
Turning Meeting Notes Into a Slide Deck in Five Minutes
Say you close a client call with eleven bullet points in a text file and a 2pm deadline to send a recap deck. There is no design time. The notes need to become something a client can open and skim quickly.
Here is a worked example, start to finish. The notes are sample notes, and the five minutes is my estimate, not a measurement.
The raw material
Meeting notes are messy by nature. Sample notes look like this:
- migration to new payment provider, go-live Sept 15
- three blockers: webhook retries, currency rounding, sandbox creds still pending from their side
- QA plan: two weeks parallel run before cutover
- rollback plan exists, tested once
- open question: who owns Sunday on-call during cutover week
Eleven lines, no structure, half of it fragments. That is the input. Do not clean it up by hand.
Adding structure constraints
The trick is not asking an agent to “make a deck from these notes.” That gets you a wall of text on every slide. Instead, give it constraints up front, in the prompt:
- One idea per slide, max four bullets each
- Title slide, then problem, then timeline, then risks, then next steps
- Close with an explicit ask (here: confirm on-call owner by Friday)
That is the whole brief. I build Slaide, a plain-text slide format. The words are Markdown, and the design lives in a small YAML file called a master. The agent does not drive a slide editor or fight with placeholder boxes. It writes a .slaide file directly, the same way it would write a config file. Slaide has an Agent Skill for this. Per the docs, the agent validates the file and renders it to check its own work.
Rendering and the first look
.slaide renders three ways from the same source: a web deck, a PDF, and an editable .pptx. Run the render and open the web version. This step matters because a text file that looks fine as text can still render as a cramped slide - long words wrap badly, a five-item list overflows the frame. You have to look at the actual deck, not just the source.
In this example, one thing goes wrong: the risks slide has four bullets and the fourth one (“sandbox creds pending from their side”) runs two lines and pushes past the safe margin. Text files do not know how their own words will wrap.
Fixing what the agent got wrong
Three fixes, all small, are typical:
- Shorten the overflowing bullet to “sandbox creds pending (their side)”
- If the agent puts “rollback plan exists, tested once” and “QA plan: two weeks parallel run” on the same slide, split them into two, one per idea, as the constraint says
- Swap the accent color in the theme from the default to a neutral gray, if the client dislikes the default
That last one is a small edit in the theme’s YAML file, not a redesign. The editor is free once you sign in with a Slaide account, so for anything more visual than a color swap, open it there. For three small fixes, editing the source directly is faster.
But let me be honest about something here: a deck with hand-placed decoration or custom animation may not come through the .pptx round trip cleanly. A text-only recap deck does not have that problem. Check the exported file before you build something fancy and hand it to a client in PowerPoint.
Export and send
Render again, export to PDF for the quick look, and to .pptx in case the client wants to add a slide before the go-live review. Both come out of the same source file.
My estimate for the whole run, from raw notes to a deck in the inbox, is about five minutes. Most of that is rereading it once before you hit send.
Why this beats the usual approach
The old way is to open PowerPoint, pick a template, retype the same eleven points into text boxes, fight the auto-layout when a bullet is too long, then export. That takes longer, and the urge to “just polish it a bit” eats more time.
The new way skips the template-picking and the box-fighting because there are no boxes. It is text with a theme, and you can reuse the theme on every recap deck for the same client. Next meeting, same theme, same structure, new notes.
If you run your own coding agent already, Slaide itself costs nothing. The Agent Skill and the MCP server are part of the open core. Your AI provider may still charge you. The only paid part of Slaide is the hosted cloud AI, for people who would rather not run their own agent.
A five-minute workflow does not fix everything. A deck that needs real design work still needs a person.