Someone hit the Hacker News front page this week with “The AI Productivity Stack I Use in 2026,” and I read the whole thing waiting for a part that never showed up. It’s a good list. Notion AI for notes, Granola for meeting transcripts, Perplexity instead of Google, two agent wrappers, and the now-obligatory vector database nobody asked for. Every tool on it is real. Every tool is fine. And every single one of them lives somewhere your actual work does not.

That’s the pattern across all of them. Scroll the trending list and you’ll see three more posts in the same shape sitting at #7, #8, #10. They’re all stacks of cloud SaaS that you wire together by hand, ferrying text from one tab to the next.

The stack is mostly tabs you pay for

Here’s what an “AI productivity stack” usually means when you unpack it: six subscriptions, each one a separate login, each one trained to answer questions about whatever you remember to paste into it. You become the integration layer. You’re the API. You read the page, you summarize it for the tool, the tool gives you something back, and you carry that something to the next tool by hand.

The marketing calls this a workflow. It’s a relay race where you’re the only runner.

And none of these tools can see the thing you’re looking at right now. Your Salesforce pipeline, the half-filled expense form, the Gmail thread with the client who’s annoyed, the Notion doc your teammate just shared. The model behind Perplexity is genuinely brilliant. It also has no idea you have nineteen tabs open and a deadline.

Why does the layer where work happens get skipped?

Because it’s hard to monetize a layer you don’t own.

A SaaS company can sell you a polished chat box. What it cannot easily sell you is access to the place you already are, logged in, with all your sessions live and your cookies warm. That place is your browser, and it’s the single richest pile of context any AI could ask for, which is probably why the listicles route around it so neatly.

Think about what a browser tab knows that a cloud tool never will. It knows you’re authenticated into your company’s admin panel. It knows the exact state of the form you abandoned. It can read the dashboard that has no export button and no public API. When an AI agent runs inside that tab, it inherits all of that for free, with no OAuth dance and no “connect your account” screen that wants thirty permissions you’ll regret granting. We wrote more about that authentication gap in why cloud browser agents can’t see your tabs, and it’s the whole game.

This is the layer the stacks forget. Not another box to paste into. An agent that operates where you already operate.

What it looks like when the agent is in the tab

I switched a chunk of my own “stack” to Dassi, which is a Chrome extension that runs in the side panel and sees the page I see. The difference isn’t subtle, and it took me a few days to stop reaching for the old copy-paste reflex out of habit, because muscle memory built over years of feeding context into chat windows does not unlearn itself in an afternoon.

Now when I’m staring at a vendor’s reporting dashboard that locks CSV export behind a higher tier, I don’t screenshot it and describe the table to ChatGPT. I ask the agent in the panel to read the rows off the page and hand me the data. When a client email needs a reply, the draft happens against the actual thread, not against a paragraph I retyped. The model is the same model the listicle people are using. The plumbing is completely different.

So the productivity gain doesn’t come from a smarter LLM. It comes from deleting the relay race. No tab-switching, no re-pasting, no being the human glue between five disconnected services. That’s the productivity tax those stacks quietly charge, and almost nobody adds it up. (We did add it up, roughly, in the context tax, and it’s worse than you’d guess.)

The missing layer

Every productivity stack post is really a confession that AI still can’t reach into the place you work. So you bridge the gap with your own hands and your own attention, one paste at a time.

Dassi is free, it runs on your ChatGPT login or your own API key, and the data stays in the browser. Build your stack however you like. Just stop forgetting the tab you’re already in.