Gamma rebuilds its presentation tool to take the 'AI smell' out of generated decks

Gamma 5 is built for the person presenting to a room that can spot a machine-made deck. "They all have a distinctive tell," Jon Noronha, Gamma's co-founder and chief product officer, said of AI-generated presentations. "There's been more and more discussion by people saying they wish they could make AI-generated content where it's not clear they made it with AI." ^[4]

Gamma is pitching taste, though most of what it has shipped is control over the canvas plus checks that catch layout mistakes ^[8]. Noronha tied the rebuild to better models. "These agentic models have gotten much more capable," he said. "Because of that, we can give them the full power to write custom code, draw shapes and arrange elements on a canvas. They do a really good job now." ^[7] Before any slide is drawn, users can work through an outline with the agent, settling the argument, the audience and the visual approach ^[9]. For images, Gamma leans on illustrations. Noronha said they avoid the distorted anatomy, inconsistent lighting and implausible details that photorealistic AI images often show ^[10].

The speed claim is about formatting. Noronha said some users have cut deck-building from eight to 10 hours down to about 30 minutes, largely because they format less ^[15]. That is a 16- to 20-fold reduction ^[20], and it rests on user reports he relayed. He said improvements that fit a deck to its audience are less measurable but no less important ^[16].

A pilot usually starts with the decks a team already has. Gamma imports PowerPoint files and PDFs for editing, and Noronha said an internal benchmark of import visual accuracy rose from roughly 30% early in development to over 97% ^[13].

I'd start a pilot with a team whose decks go to customers and draw on call notes, since that is where both the design work and the connectors apply. The tradeoff lands on the same team. Its Slack channels and call transcripts are the likeliest places for sensitive customer material, and those are the sources the connectors reach ^[11].

The rest of the decision sorts on two axes: whether the deck leaves the company, and whether it depends on data held in Slack or meeting transcripts. External decks built on company data get the full Gamma 5 case, along with the full exposure. When a deck goes outside without internal data, the test is whether a team's own imported decks keep their layout and branding. Inside the company, a data-heavy deck gains mostly formatting time, and I'd expect the AI look to matter less in front of colleagues. Internal decks with no company data are where the rebuild adds least. In every box, the measure a month in is the one Gamma applies to its own models: whether people keep editing and sharing what they make, and whether they would pay ^[18].