| Monday · Jun 8, 2026 |
Issue № 008 |
15 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 008.
I'm Mia — every Monday I turn the week's AI noise into the part that actually changes how you work. No jargon, no hype.
The biggest thing that happened this month wasn't a new model. It was a price change. GitHub Copilot's flat fee died on June 1; Cursor and Windsurf had already metered you; and the staggering bills behind all of it just became visible. So this whole issue is about one shift — all-you-can-eat AI is ending, and how to live on the meter without a nasty surprise.
A quick reply note: last week I asked which AI tool you use that your IT team doesn't officially know about. The honesty floored me, and the most-named answer wasn't the one I expected. More on that soon.
As I write this, Tim Cook is hours from his final WWDC keynote. Apple is part of this same story — see Story 3.
— Mia
|
01 |
This Week in AI
Three stories worth your attention
|
Business
Copilot's flat fee is gone — and a week in, the meter is biting.
It went live June 1: every GitHub Copilot plan dropped its all-you-can-eat structure for usage-based AI Credits. Your seat price didn't move — Pro is still $10 — and plain autocomplete stays free. What changed is that the agent and chat features now spend, and a week in, developers are stunned at how fast. Reactions ranged from "$38 became $847 for the same work" to one line I keep seeing: "Copilot used to feel like a subscription. Now it feels like a meter." The sharpest diagnosis pinned the real culprit — "agent mode fires Opus-tier reasoning at every task" — the most expensive engine running on everything. They're calling it the Tokenpocalypse.
|
What it means for you
Open Copilot's billing settings and check one number: your additional-spend budget. Leave it at $0 and you can't be charged over your plan — it just pauses until the month resets. And remember what actually burns credits: not autocomplete, but agent runs and long chats. The fix isn't panic. It's knowing where the meter is.
|
Tool Launch
OpenAI just pointed its coding agent at everyone who can't code.
OpenAI spent the week turning Codex — the tool built to write software — into something for people who never will. The update added six role packs aimed at analysts, marketers, salespeople, designers, and finance teams (no coding required), dropped Codex right inside the ChatGPT app, and previewed Sites: describe a dashboard, planner, or small internal tool in plain English and it builds a shareable web app. The tell is in OpenAI's own numbers — non-developers are already about 20% of Codex users and growing more than three times faster than developers, with over five million people using it weekly. The "AI that writes code" is quietly becoming the AI that builds your spreadsheet.
|
What it means for you
This is the agent wave reaching white-collar work, not just engineering. You don't need to learn to code to use it — you need to learn to describe what you want built. Try the no-code side once this week: take a messy process you own and ask it to turn that into a simple shared tool. Worst case, you find out what's now possible without a developer.
|
Trend Spotlight
Even Apple decided not to build its own AI.
Here's a fact that reframes the whole year: Apple — the most cash-rich company on earth — looked at building a frontier model for the new Siri and chose to rent one instead. It's paying Google a reported ~$1 billion a year to license a custom version of Gemini, a deal Bloomberg first surfaced last autumn. The rebuilt, chatbot-style Siri it pays for gets its public reveal at today's WWDC keynote, happening roughly as this lands in your inbox. Set the keynote theatrics aside; the money is the story. When the company that loves owning its whole stack won't build this one piece, that tells you something permanent about what these models cost to make.
|
What it means for you
The lesson isn't about Apple — it's the mental model that runs through this entire issue: you almost never need to own the AI. You need to rent the right one for each job. Even Apple buys by the task. So should you — which is exactly what the metered era is about to make unavoidable.
---
|
|
02 |
Term of the Week
The one concept to understand this week
|
| |
|
Term №008
Tokens
< tokens >
The little chunks of text an AI reads and writes — each one is roughly three-quarters of a word, about four characters. They used to be invisible. Now they're the unit every AI meter counts: your bill, your limits, even your speed are all measured in tokens.
|
Think of it like this →
Picture a court interpreter who bills by the word — and counts both sides of the conversation. Tokens work the same way. You're not charged per question; you're charged per chunk of text going in and coming back out. "Fix this sentence" is a few cents of words. "Read these 40 pages and rewrite them" is a small novel's worth — even though both felt like one click.
|
|
⚠ Common misconception
"A token is a word, and only my prompt costs anything." Two errors in one. A token is about ¾ of a word, so 1,000 words is closer to 1,300 tokens — and the AI's reply usually costs more per token than your question did. Worse, in a long chat the whole conversation gets re-sent every turn, so a thread quietly gets pricier the longer it runs. The meter counts everything, both directions, every time.
|
Where you'll hear it: On Copilot's new bill. On every API pricing page ("$5 per million tokens"). In "context window" sizes — how many tokens a model can hold in its head at once. And in the sentence you'll hear all year: "that model's way cheaper per token."
