments (no fluff)
- Hand-picked AI tools actually worth your time
- Expert insights on what these changes mean for you
Let's dive into this week's discoveries! ⚡
🔥 This Week in AI
🔥 This Week in AI |
Apr 11, 2026 |
| 💰 | Perplexity connects your bank accounts, loans, and brokerage data -- personal finance AI has arrived |
| 🏦 | Amazon drops first-ever AI revenue figures: AWS AI at $15B ARR, custom chips crossed $20B |
| ❤️ | Oxford AI detects heart failure 5 years early from a routine CT scan -- 86% accurate across 72K patients |
| 🎬 | HeyGen Avatar V captures your face, voice, and micro-expressions in 15 seconds for unlimited AI video |
| 🤖 | Claude Managed Agents API launches -- build and deploy cloud-hosted AI agents at production scale |
| 🧠 | Meta's Muse Spark multimodal reasoning model opens API preview with parallel multi-agent orchestration |
| 📓 | Google Gemini adds NotebookLM notebook sync -- add a source in one app and it appears in the other |
| 💻 | OpenAI launches $100/month ChatGPT Pro tier with 5x more Codex than Plus -- aimed at heavy coders |
| ⚡ | 4 AI Tools This Week: Filmora 15, Mindgrasp AI, Research Rabbit, Styldod -- all inside. |
⚡ AI Tool Discovery Workflow of the Week
Every edition, we pick one real-world task and show you the exact AI tools and steps to get it done faster. No theory -- just a copy-paste workflow you can run today. This week's workflow: setting up Claude Cowork so it always writes in your voice, not like generic AI. Here is how.
⚡ Steal This Workflow
Set Up Claude Cowork So It Always Writes in Your Voice
For anyone who uses Claude daily but keeps getting output that sounds like AI, not like them. One 20-minute setup that changes every session after it.
| 01 |
Build your ABOUT ME folder -- Claude Cowork Create a folder called "Claude Cowork" with three subfolders: ABOUT ME, OUTPUTS, TEMPLATES. Open Claude Cowork and run this prompt: "Build my about-me.md. Interview me with 20 questions about how I work, what I hate, and what good output looks like. Compile the answers into under 2,000 tokens." That file tells Claude who you are before every single session. |
| 02 |
Write your anti-AI word ban list -- any text editor In ABOUT ME/, create anti-ai-writing-style.md. List every word Claude must never use: delve, harness, tapestry, seamlessly, robust, leverage, multifaceted. Add rules: max 3 sentences per paragraph, no em dashes, no bold for emphasis. This is the file that makes the difference between generic AI text and output that actually sounds like you. |
| 03 |
Set Global Instructions + add voice input -- Wispr Flow Go to Claude Settings > Cowork > Edit Global Instructions. Paste: "Before every task, read every file in ABOUT ME/. If the brief is unclear, ask before writing. Don't fill gaps with filler. Deliver the work." Then install Wispr Flow for voice input -- dictate your briefs at 150 words per minute instead of typing. |
⏱ Total setup time: 20 minutes
Result: Claude reads your context at the start of every session and writes in your voice by default. Most users cut their editing time by more than half. The Google Doc below has the full interview prompt, the anti-AI word list template (80+ words), and the Global Instructions text to copy-paste.
Perplexity launched a Plaid integration this week that lets users
connect bank accounts, credit cards, loans, and brokerage data directly to its Computer agent
PERPLEXITY
| 💰 |
Perplexity Just Became Your Personal Finance AI |
Perplexity launched a Plaid integration this week that lets users connect bank accounts, credit cards, loans, and brokerage data directly to its Computer agent. The result: a full personal finance hub built through plain text prompts, not spreadsheets or manual setup.
You can ask it to build a budget, track net worth, map out a debt payoff plan, or create a retirement dashboard. The same agent already fills out IRS tax forms autonomously. Perplexity's ARR crossed $450M in March -- a 50% jump in a single month -- with this agentic pivot as the primary driver.
What it actually does:
• Plaid's 12,000+ bank network feeds your checking, savings, credit, and brokerage data into Computer in read-only mode
• Type "show me where my money went last month" and it generates a categorized breakdown with charts
• Same system handles US tax returns -- it reads your documents, fills IRS forms, and reviews professional-prepared returns
Why it matters:
Perplexity spent years trying to out-Google Google. Its Computer launch changed the game -- now it is competing with Mint, TurboTax, and every finance app on your phone. The pattern is the same one that made Notion dangerous: one AI agent that keeps absorbing categories that used to require separate apps.
AMAZON
| 🏦 |
Amazon's $200B AI Bet Finally Has Numbers Behind It |
Amazon CEO Andy Jassy published his annual shareholder letter this week with a notable first: actual AI revenue figures. AWS's AI business crossed $15B in annualized revenue -- a number Amazon had never disclosed publicly. That is 260x where AWS itself stood at the same stage of its growth.
The custom chip story is even larger. Trainium, Graviton, and Nitro chips together crossed $20B in annual revenue. Jassy floated selling "racks of them to third parties" -- a direct shot at Nvidia's data center dominance. Two unnamed AWS customers tried to buy Amazon's entire Graviton supply for 2026. Amazon declined to protect other clients.
