A week of short-form video with AI.

Disclosure: the editing step in this workflow uses Phillipe, the AI video editor I co-founded. I explain why I built it separately. The planning and filming steps work with any editor.

The short answer

Batch everything. Plan the week’s ideas in one sitting, film them in one session, let an AI video editor produce the rough cuts, then spend your own time on review and on the parts that need judgment. The AI handles the mechanical edit. You handle the ideas, the hooks, the review and the final call.

The examples below assume five videos a week. Scale them to whatever cadence you can actually keep. A rhythm you can hold for months beats a burst you abandon after two weeks.

The shape of the week

• Day one: plan. Choose five ideas and outline each one.
• Day two: film. Record all five talking clips, then the supporting footage.
• Day three: edit. Upload, start the edits, review the rough cuts and request changes.
• The rest of the week: publish and listen. Post on your schedule, reply to comments and collect next week’s ideas.

Separating the days matters because each one uses a different kind of attention. Planning needs quiet thinking. Filming needs energy. Reviewing needs a critical eye. Mixing all three into every single video is why making one video can take a whole day.

Step 1: plan from real questions

The best source of short-form ideas is the questions people have already asked you: direct messages, comments, client calls and conversations after a talk. A question someone actually asked comes with an audience built in.

For each idea I write four lines, not a script:

• The hook. The first sentence, the one that earns the next three seconds.
• The point. One sentence. If I cannot say it in one sentence, it is two videos.
• The proof. An example, a number I can verify or a short story.
• The close. What the viewer should do or think next.

This is the part I do not hand to AI. I will use a model to pressure-test an outline or suggest a sharper first line, but the idea and the opinion have to be mine. It is the same rule I follow in business work: use AI to build, not to think. A feed of averaged ideas is easy to scroll past.

Step 2: film everything in one session

Setting up costs more time than recording, so I set up once and record every video in the same session.

• Film vertical, at eye level. Shoot 9:16 from the start so nothing important gets cropped later.
• Face the light. A window in front of you beats a lamp behind you.
• Protect the audio. A simple clip-on microphone, or the phone close to you in a quiet room, matters more than the camera.
• Restart sentences freely. When I stumble, I pause and say the whole sentence again. An editor that removes repeated takes turns this into the fastest way to a clean line.
• Leave a beat between thoughts. Clean pauses are easy to cut. Words that run into each other are not.

Then I record supporting footage. For each idea, I capture a few short clips of what I am talking about: my screen, my hands doing the task, the place or the product. A few seconds each is enough. B-roll works when it shows what the words say, so I shoot it after the talking clips, with the outline in front of me.

Step 3: let AI do the mechanical edit

This is the step that used to swallow the week. Now it is mostly uploading and reviewing. Here is how it runs in Phillipe:

• Upload and label. The talking clip goes in as A-roll and the supporting clips as B-roll. Labelling takes seconds, and it is the most useful context you can give any editor, human or AI.
• Pick a style. I use Kinetic, Phillipe’s signature style with sculpted typography, for videos where every word carries the idea. Cinematic fits when the video is a story.
• Add direction only when it matters. Most of the time I leave the note empty. When I write one, it is concrete: a length, what to keep, what to avoid. If you want to get better at this, read how to give an AI video editor instructions it can follow.
• Check the estimate, then start. You see the credit estimate before the edit runs.

Phillipe transcribes the speech with word-level timing, cuts pauses, filler and repeated takes, keeps the complete story and places each B-roll clip on the spoken beat it belongs to. That is the rough cut, and it comes back to you for review.

Step 4: review the rough cut like an editor

The rough cut is where your judgment matters most, so I review it before any styling. I watch each video once, all the way through, and ask three questions:

• Does the first sentence work as a hook, or is there a stronger opening later in the clip?
• Is the point still complete, or did a cut remove something the argument needs?
• Does each piece of B-roll show what I am saying at that moment?

Once the story is right, the style pass adds word-timed captions, pacing, supporting footage and music under my voice. Reviewing the story first means I am never polishing captions on a sentence that should not be there.

Step 5: ask for changes in plain language

On paid plans you can ask for changes in chat. Good requests are specific, and they say what to keep as well as what to change:

• “Open with the sentence about pricing and keep everything else in order.”
• “Remove the second example. The video works without it.”
• “Keep it under 30 seconds and use my original audio.”

I group my changes into one clear message. Every correction is charged in full in credits, so one precise request beats three vague ones. Then I make the final call and download the vertical video.

Step 6: publish, then listen

The text under each post and the replies to comments are mine. They are short, and they are where people decide whether to trust you. I do not automate them.

Every good comment is a candidate for next week’s list. That closes the loop: the questions people ask about this week’s videos become next week’s ideas. For a version of this system written specifically for personal brands, see the weekly short-form video workflow for personal brands on the Phillipe blog.

What I automate and what I keep

• Automated: finding clean takes, cutting pauses and filler, word-timed captions, placing B-roll on the right beats and the first pass on pacing and music.
• Kept: the ideas, the hooks, the story review, the final call, the post text and the replies.

The line is the same one I draw in every AI project. Automate the mechanical work that repeats. Keep the decisions that make the result yours. I wrote more about drawing that line in the AI operator’s guide to automating video editing.

What it costs

Phillipe’s free plan includes one finished video a month. That is enough to test this workflow on your own footage, but not enough to run it every week. A weekly cadence needs a paid plan. How far the credits go depends on the length of your footage, and you see the estimate before every edit, so you can plan the week before you spend anything.

The easiest way to start is to take one clip you have already filmed and run it through phillipe.ai. If the rough cut saves you the hours the edit would have taken, the rest of the workflow is worth building around it.