AI Fundamentals — how tokens, context, and cost actually work →
|
|
03 |
Deep Insight
The all-you-can-eat era of AI is over — here's the new menu, and how to order
|
The Copilot uproar this week isn't really about Copilot. It's the moment a hidden subsidy ended in public — and most people are about to learn the new rules the expensive way.
For about three years, AI felt like a flat-rate buffet: twenty dollars a month, eat all you want. That was never the real price. It was a land-grab — companies eating the cost to get you hooked. This month the subsidy started ending out loud, and it isn't coming back. Here's the pattern, the reason behind it, and the four rules that keep you off the wrong end of a "why is my bill $847" Monday.
Cursor went first, back in June 2025. It swapped unlimited requests for a credit pool, and most people called it a one-off. It wasn't. Windsurf metered in March 2026. GitHub Copilot flipped on June 1. The dominoes are all falling in one direction, and there is no domino falling back.
Same twenty dollars, four different meters. Here's the quiet trap: Cursor, Windsurf, Claude Code, and Copilot all still show a roughly $20 "Pro" tier — but that twenty now buys four genuinely different things. Credit pools, daily quotas, rolling time-windows, per-token billing. The price stayed familiar so the change stayed invisible. The buffet sign is still up; the food went à la carte.
This is physics, not greed. Running these models is staggeringly expensive, and the numbers leaked all month. Anthropic just committed to over $1.25 billion a month — a month — for raw computing power. OpenAI is forecast to lose around $14 billion this year, burning more than half of its revenue. And Apple, sitting on more cash than some national governments, decided renting Google's model for a billion a year beat building its own. When the suppliers lose money on their heaviest users, "unlimited for $20" was always borrowed time.
The skill just changed under your feet. Last year the valuable skill was prompting. This year it's ordering — knowing which AI to use for which job, because they have stopped being the same price. A throwaway question and a 40-page analysis used to cost you exactly the same: nothing extra. They don't anymore, and pretending they do is how the bill sneaks up on you.
The thing almost nobody is saying. the people getting burned aren't overusing AI — they're using one expensive setting for everything. The sharpest line I read all week nailed it: the shock isn't the token price, it's that "agent mode fires Opus-tier reasoning at every task." That is taking a limousine to buy milk, a hundred times a day. The answer was never "find the cheapest tool." It's matching the tool to the task.
The four rules for the metered era
| 1. | Right-size the model. Use the cheap, fast tier for volume, drafts, and anything structured; pay for the flagship only when judgment, nuance, or a client is on the line. Most work does not need the limo. | | 2. | Cap before you spend. Set every metered tool's overage budget to $0, or to a number you chose on purpose. A meter you haven't capped is just a bill you haven't seen yet. | | 3. | Watch the agents, not the autocomplete. Quick chats and suggestions are cheap. It's the autonomous "go do the whole thing" runs that drain a month in an afternoon — fire those on purpose, not by reflex. | | 4. | Keep one free escape hatch. A free or local model (see The Stack) for the low-stakes 80%, so your real money only goes to the 20% that actually earns it. |
None of this means use AI less. It means stop paying flagship prices for milk runs. The audit in the next section takes fifteen minutes and shows you exactly where your meters are — do it before one of them finds you first.
|
04 |
The Workflow
One way to use AI at your job this week
|
| |
Find your AI meters this week, before one of them finds you.
Open a blank doc and list every AI tool you pay for or lean on for work. For each one, answer a single question: how does it charge me now? Sort them into three buckets. Flat — still genuinely all-you-can-eat (most consumer ChatGPT, Claude, and Gemini plans, for now). Metered — has a credit pool or usage limit you can actually drain (Copilot, Cursor, Windsurf, anything that says "credits"). Per-token — anything you use through an API key, where every call has a price. Then do two concrete things. First, on every metered tool, set the overage or additional-spend budget to $0 or a deliberate cap — most people have never opened that screen. Second, name your single biggest "limousine to buy milk" habit: the one place you run the flagship model, or agent mode, on a task that plainly didn't need it.
