Slidewriting deserved a real AI editor. So I built one.

The years in slides

I spent my career in consulting, corporate strategy, and finance — fields where the unit of work is a deck. Strategy reviews, QBRs, board updates, diligence readouts: I have written thousands of slides and rewritten many times more. Later I ran internal AI projects — enablement tools built to make exactly that kind of work faster. Testing every AI presentation tool on real workflows was part of the job. That is where the frustration started.

What every AI tool got wrong

Every tool I tested made the same three mistakes.

They optimized for looking impressive, not for being right. Dramatic gradients on a first draft nobody would ever present, slide after slide of filler text. In consulting and finance a deck lives or dies on whether the table reads clean and the chart says one true thing. The "wow" is worthless if the content is not meaningful.

They could not produce a real chart or table. What looked like a chart was usually a stack of shapes and text boxes — convincing in a screenshot, useless the moment the numbers change. A chart that is not linked to data is not a chart; it is a picture of one.

They did not understand the job. Slidewriting is not "generate a deck from a prompt." It is the Monday-morning refresh of a recurring report. It is rebuilding a partner deck onto your client's template. It is redacting forty slides before an external send. No generator understood any of that.

Slidewriting is a revision job

Watch where deck hours actually go and the pattern is hard to miss: most of the work is revision, not creation. Generation happens once; revision happens every week, on the same decks. The QBR gets new numbers. The pipeline review gets new takeaways. The template evolves and a whole library needs restyling. AI tools raced to own the first draft — and left the recurring majority of the work untouched.

The architecture bet

So I built the tool around the work instead of the demo. SEWPOINT is an editing engine with three commitments:

It listens in plain language — any language you work in. "Refresh every chart with Q3 numbers." "Translate the deck to Spanish." "Rebuild this on our template." One instruction can carry a whole deck's worth of intent.

It compiles instructions into native slide operations. Not images, not overlays, not a parallel canvas — a plan of concrete changes to real slide objects: text boxes, tables, data-linked charts, layouts. It learns your template as a design system — the grid, the fonts, the colors, how your charts and tables should look — not just a master slide to copy from.

It executes and verifies in your live file. The deck never leaves your Google Drive. Every change lands as a native, editable object, checked before the run finishes.

There is a reason the native part is a commitment and not a preference. HTML-canvas tools produce genuinely beautiful pages — but every edit afterward goes back through the AI: another prompt, another wait, another token bill, for a change you could have made by hand in two seconds. A deck is a working file; it gets touched twenty times before it ships. That is why traditional slide tools still win where work actually happens: dragging a box, retyping a number, nudging a label — instant, free, precise. The division of labor only works one way: AI for the heavy, deck-wide lifting; your hands for the finishing touches. And that requires the output to stay a normal, native deck.

What you are seeing is a first attempt — the architecture is the point

I want to be honest about what SEWPOINT is. This product is the first time the architecture has been pointed at a traditional presentation tool, and it was built so it would not be the last. The engine sits deliberately apart from Google Slides itself: most of the hard part — figuring out what you actually meant, planning the edit, writing it back as real native objects — is the same problem in PowerPoint or Keynote. Only the last mile changes. I started with Google Slides because that is where my own pain lived. If the idea holds, it will not stop there.

Principles

Where this goes

The near-term roadmap is simple: more depth on Google Slides, then the same engine on PowerPoint — and, as platforms allow, Keynote. An API, so recurring refreshes can run without anyone opening a chat. The long-term picture is one editing layer that speaks every slide platform's native language — so teams keep the tools they trust, and finally get the AI editor those tools never had.

If any of this resonates — as a user, a team, or someone building in the same direction — I would like to hear from you.

— the builder behind SEWPOINT
admin@sewpoint.app · LinkedIn