til there exists an agentic browser called strawberry and it has absolutely blown my mind!

let me clarify one thing - i knew it existed before today, because charles maddock and a part of his team have been coming up on my linkedin feed in one way or another but it’s only today that i gave into the hype and really tried it for myself. and boy, was it impressive!

i went in expecting yet another “Chat GPT style sidebar bolted onto chromium” style experience (Comet, looking at you here) and came out convinced this might be the first browser that actually deserves the title agentic.

what makes it different is that it isn’t just trying to be a chat window with a browser attached. it’s a browser with three things stitched together at the core:

1. a second brain - it remembers what you’ve looked at, what you cared about, and the threads you’ve been pulling on. so when you come back later and ask it something vague, it actually has the real context to answer like a colleague (or even you, yourself) would.
2. skills - reusable little capabilities you (or it) can compose. this is the bit that feels closest to how i’ve been using claude code skills, except now they live inside the thing i already spend 80% of my day in.
3. a genuine agent loop - not “summarise this YT video” parlour (yes Google that’s a dig at you) tricks, but a system that can plan, take actions across tabs, and actually finish a real task. or tasks, rather. and the reason this loop actually runs is that the workflows have clearly been researched to death and ship out of the box, so you’re not building them yourself (you could do if you wanted to) from a blank canvas.

the architecture above (brain + skills + workflows) is what lets all three stitched-together pieces work in concert. and even if you want to argue it isn’t really agentic under the hood - honestly, it doesn’t matter.

it feels like it is, and that’s the bit that counts and led to my “holy shit” moment about an hour ago

the way i understood it, it isn’t simply calling a model on every keystroke - it’s pulling from personalised memory that strawberry is getting by you as the user giving permission to connect the right tools (Slack, e-mail, CRMs etc.), dispatching the right skill, and (by the looks of it) only invoking heavyweight reasoning when it actually needs to; that’s the same architectural insight that makes agents feel snappy 🫰🏼 instead of laggy 🐌, and i’ve never seen it applied to a browser before. applied well, that is.

and honestly? for the kind of research-y, multi-tab, “i need to understand this industry/topic and absorb all I can about it” workflow i do as a PM all day, this might already be replacing perplexity or Google for me.

see, perplexity is brilliant as a search-answer engine, but it lives in its own walled garden. in fact, i haven’t touched it in a few months now.
strawberry, however, is meeting me where i already am - in the browser, on real pages, with real context about me as an individual - and just gets on with what i need getting done.

what more do you want, eh?

i’m sure i’ll come back to this once the honeymoon wears off and write a more sober take but right now, i’m fully in the “where has this been all my life” phase.

hats off to you Charles Maddock and team 🫡

p.s. what i said on linked in still stands - even on a 16 GB M4, i immediately experienced slowdown and dare i say a lag(?), during the onboarding phase itself…might be something you guys want to take a peek at!