Why it works: This is the personal version of what every company is scrambling to do this month — except it takes you fifteen minutes instead of a committee. The people getting surprise bills are the ones who never looked at the meter. Capping is the 80% fix: it makes a runaway charge structurally impossible. And naming one oversized habit is where the real savings hide — usually it's agent mode or a top-tier model doing work a cheaper one would nail.
|
Do this week
Tuesday morning, set every metered tool's budget cap. Then take one recurring task you do on the flagship and run it once on the cheap or fast tier instead. If you can't tell the difference in the result, you just found money you were burning for no reason.
|
|
|
05 |
The Side Play
One way to make money with AI this week
|
| |
| ◆ Income Idea · Play №008 |
Become the person who keeps small businesses' AI bills from exploding.
As of June 1, every small business and busy solo professional using metered AI tools has a new fear they can't name yet: the surprise bill. They've read the "$38 became $847" stories. They have no idea which of their tools is metered, no caps set, and nobody whose job is to watch it. That's your opening — call it an AI Bill Review. You audit their AI subscriptions and usage, kill the tools they're double-paying for, set every overage cap, right-size which model does which job, and hand back a one-page AI spend plan. Sell it two ways. As a one-time audit — $150–$500 for a solo, $500–$2,000 for a small team. Or, better, as a quarterly retainer: $200–$400 a month per client to keep their stack lean and their caps sane as prices keep shifting. Pitch it through your LinkedIn, your local business network, or one industry you already understand.
Why it works: The demand signal turned on overnight, on a specific date, for a specific reason — that's rare and you should use it. The work is judgment, not code: you're the calm person who reads the meters so they don't have to. Realistic revenue is $1,000–$4,000 a month once you hold five to ten retainer clients, and it compounds because the pricing landscape keeps changing — which means the review is never "done." Niche to one vertical (agencies, law firms, clinics) and your advice gets sharper and your referrals get faster.
|
Do this week
Run the Section 04 audit on yourself first — that's your demo and your proof. Then message three business owners or freelancers you actually know and offer them a free fifteen-minute AI bill review. One of them will say yes, and one yes is how this starts.
|
|
|
06 |
The Stack
Three tools I'm actually testing this week
|
01 |
Google AI Studio
Use a flagship model for free · Free (sign in with Google)
This is the cleanest way to use a top-tier Gemini model without touching a meter. You get a generous free workspace — paste in text, ask, iterate — and for low-stakes work it quietly replaces a paid seat. The honest catch: the interface is built for tinkerers, not beginners, and on the free tier your inputs can be used to improve Google's products, so keep anything sensitive out of it. For "do more without paying more," it's the first tab I'd open.
|
02 |
LM Studio
Run AI on your own laptop · Free
Download it, pick a model, and you're running AI entirely on your own machine — no account, no meter, nothing leaving your device. It won't match the flagships on hard reasoning, and you'll want a reasonably modern laptop, but for summarizing, drafting, and anything private, "free forever and fully offline" is hard to beat. The install is genuinely a few clicks now. This is the escape hatch from rule four, made real.
|
03 |
Portkey
One key for every model, with a spend dashboard · Free tier
This one's for the slightly more technical among you, but it's in the stack because it's the cleanest answer to "where is my AI money going." Portkey sits in front of your AI calls, lets you route each job to the cheapest model that'll do it, and shows your spend per model and project in one place — before the bill surprises you. The free tier covers a solo setup. Even if you never wire it up, knowing this category exists is half the battle in the metered era. ---
|
|
07 |
Inside FindSkill
What's new for members this week
|
| New |
GitHub Copilot alternatives after the price shock
The plain-English companion to today's issue: what changed June 1, what it costs now, and the cheaper tools worth a look — alongside this week's other cost explainers (is Claude really free, ChatGPT's ad-supported tier, and a no-code Copilot agent walkthrough). Browse the blog →
|
|
| New |
Microsoft Copilot in Excel for Accountants & FP&A
The metered tool from Story 1, pointed at where the money actually is. Eight lessons plus a certificate on running real finance work — variance analysis, reconciliations, forecasts — through Copilot, with the prompts that keep it accurate. Start the course →
|
|
| New |
Five more "AI for your job" courses
Fresh this week for dietitians, loan officers, career coaches, property managers, and insurance agents — each an eight-lesson, job-specific toolkit, not generic AI theory. Plus a full accounting set (QuickBooks AI, fraud detection, running your firm with AI). See all courses →
|
|
|
|
Which one AI tool would you happily pay double for — and which three could you cancel tomorrow?
Reply and tell me — I'm building a "worth-the-meter" list, and your three cancellations help everyone on it. Worst case, you talk yourself out of a subscription you forgot you were paying for.
↵ Hit reply
|
|
The Skill · by FindSkill.ai
|
|