| Friday · Apr 24, 2026 |
Issue № 001 |
16 min read |
|
|
The FindSkill Weekly Brief
The Skill
Just for FindSkill Pro members. The AI news that actually matters for your work.
|
|
Pro Members Only
|
A Private Brief
|
Not Published Anywhere
|
|
|
|
Mia
AI Learning Editor · FindSkill.ai
|
Welcome to Issue 001 — the private brief only FindSkill Pro members get.
I'm Mia, and every Monday I'll land in your inbox to make sense of the week's AI news — without the jargon, the hype, or the "10 things you MUST know" energy. This doesn't get published publicly. It goes to Pro members, and that's it.
Think of this less like a newsletter and more like a letter from a friend who spends too much time on AI Twitter so you don't have to. When you hit reply, I actually read it. So tell me what lands, what doesn't, and what you want more of.
This week: three stories, one term worth knowing, a prompt torn apart, and two brand-new recurring segments — one for using AI at your job, one for making money with it.
— Mia
|
01 |
This Week in AI
Three stories worth your attention
|
Model Release
Anthropic ships Claude Opus 4.7 — and it quietly changes how agents work
Anthropic dropped Opus 4.7 this week with a focus on long-horizon reasoning. The headline isn't the benchmarks — it's that the model now holds context across 20+ tool calls, which is exactly the bottleneck agent builders have been hitting for two years. Early users report roughly 3x improvements on multi-step workflows. Same pricing as 4.6.
|
What it means for you
If you've been waiting to build an AI agent for your own workflow — expense categorization, email triage, research digests — the "it forgets halfway through" problem is finally solvable. Start small.
|
Policy
The EU AI Act's second enforcement phase just kicked in
Most of the coverage focused on fines. The more interesting shift: companies selling AI-powered tools into the EU now have to publish model documentation — what's in the training data, where the model performs poorly, known limitations. Think nutrition labels for AI.
|
What it means for you
When a tool starts publishing its model card, read it. Free, legally-mandated honesty — use it to pick tools that are transparent about their limits.
|
Trend Spotlight
A quiet revolution: AI that remembers everything you've ever read
Three "persistent memory" AI notebooks shipped this month. The pitch: dump in every article, podcast, and PDF you consume, then ask questions across the whole archive later. I've been testing one for two weeks. It's not there yet — but the trajectory is clear. Computing is slowly becoming remembering.
|
What it means for you
The skill of "capturing input systematically so AI can work with it later" is about to become a meta-skill. Full head-to-head of the top three coming in Issue 004.
---
|
|
02 |
Term of the Week
The one concept to understand this week
|
| |
|
Term №001
RAG
< retrieval-augmented generation >
RAG is how an AI looks things up before it answers you. Instead of only relying on what it learned during training, it pulls in fresh info — from your documents, a website, a database — and uses that to respond.
|
Think of it like this →
An AI without RAG is like a student taking a closed-book exam. An AI with RAG is the same student, but you've handed them exactly the right textbook, opened to the right page, right when they need it.
|
|
⚠ Common misconception
"RAG makes AI smarter." It doesn't. RAG makes AI better informed. A well-RAG-ed AI can cite fresh sources but still reasons exactly as well (or poorly) as it did before.
|
Where you'll hear it: every AI tool that "talks to your docs." Notion AI, ChatGPT's custom GPTs, any enterprise AI search — they all use RAG under the hood.
Full RAG course →
|
|
03 |
Deep Insight
A prompt I actually use, torn apart
|
Most "prompt guides" online are theater. This is one of the five prompts I run every week — and why each line is there.
I use this one for pulling structure out of messy notes: voice memos, half-finished thoughts, meeting scrawl. Most note-organizer AIs give you sycophantic bullet-point mush. This one doesn't.
PROMPT You are a ruthless editor, not a cheerleader.
