Everyone Added a Claude Skill for Decks. They're Still Unhappy With the Output.

Over the past few months, nearly every team I talk to tells me the same story. They heard about Claude skills, got excited, and spent a weekend writing one for presentations. They pointed it at their brand guidelines, described their slide layouts, maybe even pasted in examples of their best decks.

And then they used it for a real deck — a QBR, a sales pitch, a board update — and ended up rebuilding most of it by hand.

The skill worked. The output is still disappointing. After hearing this a dozen times, I think it's worth explaining why, because the reason matters for anyone trying to solve this problem.

What a skill actually is

A skill is a set of instructions. It's a very good mechanism: you write down how something should be done, and Claude follows those instructions when the task comes up. For presentations, that usually means rules like "use our navy for headlines," "title slides have the logo bottom-left," "never more than three bullets per slide."

Instructions are genuinely useful. They catch the obvious failures. The deck comes back with roughly your colors, roughly your structure, roughly your tone.

Roughly is the problem.

Instructions describe a brand. They don't contain it.

Your brand guidelines say "use navy for headlines." Your best decks say something much richer: when the headline breaks to two lines, when the number gets its own slide, when the chart gets a callout instead of a caption, when a slide is allowed to be quiet. None of that is written down anywhere. It lives inside the decks your best people made, and inside the judgment of the people who made them.

A skill file can hold the rules you can articulate. It cannot hold the judgment you never wrote down. So the generated deck follows every rule you gave it and still feels like it came from a company that merely read your brand guidelines — because it did.

This is the gap we wrote about in Brand as Software: a guideline can describe the brand, but it cannot participate in the work.

The second problem: text-first output

Even when the content is right, skills produce presentations as structured text that gets rendered into slides. That pipeline has a ceiling. Real decks are designed objects — spacing, hierarchy, image treatment, the relationship between elements on the slide. When layout is an afterthought applied to text, you get the AI fingerprint: predictable grids, default-feeling typography, slides that are technically correct and visually anonymous.

Your audience can't name what's wrong. They just know it doesn't look like you.

The third problem: every deck starts from zero

A skill is static. It doesn't learn that last quarter's board deck landed well, or that your sales team finally cracked the competitive slide, or that marketing updated the photography style in March. Every generation starts from the same frozen instructions. The system never gets better, and the thirty percent you fix by hand every time never compounds into anything.

What actually works

The teams who are happy with AI-generated decks did something different: they stopped writing instructions about their brand and started building a system from the brand itself.

That's what we built Decky to do. Instead of a skill file describing your decks, Decky learns directly from your best presentations, your templates, your website, and your brand assets — the artifacts that already contain the judgment you never wrote down. Then:

• Every deck starts from your recognizable center, not from a generic model's idea of a business presentation. A QBR and a product launch get different energy, and both still look like you.
• The output is a real, editable file — Google Slides or PowerPoint — not a text rendering you have to rebuild. The seventy-percent problem goes away because the last thirty percent is editing, not reconstruction.
• The system compounds. When your team makes a great deck, it becomes part of what Decky learns from. The next deck starts from a higher floor.

Skills are a good feature, and for a lot of tasks instructions are enough. Presentations just aren't one of those tasks — because a deck isn't a document, it's your brand performing in front of an audience. That takes a system, not a skill.

If you've written the skill and lived the disappointment, try Decky on your next real deck. The difference shows up on slide one.