The numbers that matter:
• AWS AI arm: $15B annualized revenue -- 260x AWS's own trajectory at the same point in its history
• Trainium + Graviton + Nitro chips: $20B in annual chip revenue combined
• Demand so high two customers asked to buy Amazon's entire Graviton output for 2026 -- Amazon refused to let them
Why it matters:
If you have been tracking model releases as the main scorecard for the AI race, Amazon looks quiet. The chip numbers tell a different story. Nvidia has held AI compute for years, but the $20B in Amazon silicon revenue shows the supply side is finally getting serious competition -- at the exact moment enterprise demand has never been higher.
| ❤️ |
Oxford AI Spots Heart Failure 5 Years Before It Develops |
Researchers at Oxford published a system that reads routine cardiac CT scans -- scans patients are already getting for chest pain -- and detects invisible textural changes in the fat surrounding the heart. Those changes signal that the heart muscle underneath is inflamed and at risk. The model flagged patients at high risk with 86% accuracy across 72,000 patients. In the highest-risk group, 1 in 4 developed heart failure within 5 years -- 20 times the rate of those flagged as safe.
WHY IT MATTERS:
Heart failure's biggest problem is timing -- doctors usually cannot act until the damage is done. A warning built into scans patients are already scheduled for shifts the equation from reaction to prevention. Oxford is seeking NHS regulatory approval now and expects to extend the model to all chest CTs within months.
| 🎬 |
HeyGen Avatar V: 15 Seconds to Your AI Digital Twin |
HeyGen launched Avatar V this week. Record a 15-second webcam clip, and the model captures your face geometry, micro-expressions, lip movements, and gestures. Use that identity reference to generate unlimited studio-quality video -- in 175 languages with accurate lip sync -- without touching a camera again. The identity holds across different angles, outfits, and scenes without drifting.
WHY IT MATTERS:
Identity drift has been the main reason AI avatars stay in the lab rather than in production. Creators building course content, LinkedIn videos, or product demos need the same face across every clip. Avatar V solves that with a 15-second capture -- making consistent AI video finally practical for individual creators, not just enterprise production teams.
⚡ Coming in Hot
AI Tools of the Week
4 tools worth adding to your workflow right now.
| 🎬 |
01 Filmora 15 AI VIDEO EDITOR Filmora 15 has Sora 2 and Veo 3.1 text-to-video built directly into its timeline editor. Generate a clip, drop it into your sequence, and export -- no switching apps. Voice cloning, AI lip-sync translation for 40+ languages, and 4K upscaling are all included. At $4.17/month it is 90% cheaper than Adobe Premiere while doing things Premiere still cannot. |
| 🧠 |
02 Mindgrasp AI AI RESEARCH Upload any document, video, audio, or URL and Mindgrasp generates concise notes, flashcards, summaries, and answers to your questions -- all from the source material. Students use it for lecture recordings. Researchers use it to extract key arguments from 100-page papers in minutes. Strong free tier to start with, no credit card needed. |
| 🔬 |
03 Research Rabbit FREE Research Rabbit maps academic papers visually -- add one paper and it shows you everything that paper cites, everything that cites it, and clusters of related work you would never have found searching manually. Think Spotify's "discovery mode" but for academic literature. Free for everyone, no subscription needed. The tool researchers call Zotero's smarter sibling. |
| 🏠 |
04 Styldod VIRTUAL STAGING Upload a photo of any empty room and get a fully furnished, professionally staged version back within hours. Real estate agents use it to market properties before furniture has even arrived. Pay per image -- no subscription required. The before/after difference in listing click-through rates is significant enough that major agencies have made it standard practice. |
⚡ Quick Hits
🤖 Claude Managed Agents is live
Anthropic launched composable APIs for building and deploying cloud-hosted AI agents at production scale -- decoupled harness, sandbox, and session interfaces all included. Get the technical details →
🧠 Meta Muse Spark opens API preview
Meta's new multimodal reasoning model supports parallel multi-agent orchestration and is now available to approved developers via private API. Think GPT-4o's vision capabilities with a multi-agent layer on top. Request API access →
📓 Gemini + NotebookLM now share one source library
Google added notebook functionality to the Gemini app with NotebookLM sync -- add a PDF in Gemini and it appears in NotebookLM automatically. Useful if you use both apps for research without wanting to manage two separate document libraries. See what changed →
💻 OpenAI launches $100/month Pro tier with 5x Codex
The new tier targets developers running heavy agentic coding workflows who have been hitting Plus rate limits. Comes as frustration with Claude's usage caps was already pushing developers to look for alternatives. See the pricing breakdown →
🔊 ElevenLabs voice AI now runs fully on your own hardware
On-device models are now in early access -- inference and audio processing run entirely within your infrastructure. Built for edge hardware like vehicles and wearables. On-premise GPU deployment for government and regulated industries. Read the announcement →
That's all for this edition! What tools are you most excited to try?
Reply to this email - we read every response and use your feedback to improve future editions.
See you on Wednesday,
The AI Tool Discovery Team
Feedback
What'd you think of today's newsletter?
Vote below to let us know how we're doing.