How to get the most out of the agent

Poolday is a superintelligent AI media production agent — image, audio, video, decks. It can take your existing footage and assets, or generate net new from all the generative AI models out there.

It’s not a chatbot. You’re giving an agent instructions on how to accomplish a task for you. Be directive and specific about the things you care about.

Every word in your prompt matters. The agent treats what you write as a rule, so be really clear about your constraints.

If you want to create something ambitious, your brief needs to be just as ambitious.

You can approach a project however you like, but this path consistently works. It’s built around one idea: never sit idle while the agent works. While it builds, you do the creative work only you can do:

A sample production, start to finish

Here’s what the workflow looks like on a real brief. Say you’re launching an analytics dashboard and you’ve already built brand:acme in phase 01. Your opening brief:

Input: brand:acme, the attached screen recordings (dashboard.mp4,
mobile-flow.mp4), and our launch tagline: “Ship reports in minutes,
not days.”
Desired output: a launch video for the new analytics dashboard.
Arc: cold open on the pain (a report deadline), reveal the dashboard
as the turn, close on the tagline and CTA. First two seconds: a cursor
hesitating over a blank spreadsheet — the viewer should feel the
dread of a report due tomorrow.
Audience: heads of operations at mid-size companies. They’ve seen a
hundred SaaS ads; they trust product proof, not adjectives.
Hard constraints: real product footage only in the demo section — no
generated UI. Tagline verbatim on the end card. Logo clear-space per
the brand kit.
Platform / format: LinkedIn, 16:9.
Length: around 45 seconds.

The demo footage and tagline are fixed — you choose pacing, music,
and transitions. Give me three music directions to choose from
before you build.

The agent proposes three music directions; you pick one and it builds. An hour later the first cut is back. You watch it once as a viewer, once as a director, and answer with mechanisms, not vibes:

Strong middle. Two changes:
1. The cold open lingers — cut from the spreadsheet to the dashboard
   reveal at 0:04 instead of 0:07, hard cut, no dissolve.
2. The end swells too late — bring the music peak forward so it lands
   exactly on the tagline, and hold the logo a beat past where it
   feels done.

Two or three rounds like this and the video is finished. That’s the seven prompts of a first video — and the reason your third one takes one or two.

The complete reference. Skim it once now so you know what exists; come back to it whenever you need the details.

Everything it can do

The agent is a general-purpose video production and editing system: it can create finished videos from nothing but a brief — concept, script, visuals, sound, and edit — or work from your raw footage.

And more. Behind the scenes, the agent coordinates multiple specialized sub-agents, each responsible for a specific capability; together they plan, execute, and refine with minimal input from you. If you’re unsure whether something is possible, ask — it has full knowledge of what it can and can’t do.

The two dials on every job

Both live in the UI, and both can be changed mid-conversation. Mode sets how often the agent pauses for you:

Production tier sets how big the team can grow. Match it to the stakes of the run:

The reference

What you can drop in

Files up to 1 GB each, straight into the chat:

videoaudioimagesscriptsbriefsCSVssubtitles.ttf.otf.woff.woff2.cube LUTsLottieAE templatesDaVinci / Premiere projects

Font files become reusable brand fonts, wired in properly; .cube LUTs bring a color look from your Premiere, Final Cut, or DaVinci workflow. Lottie files, After Effects templates, and DaVinci or Premiere project files come in too, for pixel-perfect configuration. Links work as well — YouTube, TikTok, and Instagram content can be pulled in remotely, up to roughly 20 minutes or 100 MB. For anything larger, upload the file directly. One exception to know: Google Drive and Dropbox links can’t be pulled in — attach those files directly instead.

Brand kits

Import your brand guidelines, Figma files, or just your website. Your brand becomes a saved kit — palette, fonts, logo, motion style — that your whole team can apply to any future video with one brand:name mention. Figma imports run through the Figma connector, and fonts your brand owns can be dropped straight into the chat. The full anatomy of a strong kit is in Getting started, phase 01.

Connectors

Connect the tools your brand lives in under Capabilities → Connectors: find the tool, hit Connect. For Figma, paste a personal access token and the agent can import your files. For Notion, connect your workspace and it can read your docs. Fonts don’t need a connector — drop the files straight into the chat.