Below is a raw brain-dump. Your job:
1. Extract the 3 strongest ideas
2. Write each in one sentence a 10-year-old understands
3. Flag any idea that sounds smart but says nothing
Brain-dump:
<<< paste notes here >>> |
Four things are doing the work here:
| i. | "Ruthless editor, not a cheerleader." Kills the sycophancy that ruins most AI outputs. Models are trained to be pleasant by default. Explicitly flipping that switch gives you cut-throughs instead of compliments. | | ii. | "The 3 strongest ideas." Forcing a number forces ranking. Without it, you get a soup of every idea, equally weighted. With it, you get priorities. | | iii. | "One sentence a 10-year-old understands." The compression test. If the AI can't hit this, the idea isn't clear yet — often including to you. Use it as a diagnostic. | | iv. | "Flag any idea that sounds smart but says nothing." The secret weapon. AI loves this instruction — it'll call out your own vague thinking. About 20% of what I feed this prompt comes back flagged. Those are the ideas I needed to rewrite anyway. |
What this looks like in practice Input · raw note okay so thinking about pricing — maybe lifetime is doing too much heavy lifting, monthly churn is concentrated in months 1-2, also there's something here about how the people who buy annual never churn which is wild, need to figure out if that's a filter effect or a commitment thing, also considering whether to bundle something exclusive for monthly to reduce early churn, community access maybe? | | | Output · after the prompt 1. Monthly churn is concentrated in the first two months — fix that window, not the rest.
2. Annual and lifetime buyers never churn, which probably reflects who buys them, not what they buy.
3. Bundling community access into monthly plans might reduce early churn — but this is untested speculation, not a conclusion. |
Try it on your own messy notes this week. Reply and tell me what it flagged — I read everything.
|
04 |
The Workflow
One way to use AI at your job this week
|
| |
Let AI audit your week before you plan the next one.
Sunday evening — or whenever you plan your week — try this: paste your last 7 days of calendar events into Claude or ChatGPT, alongside your top 3 goals for the quarter. Ask: "Rate how well I spent my time against these goals, from 1 to 10. Be specific about which hours helped and which were theater."
Why it works: you already do this review in your head, poorly. AI does it in 60 seconds with no emotional attachment to the meetings you sat through. It'll tell you what you've been avoiding. You'll flinch a little. Then you'll adjust.
|
Do this week
Run the audit once. Even if you hate the output, you'll learn something about how you actually spend your time.
|
|
|
05 |
The Side Play
One way to make money with AI this week
|
| |
| ◆ Income Idea · Play №001 |
Sell an AI workflow pack for a role you already understand.
Pick one profession you know well — teachers, real estate agents, therapists, freelance writers, or whatever you do now. Build 8–10 specific AI prompts and mini-workflows that solve real problems for that role. Package as a Notion page or PDF. Price at $29–49. Sell on Gumroad or Lemon Squeezy.
Why it works: most people in specialized jobs don't know AI well — and they don't want generic guides. They want "AI for therapists," not "AI for everyone." A niche prompt pack can hit $1–3k/month with almost no maintenance, because prompts keep working as long as the models do.
|
Do this week
Write down three jobs you've held or know intimately. One of them is your pack.
|
|
|
06 |
The Stack
Three tools I'm testing this week
|
Not affiliate picks. Not sponsored. Just what's open in my browser tabs.
01 |
Granola
Meeting notes · Mac only
Joins your meetings silently, takes notes based on what you type plus what's said, produces a summary that sounds like you wrote it. I've tried six of these. First one that doesn't feel like homework.
|
02 |
NotebookLM
Research RAG · free
Google's quietly-best AI product. Upload up to 50 sources, ask questions, get citations that actually point to the right passage. Best free demo of RAG that exists.
|
03 |
Cursor
Code editor · $20/mo
If you write any code — even tiny scripts — this replaces your text editor. Reads your whole codebase as context, so suggestions are actually good instead of generic.
|
|
07 |
Inside FindSkill
What's new for members this week
|
| New |
Product Photos with ChatGPT Images 2.0
Ship e-com product shots that actually convert, no studio needed. Start the course →
|
|
| New |
AI Therapy Notes: The HIPAA-Safe Workflow
For therapists who want AI on notes without a compliance disaster. Start the course →
|
|
| New |
Virtual Staging with ChatGPT
For real estate agents: stage empty listings in minutes, not days. Start the course →
|
|
|
|
What's one AI thing you tried this week — and did it actually help?
Reply to this email. I read every one. Worst case, you help me pick next week's Term of the Week.
↵ Hit reply
|
|
The Skill · by FindSkill.ai
|
|