Appreciate how practical this is. Iterating on prompts and combining AI vision/voice with tools like Gemini Canvas really turns abstract ideas into actionable insights in minutes.
I like the concept of this newsletter! Some notes:
-The "10 new ideas for automation" prompt yielded interesting results—it suggested 10 approaches I'd actually tried before, which showed it understood my needs. However, I'd already discovered that automating these particular tasks with AI either unnecessarily messified the process, created more work than doing them manually, or required the same effort without delivering any real benefits. Ex. it recommended "AI generates base visuals (icons, 3D data flows, IFTTT-style screens) → you refine." which has honestly been something I've tried to make work for the last 2 years and have not been able to do well (it always looks too much like AI which doesn't work for an enterprise setting)
-I love the idea of Gemini infographics but have not been impressed with the output (it uses the same style every time, right?). I am excited because I see a really strong AI application for diagrams. I actually have been finding Microsoft's "SmartArt" feature on Powerpoint to be really useful, and I feel like if that system can incorporate AI, it could make a Napkin-style feature set native in Powerpoint
On the 10 ideas for automation, I think it may have surfaced the same ones because it pulled from your memory. You could push it a bit harder to dig for novel ones, though of course it can’t fully understand how you work, so it’s limited in spotting where friction really is.
What kind of visuals are you trying to produce? Do you have a standard you usually aim for? very curious to see an example.
And yes, agree on Gemini, it tends to fall back on the same format most of the time. At least for now.
Yeah, that’s a good point! For images—I was really hoping to make AI work for building McKinsey-style graphics for my decks at work. I tried using a mix of Ideogram, ChatGPT, and Nano Banana to brainstorm the best way to show a process. But everything they produced looked kind of odd and half-baked.
One workaround that did help was using AI to generate Google Image queries. I had it search across consulting sites like McKinsey, Accenture, EY with a keyword (like “Customer Research”), and that actually surfaced some useful examples. Once I found a design I liked, I’d try to recreate the shapes/diagrams with the AI tools and then add the text myself. The results still looked too “AI-ish,” though, so I usually ended up just building the shapes in PowerPoint from scratch.
Where AI has been valuable is in the early stages—helping me come up with the initial concept, like what arrows to use, where things should flow, and what text might sit at each step. Then I can take that scaffold and finish it myself.
Another use case was a logo for my Substack. I actually liked a lot of what Ideogram came up with, but couldn't change the color. I tried Vectorizer to turn it into a vector but always in those cases have to end up creating the vectors myself (unless it's super super simple). I think there's been some research on AI that can directly create vectors so I'm excited to see how that progresses. Would likely be more similar to code generation than diffusion
For the logo, have you tried changing the color with Nano Banana? I’ve found it works well for that kind of tweak.
Your process actually sounds like a great way to use AI for images. In my experiments, giving the model a clear example does push it much closer to the style I want, especially when you have a clear vision of what it should look like like you do.
That’s why I’m intrigued where it still doesn’t work & still curious what “AI-ish” looks like in this context since we’re talking diagrams. Could you DM me one example of what you were trying to build, along with the McKinsey slide you wanted to mirror + what AI generated? I’d love to try recreating it.
I just DMed you! And great idea on Nano Banana, that could actually save me some headache as until now I have not been able to say "Change this #hexcode to this #hexcode" with any real consistent results
This newsletter is packed with actionable ideas...I especially love the prompts for reflection and workflow optimization.
Thank you, Sharyph, glad they're useful!
Appreciate how practical this is. Iterating on prompts and combining AI vision/voice with tools like Gemini Canvas really turns abstract ideas into actionable insights in minutes.
Glad it’s useful!
Thanks for sharing this Daria.
I like the concept of this newsletter! Some notes:
-The "10 new ideas for automation" prompt yielded interesting results—it suggested 10 approaches I'd actually tried before, which showed it understood my needs. However, I'd already discovered that automating these particular tasks with AI either unnecessarily messified the process, created more work than doing them manually, or required the same effort without delivering any real benefits. Ex. it recommended "AI generates base visuals (icons, 3D data flows, IFTTT-style screens) → you refine." which has honestly been something I've tried to make work for the last 2 years and have not been able to do well (it always looks too much like AI which doesn't work for an enterprise setting)
-I love the idea of Gemini infographics but have not been impressed with the output (it uses the same style every time, right?). I am excited because I see a really strong AI application for diagrams. I actually have been finding Microsoft's "SmartArt" feature on Powerpoint to be really useful, and I feel like if that system can incorporate AI, it could make a Napkin-style feature set native in Powerpoint
Thanks for sharing!
Hey Tyler, thanks so much for sharing!
On the 10 ideas for automation, I think it may have surfaced the same ones because it pulled from your memory. You could push it a bit harder to dig for novel ones, though of course it can’t fully understand how you work, so it’s limited in spotting where friction really is.
What kind of visuals are you trying to produce? Do you have a standard you usually aim for? very curious to see an example.
And yes, agree on Gemini, it tends to fall back on the same format most of the time. At least for now.
Yeah, that’s a good point! For images—I was really hoping to make AI work for building McKinsey-style graphics for my decks at work. I tried using a mix of Ideogram, ChatGPT, and Nano Banana to brainstorm the best way to show a process. But everything they produced looked kind of odd and half-baked.
One workaround that did help was using AI to generate Google Image queries. I had it search across consulting sites like McKinsey, Accenture, EY with a keyword (like “Customer Research”), and that actually surfaced some useful examples. Once I found a design I liked, I’d try to recreate the shapes/diagrams with the AI tools and then add the text myself. The results still looked too “AI-ish,” though, so I usually ended up just building the shapes in PowerPoint from scratch.
Where AI has been valuable is in the early stages—helping me come up with the initial concept, like what arrows to use, where things should flow, and what text might sit at each step. Then I can take that scaffold and finish it myself.
Another use case was a logo for my Substack. I actually liked a lot of what Ideogram came up with, but couldn't change the color. I tried Vectorizer to turn it into a vector but always in those cases have to end up creating the vectors myself (unless it's super super simple). I think there's been some research on AI that can directly create vectors so I'm excited to see how that progresses. Would likely be more similar to code generation than diffusion
For the logo, have you tried changing the color with Nano Banana? I’ve found it works well for that kind of tweak.
Your process actually sounds like a great way to use AI for images. In my experiments, giving the model a clear example does push it much closer to the style I want, especially when you have a clear vision of what it should look like like you do.
That’s why I’m intrigued where it still doesn’t work & still curious what “AI-ish” looks like in this context since we’re talking diagrams. Could you DM me one example of what you were trying to build, along with the McKinsey slide you wanted to mirror + what AI generated? I’d love to try recreating it.
I just DMed you! And great idea on Nano Banana, that could actually save me some headache as until now I have not been able to say "Change this #hexcode to this #hexcode" with any real consistent results
Thanks for the practical tips! Especially using the phone's camera to let AI interact with my surroundings.
I love that feature too! use it even when I buy new home appliances or other such things to get help setting them up