Figma is extending its AI agent into FigJam and Figma Slides, announced at Config 2026 and rolling out through an early-access waitlist, bringing the same generate-and-assist capability the design tool has to the whiteboard and the deck. Both already have their own separate AI features, but the agent itself is the new arrival. The reflexive reaction to any AI-in-a-tool announcement is either uncritical excitement or reflexive dismissal, and both are lazy. The useful question is narrower: for the actual work people do in FigJam and Slides, where does an agent genuinely help, and where is it a novelty that produces output you'll throw away. This piece frames what these agents do, sorts the likely useful uses from the likely noise, and gives you a way to judge it on your own work. Because the agent is still rolling out to these surfaces, the exact capabilities will keep firming up, but the judgment framework holds regardless.
What the agents plausibly do
Figma is extending its AI agent into FigJam and Figma Slides, announced at Config 2026 and rolling out through an early-access waitlist, bringing the same generate-and-assist capability the design tool has to the whiteboard and the deck.
In broad terms, an agent in FigJam can generate and organize whiteboard content, boards, diagrams, groupings, summaries, from a prompt, and an agent in Slides can generate and shape presentation content, drafting slides, structuring a deck, producing supporting material. The pattern matches AI assistance elsewhere: it produces a starting point and handles some of the mechanical shaping, fast.
The honest lens for judging them is the same one that applies to all AI-in-tools: an agent is genuinely useful when it removes real mechanical work or gets you past a blank start, and it's noise when it produces plausible-looking output that needs so much fixing you'd have been faster doing it yourself, or when it replaces thinking that shouldn't be replaced.
Where a FigJam agent genuinely helps
FigJam is a thinking and collaboration tool, which sets up a clear line between help and noise.
Genuinely useful: getting past the blank board and handling the mechanical setup. Generating a starting structure for a retro, a brainstorm, or a workshop, so you're editing a scaffold instead of staring at nothing, is real help, because the setup is mechanical and the blank start is a genuine friction. Likewise, organizing and summarizing a messy board after a session, clustering sticky notes, pulling out themes, drafting a summary of what a chaotic workshop produced, is exactly the tedious post-work an agent can do well and people usually skip.
Likely noise: generating the actual thinking. If the agent produces the ideas in a brainstorm or the content of the collaboration, it's replacing the point of the exercise. A retro's value is the team's honest reflection, not an AI's guess at it; a brainstorm's value is the group's ideas, not generated filler. An agent that does the thinking for a thinking tool hands you plausible output that's missing the only thing that mattered, which was the humans doing the work.
The line in FigJam: use the agent for setup and cleanup, the mechanical bookends, and keep the humans doing the thinking in the middle.
Where a Slides agent genuinely helps
Slides is a communication tool, and the line runs slightly differently.
Genuinely useful: drafting structure and handling formatting. Turning your rough content into a first-draft deck structure, so you refine rather than build from scratch, is real help, and handling the tedious formatting and layout consistency across slides is exactly the mechanical work that eats presentation time. If you have the substance and need it shaped into slides, an agent can do the shaping.
Likely noise: generating a deck from a thin prompt. Ask the agent to "make a presentation about X" with little input and you get a generic, plausible-looking deck that says nothing specific, the presentation equivalent of the same-looking AI output that shows up everywhere. A deck's value is the specific argument and the specific evidence, which a thin prompt can't supply, so the agent fills the gap with generic content that looks like a presentation and persuades no one. This matters more for high-stakes decks, where a generic AI-generated presentation actively undercuts you, a risk worth weighing against the craft of presenting design work to stakeholders.
The line in Slides: use the agent to structure and format content you have, not to generate content you don't.
The useful-vs-noise test you can apply
Rather than a verdict that ages the moment the features update, here's the test that stays true: an AI agent in any tool is useful in proportion to how mechanical the task is and how much real input you bring, and it's noise in proportion to how much judgment or specific substance the task requires that only you have.
Setup, cleanup, formatting, and structuring-what-you-have sit on the useful side, because they're mechanical and you supply the substance. Generating the thinking, the ideas, or the specific argument sits on the noise side, because that's the part that needed you. Apply that test to any FigJam or Slides agent feature and you'll know whether to use it before you waste time finding out, which is the same discipline of knowing where AI output is trustworthy that runs through designing AI features users trust.
A note for teams: don't let generated decks become the default
There's a team-level risk worth flagging. Once generating a deck is one prompt away, the temptation is to make AI-generated decks the default for everything, including the high-stakes presentations where a specific, well-argued case actually matters. The danger isn't that any single generated deck is bad; it's that a culture of generating decks quietly lowers the bar, so important presentations that deserved real thought get the same generic treatment as a throwaway internal update. Keep the distinction explicit: use the agent freely for low-stakes, internal, or first-draft decks, and insist on genuine human argument for the presentations that carry weight. The tool makes generating easy; deciding which presentations deserve more than easy is a judgment a team has to protect on purpose.
Start Monday
Next time you set up a FigJam board or start a deck, try the agent for exactly one thing, the mechanical part: generating the board structure, or formatting a deck you've drafted, and do the thinking yourself. Notice where it genuinely saved time and where its output was generic filler you replaced. One honest test on real work tells you where these agents belong on your useful-versus-noise line, which is worth more than any feature announcement.
The FigJam and Slides agents are neither the productivity revolution nor the pointless gimmick the two reflexive reactions claim. They're genuinely useful for the mechanical bookends of the work, setup, cleanup, formatting, structuring, and noise when asked to supply the thinking or the specific substance that was the point. Use them for the former, keep the latter human, and judge every new feature by how much of your own input it requires.