Scribble is Pencil's AI creative assistant. It produces brand-safe content at scale, with human approval built in.

You tell Scribble what you need in everyday language. It works out which tools to use, shows you what it plans to do, and gets to work once you say yes. While it runs, you can carry on with something else. When it finishes, you review the work and approve it.

Your workspace has one Scribble, shared by everyone on the team. It applies your own context and preferences to your requests, so your work stays separate from everyone else's.

Scribble is available on Pencil Pro, and runs as an open beta. A Workspace Admin switches it on under Settings > Beta features, a dedicated page separate from AI governance. Once it is on, everyone in the workspace has access.

Scribble is enabled one workspace at a time. There is no bulk option, so a Workspace Admin switches it on in each workspace that needs it.

Enabling Scribble cannot be undone. The in-product description says so plainly: "Once enabled, this beta cannot be disabled."

Switching Scribble on also migrates the custom agents your team has already built into Scribble's agents and skills. Nothing is lost, and your Workspace Admin can review how each one was mapped. For more on this, see What happens to my existing work? below.

Once a Workspace Admin has switched Scribble on under Settings > Beta features, it sits front and centre on your workspace homepage. Type a request straight into the chat box to start a conversation. You can also open Scribble from the chat page, where all of your conversations live.

Scribble will soon also work in Brief Mode, Workflows (both as an overarching Workflow architect, and as a self-contained all-purpose Node), as well as Feeds and Sheets.

If it's your first time using Scribble, you may want to give it a little test. Try re-running a finished brief with Scribble. Then you can compare Scribble's output against previously completed work that you know is good. This will help you gain a working intuition about what Scribble can and can't do at the moment. Bear in mind that Scribble's capabilities will continue to improve and expand into new domains over time.

Scribble proposes before it acts. It shows you the plan before any step runs, and nothing starts until you approve it. You can redirect it or stop it at any point. You set the creative direction and give final approval on every asset.

Memory is how Scribble holds on to the way you and your team like work done. As you run briefs, hand it assets and ask for changes, it records durable facts about your brand, your products and your preferences, then applies them to later requests. You do not have to set out the same standing instructions at the top of every conversation.

When Scribble saves a fact about you during a conversation, a Remembered (you) chip appears in the chat. It has two buttons. Forget (the X) deletes the memory. Promote (the up arrow) moves it from your personal memory into your workspace memory.

This is useful when Scribble needs to remember preferences or ways of working that are fundamental across your whole team.

The test is whether a fact would still be true and useful on your next brief.

Brand guidelines, tone of voice and reference material belong in your Brand Library, under the Knowledge category, alongside Logos, Fonts and Colours. Scribble reads your Brand Library whenever a conversation starts. Memory covers how you work, and your Brand Library covers what your brand is.

No. Your briefs, assets and outputs are never used to train AI models, so running more work through Scribble does not retrain anything. Memory is a record kept for your workspace and used only for your workspace. Scribble gets better when Pencil improves the skills, agents and models behind it, tested against our own evaluations. For more on how your data is handled, read: Data Usage & Privacy in Pencil

Memory has two tiers.

Personal memory holds your own preferences, working style and corrections. It applies to every Scribble conversation you open in the workspace, and your teammates cannot see it.

Workspace memory holds facts about your company, products and audience, and how your team works. It applies to every conversation in the workspace, so it holds for your teammates too.

Neither tier carries across workspaces. A fact saved in one brand's workspace does not surface in another's.

To see what Scribble holds, ask it in chat, for example: "What do you remember about me?"

Think of Scribble as a boss agent. Instead of doing all the work itself, it recruits the specialists it needs from a library of agents and skills in your workspace, delegating each job to a pre-built agent with the appropriate knowledge and skills.

A skill is one procedure Scribble can run: resize an image, write a script, generate a voiceover. Skills can access various AI models as well as tools, connectors and third-party plugins.

The dividing line is decision-making. A skill runs a recipe. An agent works out the recipe as it goes.

An agent is a named specialist that holds an identity and its own reference material, and takes on a whole job rather than a single step. An agent can contain a number of Skills, as well as instructions about how and when to use them. Scribble chooses the right combination for each request, so you never have to name either one. Choosing for yourself is quicker when you already know exactly what the job is.

Slash commands are a shortcut to a specific skill. Type / in the chat box composer to search your workspace's skills by name, then pick the one you want. Reach for this when you already know the exact job to run, rather than describing the outcome and leaving the choice to Scribble.

To hand a job to a specific agent of your choosing, there are two options:

You can simply mention the agent by name, in the chat with Scribble (For example: "Scribble, please use the Creative Resizer agent to resize my image to 1920 x 1080 px") and Scribble will spawn the appropriate sub-agent to work on your request in the background.

Alternatively, open the plus menu in the bottom right corner of the chat message composer and choose Use agent. The specific agent will then be used in Agent Chat instead of Scribble and you can request specific work from the agent via plain language text chat, as you normally would.

An agent is worth building when your team repeats a job often enough that it deserves its own name and its own reference material.

Open Agents in your workspace and select New agent. The builder has four sections:

The pane on the right previews your agent's card and lets you test it before anyone else sees it. Save as draft keeps your configuration without making the agent available. Publish changes releases it at the scope you choose: this workspace only, or every workspace on your subscription.

For a single procedure that other agents and skills can call, build a skill instead. You can create your own Skills in the same way.

Configure each new skill's Tool connections, Instructions and set Model Preferences in the Definition Section.

For the jobs Scribble handles best, read: What can I use Scribble for?

Nothing is lost. Your conversations carry over, and Pencil moves any custom agents your team has built into the new setup automatically. Your Workspace Admin reviews the results once to catch anything that moved across wrong.

Some of the Agents your team has built pre-Scribble will be converted to Agents and others as Skills. Your Workspace Admin can see how each one was mapped after the migration runs.

The deciding factor is agency, and creative scope, not tool use. Does the agent need to autonomously make calls on what to do, or does it carry out a set procedure, protocol or 'recipe'? Anything that plans as it goes, picking its own tools and changing course based on what it finds, becomes an Agent. The previous Agents that were structured as a sequence of standardized operations to form a repeatable procedure - running the same steps on the same kind of input to produce the same kind of output - will instead now be found under Skills.

Scribble uses generations from your existing subscription allowance. The cost of each request depends on the model and the complexity of the job. For details, read: How does Scribble pricing work?