Meep #2
Discover more from Kamil’s newsletter
See all Meep posts here.
It’s been six days since my inaugural Meep post. In that time, I mostly worked on my other projects, including Direction App (check it out!). Now, I’m refocusing here.
First, a side note
Some people prefer to focus on a single thing for weeks or months at a time. I disagree. I think it’s beneficial to do a work sprint on something, but then put it aside for a while, especially when you’re trying to innovate.
The time when you put it aside is not wasted, because:
Your subconscious keeps turning it around and processing it in the background
You can still casually talk about it with people and get new ideas that improve your thinking.
If you were sealed in your basement, you wouldn’t have got these ideas.
I found & researched a competitor
I came across compose.ai. One of their features is essentially what I’ve envisioned for Meep (the grey autocomplete). It also looks like they have other features that are less unique, with seemingly a lot more competitors (rephrase, generate text etc.).
Out of curiosity, I installed their extension and did some brief testing. I decided to only look at the autocomplete part.
I also skimmed their chrome web store reviews. Here is what I found from both the reviews and my testing:
The good:
The frontend layer, at least in gmail, seems to work fairly seamlessly (suggestions get displayed, I can accept them with Tab).
Suggestions show up quickly.
It knows my name, even though I did no manual setup.
The bad:
Suggestions are often nonsensical - see gif 1, why would it repeat my name?
Chunking is weird - first it suggests a few words but with my last name cut off, later it suggests maybe 1 or 2 words only.
Doesn’t take context into account - in gif 2, I started asking for a refund, so I would like some refund-related suggestions, rather than defaulting to a generic “I would… like to send you… a message asking for… your help”. This isn’t helpful.
It only works on a handful of websites. Maybe the frontend bit is harder than I thought? It could be that I also need to identify a small set to start with.
Some people report they make Gmail unresponsive - I also observed a slow down, not sure if it was because I was recording my screen meanwhile. I hope these guys aren’t using the extension to access my computer’s power to mine bitcoin in the background 😅
Doing some more research, it looks like compose.ai was founded in 2020, so it’s possible their AI predictions use pre-GPT 3 models, which were substantially weaker. Despite these issues, they still managed to get 400k downloads, which actually makes me quite excited - the UI and value proposition seems attractive to people; how successful could your product be if your suggestions were significantly higher quality?
The research insights got me excited!
Here’s why:
This is evidence that people do understand and are willing to test this kind of product.
From my experience working with GPT, I feel strongly that I can come up with better quality suggestions (whether I can provide them quickly and cheaply enough is TBD).
What am I doing next?
Over the next few days, my plan is to:
Come up with a name and domain.
Make a simple web page that presents the product, and has a waitlist where interested prospective users could sign up.
Brainstorm: who could benefit the most from using it? Is it customer support agents? Salespeople? Lawyers? Where do they usually do their typing?
After my brainstorm, figure out where I can reach some of them, and point them to my web page with the waitlist to see if they get excited, or if they have any other feedback.