👋 Hello hello,
You ever notice how AI announcements used to feel like once-every-couple-months events? Now there are three meaningful ones before your second coffee on a Tuesday.
OpenAI reshuffled how ChatGPT's models work for different subscription tiers, and honestly it's one of those changes that sounds boring until you realize what it means for free users. Anthropic updated Fable 5's biology guardrails, which is either a minor tweak or a big deal depending on how you've been using it for health and science questions.
And then there's the Agent Plugins spec, developed with AWS, Cursor, GitHub, VS Code, and Vercel, which is the kind of move that doesn't make headlines but tends to matter a lot six months later.
🚨🚨 This Friday on PracticalyAI Live: we’re building one subscriber’s dream system — an agent that turns any video you save into content written in your voice. 🚨🚨
All details are at the end. Okay, let’s get into it.
🔥🔥🔥 Three Exclusive AI Updates
OpenAI just reorganized how its models are served inside ChatGPT, and the changes are more significant than the announcement suggests. GPT-5.6 Sol now handles both quick responses and deep reasoning for Plus and Pro subscribers. That means one model covering both modes, with OpenAI promising it's more factual and more focused than what came before.
The bigger news, however, is that free and Go-tier users are getting unlimited text chats with GPT-5.6 Luna. That's a real access expansion. This is part of OpenAI's ongoing push to stop making intelligence feel like a metered resource.
Worth checking back in if you've been sleeping on the free tier.
Anthropic updated Claude Fable 5's biology safeguards, and the number that matters here is 85 percent. That's how much they reduced false positives in biology-related responses across their product surfaces. Fable was previously refusing or falling back on a lot of questions that were totally legitimate, including everyday health and educational topics.
The fix means Fable can now handle a much wider range of biology and health questions without unnecessarily bouncing you to Opus 5. That said, Anthropic is keeping Fable on a shorter leash for anything that touches dual-use territory: virology, toxicology, and molecular design still route to Opus 5. So professional biology researchers and drug developers aren't fully served yet, but Anthropic says they're working on trusted-access pathways for that.
This one flew under the radar but it shouldn't. OpenAI, working with AWS, Cursor, GitHub, VS Code, and Vercel, just released an open standard for agent plugins. The idea: you build a plugin once, and it works across any client that supports the spec. The format packages Agent Skills and supports MCP server configurations in a shared way.
Right now, building for agents means building separately for every platform. This standard, if it gets broad adoption, changes that significantly. Think of it like what RSS did for content distribution, or what OAuth did for login. You don't have to get excited about the infrastructure to benefit from it later.
🔥🔥 Two Pro AI Tips
If you left ChatGPT a few months ago and just came back, the app you returned to is genuinely different. Here’s 5 new features you may or may not know about:
There's now an official Chrome extension that sees your open tabs and lets you ask across all of them.
There's computer use on desktop, where ChatGPT can operate your actual computer in the background while you do something else.
You can describe a dashboard or web app and it builds and hosts it with a live link.
Your skills are now in an open portable format, so they travel with you.
And the wildest one: record yourself doing a boring workflow once, and it becomes a reusable automation with new inputs.
Fair warning: most of this is on paid plans and Mac first. But if you're already paying, you might be leaving a lot on the table.
Turns out most of the AI tools you're paying for have free, open-source equivalents that most people have never heard of:
Wispr Flow → Handy (local voice-to-text dictation, runs entirely on your laptop)
Claude Design → Open Design (prompt-to-product UI canvas, plugs into whatever AI you already use)
ElevenLabs → Voicebox (open-source voice generation in multiple languages)
Perplexity → Perplexica (self-hosted, open-source AI search)
The subscriptions aren't usually buying you better AI. Mostly they're buying you convenience and polished UX. That's worth something, but it's useful to know what you're actually paying for. Here’s a quick comparison!
🔥 Claude Skills That Save You Hours
Emil Kowalski put together a GitHub repo of nine practical Claude skills specifically for designers and engineers, and it's one of the more immediately useful collections we've seen. Here's how to use it:
Go to https://github.com/emilkowalski/skills and browse the skills listed.
Open the skill file relevant to your current project (design tokens, component generation, code review, etc.).
Copy the skill prompt and paste it at the start of a Claude conversation.
Describe your specific task or paste in the relevant code or design context.
Let Claude work through it using the skill's structure, then iterate from there.
These aren't generic prompts. They're opinionated, task-specific workflows that actually reflect how designers and engineers work. Start with whichever one solves your most annoying current problem.
🎤 PracticalyAI Goes Live
Its Friday, which means we’re going live! Today we’re building the workflow that one of you asked us to build for you:
"Take a video I like on Instagram, transcribe it, and have an agent rewrite it in my language patterns so I can recreate it as my own video or article."
We’ll build it live, start to finish — how we think through it, where we get stuck, and what it looks like running. How to use skills, connectors and Claude Code. At the end, one number on screen: the hours a month this task used to cost, and what it costs now.
A friendly reminder that your submission could be the topic of our next build. If we pick yours, we build it whether you're there live or not, and you get the finished system either way. Everyone can submit.
Also, the last 15–20 minutes of the session is a Q&A. Remember, there is no question too basic so bring them all.
🗓️ When: Friday, August 7 2026 @ 10AM PT (1PM ET)
📹️ Where: Virtual. (Can't make it live? RSVP anyway, every registrant gets the recording.)
Did you find today's post useful?
Until next time,
Team @PracticalyAI
