← All field notes
Applied AI

What Are Flows? 7 Playbooks Your AI Agent Runs to Make Content

What are flows? Free playbooks your own AI agent runs with Voholabs Studio's tools to make explainers, carousels and blog posts in your brand, paid as you go.

A single folded paper map with one route traced in turquoise and a small apricot pin at the destination, on a flat paper-white background
One route, already drawn: what a flow hands your agent.

Ask your AI agent for "an explainer about our pricing" and, with a flow, it comes back about ten minutes later with a finished video in your colours. That is what flows are: a flow is a written playbook your own agent reads and runs with Voholabs Studio's tools, so one short request becomes a finished piece of marketing without you leaving the chat.

If you already use Claude or ChatGPT every week but still stitch together five tools to get one post out, flows are the layer that joins them up. Your agent can already write, and a flow adds the method, your brand and the tools to finish the job. This piece covers what a flow is, how a run looks, the seven that exist today and what one costs.

What is a flow?

A flow is a free, written playbook for one marketing job that your own AI agent reads and follows, using Voholabs Studio's tools to research, generate and save the result in your brand.

Think of it as the method a senior producer would give a capable new colleague: what a great result looks like, where the real data comes from, the traps, and the points where they should stop and check with you. The agent reads it on demand over MCP (the Model Context Protocol, the open standard that lets an AI assistant use outside tools, introduced in November 2024). Nothing gets installed. You connect Studio once, and every flow is there the next time you ask.

In many agent builders, a flow is a drag-and-drop chain of steps you assemble yourself and that runs the same way every time. A Studio flow is written advice instead, and your agent keeps charge of the order, the judgement and the writing, skipping what it does not need and running steps in parallel.

How does your AI agent run a flow?

A flow run is a short conversation with two kinds of moment: steps you watch, and gates where the agent waits for your yes.

  1. You ask in a few words. "Write a blog post about onboarding", "a 60-second video for Reels", "an infographic from this report". No app name needed.
  2. The agent reads your brief. The brief is the set of documents in Studio that describe your business: who you are for, how you sound, your colours, fonts and logo, and what you never claim. Every flow builds on it, so the result sounds and looks like you.
  3. It researches. Search, reading the pages that rank, finding the questions people actually ask.
  4. It stops at a gate. A gate is a review point: the script, the plan or the outline, with the cost quoted. One question, such as "Shall I go ahead with this script?" A word from you carries it on.
  5. It makes the thing. Images and narration come from Studio's generation tools; layout, rendering and text are done by your agent in its own workspace.
  6. It finishes in Studio. The result lands in your media library, or as a draft on your calendar, ready to schedule.

Every step appears in the chat as a card, so you can see what is happening without reading a wall of text. The gates keep you in the loop exactly where money is spent or something is about to go public, which is the same guard rail our guide to AI agents for business recommends.

Which flows can you run today?

Seven flows cover the jobs that eat the most hours in a small marketing team, as of October 2026:

  • Whiteboard explainer: a drawn-on whiteboard video with narration, 16:9 or 9:16 for Reels, TikTok and Shorts, in your own voice if you like. About 10-15 credits for a short.
  • Carousel: slides for LinkedIn or Instagram, plus a PDF. About 8-16 credits.
  • Infographic: a sourced infographic with charts drawn from real numbers. About 3-6 credits.
  • Photo-real images: product and lifestyle shots in your brand's look. About 3 credits per image.
  • Article thumbnails: a story-first cover image for a blog post or article. About 3 credits.
  • GEO blog post: a sourced article written to be quoted by AI answer engines, saved as a draft on your blog. About 3-10 credits.
  • Brief upgrade: your brief filled in and sharpened from your website, your customers' words and your competitors, plus a narration voice made from your own. About 1 credit.

GEO, generative engine optimisation, means writing so that AI answer engines can lift and cite your page: a direct answer up top, question-shaped headings, definitions in one sentence and a source for every number. The article you are reading was written with the GEO blog post flow, from our own brief. If that topic interests you, our piece on getting found by AI goes deeper.

Why do flows start from your brief?

AI content sounds generic when the model knows nothing about the business behind it. A flow closes that gap before the first word, by reading your brief.

That is customisation in practice. The same "make an explainer" request produces a calm teal video for one company and a bold orange one for another, because each takes its colours, logo, voice and audience from its own brief. When the brief is empty, the flow learns your business from your website first and saves what it finds, so the next job starts further ahead. It never borrows another brand's look as a stand-in for yours.

Flows filter the noise as well. Instead of choosing between dozens of tools and tutorials, you get one tested method per job, the one we use ourselves. The five workflows one person can run piece shows why one method beats a pile of tools.

What does a flow cost?

Reading a flow is free; you pay only for what it generates, from a prepaid wallet in Studio, with no subscription required.

A credit is the wallet's unit: $1 buys 20 credits. An image costs about 2.5 credits, and narration about 4.7 credits per 1,000 characters spoken (about 3 in your own cloned voice). So a 60-second whiteboard short, with its icons and voice-over, comes to roughly 10-15 credits, well under a dollar. Research tools are free. Your agent quotes the cost at the gate, before anything is spent, and reports what was charged at the end, with your balance.

Why do you stay in your own agent?

You already have an agent you trust. Flows run inside it, in Claude or ChatGPT, rather than asking you to learn another app.

That matters for two reasons. Your agent already knows the conversation you are having, so a flow picks up from "do it" without a form to fill in. And the work happens where you already are: the cards, the previews and the finished video all appear in the chat. Studio is where the results are kept and scheduled, and where you top up. The open standard behind this is now widely shared: OpenAI adopted MCP in March 2025, and in December 2025 it moved to a foundation under the Linux Foundation, so the same connection works across agents.

Try one now. Connect Voholabs Studio to your agent in Settings, then ask it for "a short explainer about what we do". Connect your agent.

What to do next

A flow turns one short request into a finished, on-brand piece, with you approving each step in your own chat and paying only for what gets made. Pick the job you repeat most, connect Studio, and ask your agent for it in your own words.

Start with one flow this week.

Frequently asked questions

  • Do I need to name the flow or the tool?
    No. Ask the way you would ask a colleague, such as "make a carousel from our last post", and your agent picks the matching flow from Studio on its own. A short reply like "go" carries it on at each gate.
  • Which AI agents can run flows?
    Any agent that connects to Studio over MCP, including Claude and ChatGPT. Flows are written for agents with their own workspace, so the agent can render videos and lay out slides itself.
  • Can I run a flow before my brief is filled in?
    Yes. The flow reads your website first, saves what it learns to your brief and asks only what it could not find. Every later job then starts from that brief.
  • What happens if I do not like the result?
    Reply with what to change. Each flow stops at its gates for exactly this, and a correction you make can be saved to your brief so the next run gets it right first time.

Sources

  1. Introducing the Model Context Protocol· Anthropic
  2. Model Context Protocol· Wikipedia
Mehrdad M. Sadeghi Profile Image
Mehrdad Sadeghi
Voholabs Co Founder

AI Ops Lead at the Livepeer Foundation, building the AI-native operating system behind a globally distributed team running open video and AI infrastructure.