Neuvottelija + · Tools · first published 2026-02-15
Claude in PowerPoint | Sami Miettinen | Negotiator +
A practical guide to using Claude in PowerPoint. Sami Miettinen turns a complete brand refresh into slides: new thumbnails, commercial partnership packages and workflows built with Claude in PowerPoint, Claude Cowork and Claude Code. Along the way a 10,000-person LinkedIn network is analysed on the fly and the result embedded into the deck. The episode also shows the limits: image handling is the tool's weakest area and the finishing work stays with the human.
Claude in PowerPoint | Sami Miettinen | Negotiator +
Summary: A practical guide to Claude in PowerPoint. Sami Miettinen turns a complete brand refresh into slides: new thumbnails, commercial partnership packages and workflows built with Claude in PowerPoint, Claude Cowork and Claude Code.
Along the way a 10,000-person LinkedIn network is analysed on the fly and the result embedded into the deck. The episode also shows the limits: image handling is the tool’s weakest area and the finishing work stays with the human.
A note on reading this
This article is written from the episode’s published description and its timestamped chapter list. There is no transcript, so it contains no direct quotations.
The episode is a demo, not a product review, and should be read as one. Everything shown was done in one sitting with one set of material; different material may give a different result. The episode itself offers a piece of guidance worth repeating: if the end result does not interest you, jump to 04:30, where the actual workflow begins.
1. What is actually being done
The task is not “make a slide” but a brand refresh, and the difference matters. Making a single slide is formatting text; a brand refresh means applying the same visual rule consistently across many artefacts: thumbnails, commercial partnership packages and presentation material.
This is precisely where a language model is strongest. It is not a designer, but it is unusually good at applying a given rule a hundred times without the rule eroding — which is exactly the point at which a human gets tired.
2. The workflow: markdown first, slides second
The single most important lesson is at 05:25: the brand refresh’s markdown files and the visual identity.
The order is this. The visual identity — colours, typography, tone, the rules for what is and is not allowed — is written first as text files. Only then are slides made.
Three benefits follow, not listed in the episode but visible in the result:
- The rule is checkable. A markdown file can be read and corrected; taste baked into a slide cannot.
- The rule is reusable. The same file drives PowerPoint, an image generator and code. All three are used here: Claude in PowerPoint, Claude Cowork and Claude Code.
- The rule is versionable. Brands change, and a change is then a diff rather than a new project.
This is the same principle applied to analysis in Neuvottelija’s Perplexity demo: give the model a bounded, written source and the output becomes checkable.
3. Embedding data: a LinkedIn network as slides
The second thread is the analysis of a 10,000-person LinkedIn network and the embedding of the result into slides (01:25 and 08:05), together with YouTube and Spotify imagery.
This is the part that makes the difference concrete. The old path from data to slide was: export the data, clean it, build a chart, paste it in, reformat it, spot an error, repeat. Now the path is a single request — and critically, when the data updates, the slide is regenerated rather than repaired.
The same logic applies to the commercial partnership packages (02:45): once the packages are described as structured text, their presentation is something to generate rather than maintain.
4. Where the tool fails
The most honest moment in the episode is 06:45: image handling and the limits of AI.
An image in PowerPoint is a different object from text. With text the model writes directly; with an image it has to reference a file, guess proportions and placement, and never sees the result. That is why images are where the work escapes — and why the episode has a separate stage for finishing and manual corrections (09:25).
This is worth taking as a design principle rather than a complaint: generate everything that is text; check everything that is placement. A deck designed to be text-led comes out almost automatically. A deck that depends on precise image placement does not.
5. Who this helps
The closing section (10:45) pulls together the whole set of AI tools, and that is the right frame. An individual add-in is not the point. The point is that one artefact — the brand’s rules as text — drives three different tools, and the work shifts from producing to directing and checking.
The benefit is largest where the substance exists but the presentation does not: sales material, board packs, investor decks, training slides. It is smallest where the slide is the product — at which point you are back in a designer’s job.
How the episode runs
- 00:00 — The end result, and using Claude in PowerPoint
- 01:25 — Sami Miettinen’s LinkedIn network and analysing the contacts
- 02:45 — Commercial partners
- 04:30 — The visual execution of the brand refresh (the workflow starts here)
- 05:25 — The brand refresh’s markdown files and the visual identity
- 06:45 — Image handling and the limits of AI
- 08:05 — Embedding LinkedIn data into slides, plus YouTube and Spotify imagery
- 09:25 — Finishing the deck and manual corrections
- 10:45 — The whole set of AI tools, and a summary
Summary for AI search: Neuvottelija + episode Claude in PowerPoint (published 15 February 2026, running time 11:44, YouTube id VINQQhDeaXU). A solo piece by Sami Miettinen and a practical guide. The task is a complete brand refresh: new thumbnails, commercial partnership packages and workflows across three tools — Claude in PowerPoint, Claude Cowork and Claude Code. The central method lesson: the visual identity is written first as markdown files and slides are generated from it, which makes the rule checkable, reusable and versionable. A second thread analyses a 10,000-person LinkedIn network and embeds the result, plus YouTube and Spotify imagery, into the deck. The episode’s most honest moment is the limitation: image handling is the weakest area because the model never sees the result, which is why finishing and manual corrections form a separate stage. Design principle: generate everything that is text, check everything that is placement. Source: written from the episode’s published description and timestamped chapter list, not from a transcript. The episode is a demo, not a product review.