Parallel jobs

Don’t watch the screen. Brief it, answer its questions, then leave. Run as many conversations at once as you want — our best customers keep five or more going — and you can ask the agent itself to split a project and start the parallel conversations for you. One conversation can also fan out on its own: ask for three variants and the agent builds the base once, then clones and customizes each in parallel.

Live editor

Every delivered video has real-time controls that update the preview instantly, with no re-render. Ask for the controls you want — “give me controls to adjust the volume of music and SFX” — and you get them: text fields, sliders, toggles, color pickers, dropdowns, and image, video, and audio pickers, grouped by scene so you only see what’s relevant to the moment you’re scrubbing. It’s also what makes client handoff and A/B testing work: same video, swap the headline or colors yourself.

Point and select

Point at anything in the video to select it. Drag a box to grab a few elements at once. Then tell the agent what to change.

Commands

Found a brief that works? Save it as a command. Define a workflow once — brand rules, a format, a checklist — and invoke it any time by typing /name. Combine several commands in one message. Update a command once and every future use inherits the change. Commands belong to your organization, so the whole team briefs the same way.

Skills

Skills are knowledge you teach the agent about your specific preferences. They help it understand what you mean when instructions are subjective. “Pick the best moments from this video” means different things to different people:

Lifestyle creatorSmiling faces, laughter, dynamic movement.

Gaming channelCrowded scenes, explosions, intense action.

PodcastOnly the guest’s answers; skip the interviewer.

Create a skill by describing what you want the agent to remember: “Create a skill for picking highlights. I want moments with high energy: people laughing, fast movement, or dramatic reactions. Skip anything slow or with long pauses.” From then on, “pick the best moments” applies your criteria automatically.

Skills aren’t only about taste. Teach the agent your brand guidelines, a technical domain so scripts come out credible, or a procedure your team repeats — “every product launch video follows this structure.” Complex skills can have sub-skills for deeper detail. Skills are saved to your organization, so everyone on your team gets them automatically. They sit alongside a built-in library of production expertise the agent already ships with, and they can be updated or refined over time.

Memory

You don’t have to re-teach it. The agent keeps a memory of your feedback and preferences across conversations: corrections you make, styles you approve, rules you set. Ask it to remember something and it will; ask it to forget, and it does.

MemoryWhat the agent observes — automatic, always on.

SkillsWhat you teach — deliberate, permanent, org-wide.

For anything you want locked in permanently, save a skill.

Templates & variants

Any video can be saved as a template and cloned. Need the same ad for three audiences? The agent builds one base, then clones and customizes each variant in parallel — headlines, images, music — dramatically faster than three from scratch. Every render and version is tracked, so you can manage multiple outputs of the same project side by side. The sentence to remember: “another one like the launch video, but for Germany.” And relax — a new video made from a format never changes the old ones; every clone is an independent copy.

Talk to it

Ask the best way to approach your problem, and push back when you disagree. It knows everything Poolday can do, and it can give you strategic advice as well as execution. This is allowed and encouraged.

Start in the chat

Poolday supports skills, custom apps, and ways to build your own workspace. Save that for later. For now, stay in the chat and ship your first videos.

Good to know

Honest limits, so you don’t discover them mid-project:

• Google Drive and Dropbox links can’t be pulled in — attach the files directly.
• Music and voice work best with stems. A single mixed track limits how cleanly the agent can recut, localize, or swap music — provide separate tracks when you have them.
• Captions burned into footage can’t be cleanly removed or translated. Keep text as separate layers where you can.
• Word-level cuts land close on the first pass and may take one revision round to be frame-perfect.

What it doesn't do

Post to social platforms. Poolday produces the video files; download the finished video and publish it where it needs to go.

The agent can execute almost anything. The difference between an average video and a great one is the direction it receives. Four habits make that difference:

The agent supplies the craft; you supply the judgment.

Your next step

Open a conversation, drop in your assets, and brief the agent. Stuck, blocked, or something behaves unexpectedly? Ask the agent first; it knows its own limits. Still stuck? Reach out to your Poolday team directly — and if you’re not on our shared Slack yet, ask us to add you.