In partnership with
MATTYMOE AI · AI workflows for people who’d rather own than rent
Most newsletters will report these two stories this week and stop. I deploy AI into real businesses for a living, so I’m going to tell you what they mean for your Tuesday, and hand you something you can actually run before either model is even public.
⚡ the 60-second version
Anthropic built a model too powerful to sell. Claude Mythos found over 10,000 critical security flaws across partner codebases since April. Instead of releasing it, they locked it to about 150 vetted organizations across 15+ countries running power, water, and hospitals. The public gets Claude Security instead, a tamer tool built on the regular model. why it matters to you: the ceiling of what AI can do just jumped. and the floor, the cheap public version you actually use, always rises to meet it a few months later.
GPT-5.6 is days away (rumored, not out, I’ll confirm the morning it ships). Prediction markets put it above 68% for June. The leaks all say the same thing: it’s built to run multi-step work on its own, with less babysitting, reportedly near Mythos-level at a fraction of the cost. why it matters to you: this isn’t a better email writer. it’s a model built to run whole workflows, not single tasks.
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
🧠 the real story
The last 18 months made the single task cheap. Write the caption, draft the email, clean the spreadsheet. You already know AI can do a task. That’s not news anymore.
What’s coming makes the whole workflow cheap. Not “write me a follow-up” but “watch my inbox, draft replies in my voice, flag the three that need me, schedule the rest.” Mythos proved one model can do the work of an entire team. GPT-5.6 is the version of that they’re actually going to hand you.
The unit of work getting automated is no longer the task. It’s the job.
🛠️ how to use this today (the part nobody else gives you)
Here’s the move, and it works before GPT-5.6 ships, on the $20 AI you already have. The owners who win the week it drops are the ones who already turned a process into a prompt. So let’s do one right now.
Pick the repetitive process you dread most. Then fill in this template. I road-tested it on a roofing client’s post-estimate follow-up, output below so you can see what good looks like.
The template (paste into Claude or ChatGPT):
You are running [PROCESS NAME] for my business, [BUSINESS + WHAT YOU DO]. My tone is [warm / direct / no-nonsense]. Here is the process, step by step: [list your steps the way you’d train a new hire]. When [common thing goes wrong], do this: [your rule]. Produce [the exact output you want: emails, a checklist, a schedule]. Ask me anything you need before you start.
Worked example (roofing, post-estimate follow-up):
You are running post-estimate follow-up for Big Bear Roofing in Charleston. My tone is friendly and local, never pushy. Process: 1) thank them for the walkthrough same day, 2) restate the scope and price in plain English, 3) answer the two questions every homeowner asks (timeline and insurance), 4) follow up day 3 and day 7 if no reply. When they go quiet past day 7, send one soft check-in and then stop. Produce all four emails, ready to send, in my voice. Ask me anything you need before you start.
That’s it. You just built a follow-up system that used to be a “someday I’ll write that down” project. Save the output. Reuse it forever. The week the agentic models land, you paste this in and it runs the whole loop while you’re on a roofFirst, the news that actually matters this week
Two AI stories are worth your two minutes.
One: Anthropic just expanded Project Glasswing, and the headline is wild. They built a model, Claude Mythos, so good at finding security holes (10,000+ critical flaws across partner codebases) that they’ve decided not to sell it to the public. It’s locked to roughly 150 vetted organizations running critical infrastructure. Read that again: the most powerful version is the one you can’t buy. What you can use is Claude Security, a new tool built on the public Claude model that scans your code and suggests fixes.
Two: the internet is buzzing about “GPT-5.6.” Worth knowing, but as of this morning it’s rumor, not reality. Leaks and prediction markets, no official launch. I’ll tell you the day it’s real. I won’t tell you it’s real before it is.
So why does any of this matter to a small business? Because it confirms the one trend you can actually bank on: AI capability is racing ahead, and the cheap public tools you already pay for keep inheriting that power. You don’t need the model that’s too dangerous to sell. You need to stop paying eleven middlemen to use the one sitting in your browser right now.
If you run a small business, you already know the trap. Every problem has a $30/month subscription waiting to “solve” it. Stack twelve of those and you’re not running a business, you’re a tenant. You don’t own your tools, you rent them, and the rent goes up every January whether you used the thing or not.
Here’s what nobody’s saying out loud: most of those subscriptions are now just expensive middlemen sitting between you and an AI you already pay $20 for. The $99/month “AI writer”? It’s a friendly wrapper around the same model you can prompt directly. You’re paying rent on a doorway. So this week, no theory. Here’s the exact swap list.
Why the same news that scares CISOs is good for you
Here’s the thread connecting the Mythos headline to your bank statement. The reason Anthropic won’t sell Mythos is that one model is now powerful enough to do the work of an entire security team. That same leap, the part nobody locks behind a vault, is showing up in the cheap, public AI you already pay $20 for. A subscription was always a bet that you’d never learn to do the thing yourself, because the thing was genuinely hard. Logos, copy, landing pages, spreadsheets of leads. You paid software because the alternative was hours you didn’t have.
That bet broke in the last 18 months, and it’s about to break harder. The same $20 subscription now drafts the copy, builds the landing page, cleans the spreadsheet, and writes the automation. And the next wave is already at the door, whether that’s a confirmed Claude release or the GPT-5.6 everyone’s whispering about (rumored, not out). Either way the direction is one way: the public tools get more capable, and the pricey middleman software gets more pointless. You’re not replacing one tool with another. You’re replacing ten tools with one skill: knowing how to ask.
See you next week,
Matt
MATTYMOE AI. Helping small businesses replace expensive software with AI they actually own.
Follow along: @mattymoe.ai on Instagram