The AI sidebar in your app only knows what that app knows

A lot of apps have added a little AI chat sidebar in the last couple of years. A box in the corner where you can ask questions instead of digging through the UI yourself.

One person in a recent conversation described giving up on these entirely. Not because the models behind them are bad, but because of where they sit: “the sidebar AI will never have the context that you have inside your own AI.” The sidebar only ever sees what’s inside that one app. Ask it something that requires connecting dots across systems, and it starts, in his words, making “up shit that doesn’t make sense.”

Compare that to a connector-based assistant — using MCP, inside something like Claude. Same person’s example: you want to rewrite a policy, so you ask it to find every place that policy statement currently appears across your connected systems and update it. Or an assistant that already knows your infrastructure, and you ask it to add all your approved apps into a document. Try that with a sidebar chat and you’re back to copy-pasting between tabs.

That’s the real distinction. It’s not that in-app AI chat uses a worse model — though he also suspected they’d gone with the cheapest option. It’s that the sidebar is structurally boxed in. It can only ever answer from the data the app itself holds. A connector-based assistant can see across everything you’ve connected it to, which is exactly what compliance work actually requires: policies, systems and documents that live in more than one place at once.

So the useful question isn’t “does this tool have an AI chat now?” It’s “what can that AI actually see?” A convenient-looking feature with no cross-system context is still, underneath, a dead end.