In partnership with
You're about to have a brand kit and a daily AI briefing — before lunch. Read all the way through; there's something at the bottom for people who finish.
SIGNAL PREVIEW
📡 Brand kit built in 14 minutes — three inputs →
📡 Claude now drops a daily brief automatically →
🛠️ Prompt included: 4 dark social slides, 10 minutes →
---
⚡ GPT-5.5: work agent, not chatbot → Deep Research: 160 searches, one query → Claude goes native on AWS → Google's AI morning brief: real, waitlisted →
Jump to the one that matters to you ↓
This Week's Poll
What does your current AI setup actually look like?
📡 ChatGPT's Image Model Just Became a Legitimate Work Tool — Here's What Real People Are Doing With It
What's new: GPT Image 2, released April 21, just recorded the largest quality jump in Image Arena history — 1,512 points, 242 ahead of the next closest model. The reason isn't aesthetics: text inside images finally works. Menus, infographics, social graphics with real copy — outputs that would have been unusable six months ago are now production-ready on the first attempt.
Dan's take: The model that ships today does something the previous version couldn't — it thinks before it generates. It plans the layout, checks text accuracy, and self-corrects before producing a pixel, which moves it from "AI art toy" to something closer to a junior designer who follows a brief. The Reddit threads proving this are more convincing than any benchmark.
The test: I gave it a brand brief — a perfume brand called L'Aura, three vibe words, one admired brand — and got a 9-section brand kit plus 5 platform campaign images in 14 minutes. The more interesting use cases are from people I don't know: a Shopify store cutting photo shoot costs by 80%, a demolition engineer annotating technical photos, a teacher building classroom diagrams that actually render text correctly. Structured input, production-ready output — the pattern holds everywhere.
Brand kit image for L'Aura
That brand kit above is what 14 minutes and four inputs produced.
I ditched ChatGPT for visuals for a while and switched to Nano Banana Pro.
GPT Image 2 changed that.
So I built Signal Studio.
It’s a custom GPT that helps you turn a basic brand idea into real marketing visuals.
Give it your brand name, what you sell, and three words that describe the vibe.
It can create brand kits, social posts, ad creatives, product mockups, email headers, carousels, banners, and campaign images in the right sizes for each platform.
No long prompt needed.
It’s free on any ChatGPT plan.
Pro users can generate up to 8 images in one response.
📡 Claude Connected to Gmail, Calendar, and Notion — Now Drops a Structured Brief Into Notion Every Day at 5:45 PM
Gmail morning brief
What's new: Your inbox and calendar can now feed an AI agent that works while you're not. A non-technical person can connect Claude to Gmail, Google Calendar, and Notion through Claude's connector menu — each one is a new browser tab, login, done — then use Claude Code's /schedule command to create a cloud-based routine that drops a structured evening brief into Notion automatically every day. This requires Claude Pro, Max, Team, or Enterprise — free users cannot access /schedule or cloud routines.
Dan's take: I'll be honest — when I opened Claude Code in terminal, I assumed I was doing developer things. It still uses a terminal, but the setup felt more like logging into apps than writing code. The moment Claude Code confirmed it could already see the Gmail, Calendar, and Notion connections I'd made through claude.ai — and asked what I wanted it to do with them — that's when I understood what "agentic" actually means for normal people.
On permissions: Claude requires explicit Google auth for each connection. Connector data is not used for training.
The test: I gave Claude Code one prompt describing four sections — Today's Meetings | Urgent / Action Required | Emails to Reply | FYI — and told it to create a new Notion page titled "Evening Brief — [today's date]" under a specific parent page. Took approximately 1 minute 12 seconds from submitted prompt to confirmed routine. Google CC (labs.google/cc) is building the same idea — currently waitlisted, U.S. and Canada only, paid users prioritized.
Want the full setup?
Reply ROUTINE and I’ll send you the step-by-step video showing exactly how a non-technical person can set this up.
No coding lesson. No developer setup. Just: connect Gmail, Calendar, and Notion → give Claude one routine prompt → get your daily brief dropped into Notion automatically.
