Comer is great except for sensitive stuff, even they claim that’s not fully safe yet, so i just preferred sticking to where I can do everything and turn ai off or on as needed. It’s more preference I think
Claude continues its workaround 'Gemini-ness' with Claude in Chrome. I've been in the Beta for a bit, and like you, loving it, still discovering it, still stumbling into, "Damn, I didn't know I could do that!" It has its wonks, but I'm sticking with it at this point as a ready-made AI assistant that's ominpresent and not an afterthought.
Thoughtful and practical take. The emphasis on selective AI use and explicit approval as a security boundary is refreshing amid the “always-on” hype. Strategy over novelty resonates.
Thanks Joel, another excellent post. That 2-question safety test is something that all of us should be doing when we use AI, not just for project builds as well. Thanks so much for sharing.
Love it! The approval step you describe is huge. it turns AI from an unpredictable actor into a true assistant.
Using AI to remove mechanical friction while preserving judgment, security, and trust is exactly how serious leaders should be thinking about this shift.
I've been using Claude in Chrome for a little bit now, but I was missing the bigger picture, the mechanical work. And I agree that seeing the approval puts the unsettled thoughts of "will this actually do what I need it to" at rest.
Wonderful set of use cases !! I have been using this lately, as you know.. even though it is a bit slow but as you describe the use cases above it could be very handy in many cases where grunt work is needed
Liz, I haven't tried it myself, but Brave as a browser, I've heard a lot of good things in terms of security. However, when it comes to using its AI, my concern comes with whether or not you can turn it off when you are on banking websites, credit card websites, or other confidential sites.
MCP is much more direct. I use it on the cloud app on my desktop, but this allows you to give the tool permissions to work on your browser. For example, I have a problem setting up a function for an app connected to Supabase, and I couldn't find where to fix the issue, so I gave Claude access to the desktop so I could go in, click through, fix the issue, open a new window if needed. The nice thing is just like MCP, or at least the way I have MCP set up, it asks before anything is done on your behalf.
I've always wondered about the utility of AI browsers. Seems like they're best fit only for the power users among power users who don't care about anything else (privacy & security included).
“You don’t need to abandon Chrome to access AI capability. Claude in Chrome gives you selective intelligence in the workflow you already trust.”
fantastic takeaway here Joel :) i’ve been really impressed by the plugin so far.
Glad to hear! Yeah the always on ai looking at your browser makes me too nervous
rightfully so lol. even openai says they aren’t solving their indirect prompt injection any time soon!
I know I saw that! man I can’t wait for our collab
i’m back “in office” this week so i’ll have to get the ball rolling again!
I've been using comet for a while now, but you are making me second guess this. Need to stew on all this a bit more
Comer is great except for sensitive stuff, even they claim that’s not fully safe yet, so i just preferred sticking to where I can do everything and turn ai off or on as needed. It’s more preference I think
makes total sense.
I’m obsessed with Comet. It’s where I actually see real time savings with AI.
i've been enjoying it as well.. curious to compare the difference using some of @Joel Salinas suggestions here, and then weigh the pros and cons.
I'm up to my eyes in customs as building lots of founder tools so this is super helpful. Thanks Joel.
I bet, Chris! Glad to hear, I’m glad to answer any questions you get that come up
Thanks
Claude continues its workaround 'Gemini-ness' with Claude in Chrome. I've been in the Beta for a bit, and like you, loving it, still discovering it, still stumbling into, "Damn, I didn't know I could do that!" It has its wonks, but I'm sticking with it at this point as a ready-made AI assistant that's ominpresent and not an afterthought.
Haha I keep having that mindblowing moment once every few days as well!
Thoughtful and practical take. The emphasis on selective AI use and explicit approval as a security boundary is refreshing amid the “always-on” hype. Strategy over novelty resonates.
Strategy over novelty is my motto! Thanks, Petar! Hope you had a great holiday!
Ok, now I need to try it too:) I'm obsessed with Comet, but you made a great case for this plugin. Thank you Joel 🤗
I’m curious on your take!! It may not be for you, or it may :)
Will let you know! 🤗
Thanks Joel, another excellent post. That 2-question safety test is something that all of us should be doing when we use AI, not just for project builds as well. Thanks so much for sharing.
Glad it resonated!
Love it! The approval step you describe is huge. it turns AI from an unpredictable actor into a true assistant.
Using AI to remove mechanical friction while preserving judgment, security, and trust is exactly how serious leaders should be thinking about this shift.
It is great, not just an auto approval.
I've been using Claude in Chrome for a little bit now, but I was missing the bigger picture, the mechanical work. And I agree that seeing the approval puts the unsettled thoughts of "will this actually do what I need it to" at rest.
Glad to hear that, Lee!
I really like the "smart switches" vs "smart home" analogy here.
Thank you, sir!
Claude’s approval step actually stops the AI from going rogue on you, which is what most of these “AI browsers” seem to gloss over.
Yes!! 🙌 that’s why it’s my preferred avenue
Wonderful set of use cases !! I have been using this lately, as you know.. even though it is a bit slow but as you describe the use cases above it could be very handy in many cases where grunt work is needed
I’m not very tech savvy but I use Brave as my browser and it has Leo AI. Your thoughts on Brave & Leo?
Liz, I haven't tried it myself, but Brave as a browser, I've heard a lot of good things in terms of security. However, when it comes to using its AI, my concern comes with whether or not you can turn it off when you are on banking websites, credit card websites, or other confidential sites.
Thank you, Joel! I’m not sure…
I’ll do some more digging for you :)
From what I can see, no red flags with Leo AI :)
I’m not a fan of AI Browsers because of security. Will be trying out the Claude extension though.
agreed, same boat! What I like here is I turn it on and off as needed
So it accesses your online software when opening it and granting access to Claude? Wonder how that stands up against MCP?
MCP is much more direct. I use it on the cloud app on my desktop, but this allows you to give the tool permissions to work on your browser. For example, I have a problem setting up a function for an app connected to Supabase, and I couldn't find where to fix the issue, so I gave Claude access to the desktop so I could go in, click through, fix the issue, open a new window if needed. The nice thing is just like MCP, or at least the way I have MCP set up, it asks before anything is done on your behalf.
Thanks Joel, I’ll have to try it myself, I am stuck to Safari though 🫣
I get it! Apple may have something by 2035 ;)
😫
I've always wondered about the utility of AI browsers. Seems like they're best fit only for the power users among power users who don't care about anything else (privacy & security included).
Yeah, that's exactly my hesitation with Comet or Atlas when the creators themselves tell you that they cannot guarantee your security.