The last presentation I built took four hours. Maybe forty minutes of that was the deck. The rest was tabs.

A Notion page with the strategy memo. Three Slack threads where the actual numbers got argued out. A HubSpot dashboard. Two Google Docs from people who don’t use Notion. A Salesforce report I had to re-run because the filter reset. By the time I had the inputs in one place, I’d lost the thread on what I was supposed to be saying.

This is the tab-sprawl tax, and it’s the thing nobody talks about when they pitch you AI tools for “making decks faster.” The deck isn’t the bottleneck. The pre-deck is.

Where the time actually goes

If you instrument a deck-building session honestly, the breakdown is something like:

  • 45% gathering inputs across 6–12 tabs
  • 20% reconciling conflicting numbers between those inputs
  • 15% writing the actual prose for each slide
  • 15% formatting, visual polish, chart cleanup
  • 5% the part you can’t predict (the slide your boss wants but didn’t mention)

The AI tools most people reach for collapse the 15% writing slice. They do not touch the 45%. That’s the part a browser agent should own.

What “owning the pre-deck” actually means

A browser agent that lives in the same tabs you do can:

  • Open the recurring HubSpot or Salesforce report, apply the filters you actually want, and dump the result into a markdown table
  • Walk a Notion page, pull every callout block tagged “decision”, and turn them into bullet points
  • Scrape the three Slack threads where the rollout decision got made, summarize the outcome, and flag the dissents
  • Cross-check the number that appears in two different docs and tell you which one is stale

Notice that none of these tasks need a new tool. They need the existing tools to be driven by something that doesn’t get tired clicking through them. That’s the wedge a browser agent like Dassi sits in — it reads your logged-in pages in real time and acts in them, which is why it can do the boring half of preparation that an API-bound automation can’t reach.

I save the prep work as a reusable Skill in the extension and schedule it to run every Friday afternoon. By Monday morning when I sit down to actually build the board update, the inputs are already in a single markdown brief.

Where the deck tool comes in

Once the inputs exist as structured text — bullet points, tables, decisions, numbers — turning them into slides is the part that finally is a writing problem. You can paste the brief into ChatSlide and it will produce a draft deck in the layout you want, with the charts already wired to the numbers in the brief. The forty minutes of actual deck-building shrinks to ten, because the LLM is doing what LLMs are good at: rewording structured input into slide-shaped output.

The split is the point. You do not want one big model trying to also drive your browser. You want a browser agent for the gathering, and a deck tool for the writing. Each one is specialized for what it does. Each one gets boring and reliable in its lane.

The bigger pattern

This generalizes past presentations. Most knowledge work has the same shape: 70% gathering and reconciling, 30% producing the artifact. The artifact tools have gotten dramatically better in the last two years. The gathering tools haven’t, because gathering happens in your browser, across tools you already pay for, behind logins that no API will give you access to. It’s the same reason cloud browser agents can’t see your tabs — the work happens where the credentials already are.

The win isn’t a single AI that does the whole job. The win is splitting the job at the natural seam — browser agent for the part that needs to act in your real, logged-in tabs, and a specialized generator for the artifact at the end.

You’ll know it’s working when “make me a deck on X” stops feeling like a four-hour task and starts feeling like a fifteen-minute one. The browser agent did the three hours that nobody was counting.

The ChatSlide team builds an AI presentation generator that turns structured briefs into board-ready decks. Try it at chatslide.ai.