4x more context into every prompt. Zero extra effort.
You think faster than you type. Which means every typed prompt leaves out the constraints, examples, and edge cases that would have made the output actually useful.
Wispr Flow turns your voice into paste-ready text inside any AI tool. Speak naturally — include "um"s, tangents, half-finished thoughts — and Flow cleans everything up. You get detailed, structured prompts without touching a keyboard.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Free on Mac, Windows, and iPhone.
Signal vs. Noise
⚡ OpenAI just gave ChatGPT a 1M-token "do the work for you" brain — here's the catch no one's leading with.
GPT-5.5 launches not as a smarter chatbot but as a true work agent — with a 1-million-token context window and the ability to take multi-step actions across tools. The limitation worth knowing: agentic tasks still require explicit permission prompts at key decision points, which means it doesn't operate as freely as the announcement implies.
⚡ Google quietly shipped a Gemini mode that runs 160 searches per query — and the output format is the surprising part.
Deep Research Max runs up to 160 searches per query, synthesizes results, and outputs a structured, citation-heavy report formatted for presentation. It's not a summary tool — it's closer to a first-draft research function.
⚡ Claude just became a native AWS service — and Anthropic quietly locked in a 5-gigawatt AI power deal — the infrastructure math behind that number is worth a minute.
Claude is now accessible as a native Amazon Web Services offering, removing the integration friction for enterprise AWS customers. Separately, Anthropic confirmed a 5-gigawatt compute commitment — a figure that signals serious long-range infrastructure planning, not incremental scaling.
⚡ Google's AI morning assistant is real — but only if you make the waitlist.
Google CC connects Gmail, Calendar, and Drive to send a "Your Day Ahead" briefing each morning. It's currently waitlisted, U.S. and Canada only, with paid Google users given priority access.
🛠️ Use This Tuesday — Status Brief
🛠️ Turn a plain Q&A list into 4 polished social slides using GPT Image 2 + Canva Magic Layers — under 10 minutes.
Copy this prompt exactly, then paste your Q&A at the bottom:
Turn these questions and answers into 4 vertical social media slides.
Style: dark, cinematic, premium.
Use real-life scenes, not generic AI graphics.
Keep the design consistent across all slides.
Here are the questions and answers:
[paste your Q&A here]Then open Canva, upload images generated, Click edit, use Magic Layers, and resize any text that reads small on mobile. The most common mistake: publishing the first output without checking text size. Small text on mobile will cost you the scroll.
Done looks like 4 visually consistent slides ready to export.
Want to set up the evening briefing routine from Story #2? Reply ROUTINE — first 10 readers get a personal setup walkthrough from my team over email.
From the Community
Last week's winner: ⚡ I don't hold back — 54% (99 of 185 votes).
54% said they ask everything. But 49 of you flagged data trust as a real barrier — and that number is more honest than the headline.
Russ L.: "If I have a detail question about finances or taxes... I don't feel comfortable sharing this as I don't know who has access." David T.: "Treat AI as your staff member, not your boss." t.h.mcintyre: "I even argue back lol"
High workload this week meant I didn't get to reply to some of you who commented after the poll. Working through those after this issue goes out — if you left a comment, it's in the queue.
AI FAIL
💀 Fake AI wolf photo redirected a real police search in South Korea
A man generated an AI image of an escaped zoo wolf and shared it online. Authorities redirected search teams, issued emergency public alerts, and showed the fabricated image at a press briefing before confirming it was AI-generated. He was arrested and faces up to 5 years in prison.
ONE MORE THING
After the L'Aura brand kit was done, I was curious — so I took those same AI-generated images and built a fully working luxury website using Claude Design. Two messages. First attempt. One-click deploy. The homepage is live. If you want the exact step-by-step and want to ask me questions directly, reply CLAUDE DESIGN.
Hit reply and tell me: are you running any kind of AI routine already, or does Story 2 feel like the first time this has been realistic for you?
Dan Rice · AI Signal · Every Tuesday