Blog/AI slop

Where AI helps a video team, stage by stage

Planning, capture, edit, polish, publish: each stage has a part AI does well and a part a person has to decide. Here is the split.

Kevin Blanco··7 min read·Every stage

The short answer

AI helps most with the repetitive parts of a video's life: transcribing, rough cuts, captions, translations and resizing. People should own the decisions: who the video is for, what to show, the hook and the final cut. Treat AI as a fast first draft at every stage, and have one named person approve it.

01

Five stages, two jobs

A video goes through five stages: plan, capture, edit, polish, publish. At each one there's a draft to produce and a decision to make. AI is good at the first and shouldn't own the second.

Get that split right and AI gives you hours back. Get it wrong and you spend them fixing confident, generic work.

FigureDrafts above the line, decisions below it. The split we'd suggest, drawn to show the idea.

02

Stage by stage

StageAI draftsA person decidesWhy it splits there
PlanTranscripts of research calls, clusters of viewer questions, outlinesWho it's for, the one action, the journey stageThe plan sets what every later choice aims at. A draft can't know your viewer.
CaptureAudio cleanup, live captions for reviewWhat to record, whether the demo is accurate, the real errorOnly someone who has run the product can tell if it's right.
EditA rough cut, silence removal, candidate momentsThe hook, the order, what to cutA rough cut picks moments by sound and pace. A person picks them by use.
PolishCaptions, translations, resizing, thumbnail draftsProduct names, tone, which translations ship, what the thumbnail promisesNames and terms need a human eye.
PublishTitle options, descriptions, chapters from the transcriptThe final title, the link, the next step, the number to checkA title is a promise. Whoever made the video should make it.

03

The handoff

Most of the trouble sits between the stages, where a draft gets mistaken for a finished thing. Three habits keep that from happening.

  • 01

    A named approver

    One person per video, not a committee. Before the AI produces anything, you know who will say yes or no to it.

  • 02

    Reject the first draft

    Expect to change it. A first output you accept as it is probably went through unchecked.

  • 03

    A log of what you change

    After a few videos you'll see the same fixes: product names it gets wrong, hooks it keeps writing. Put those into the brief so the next draft starts closer.

04

A four-week way to start

Don't change every stage at once. Add one at a time and look at what it did.

  1. Week 1

    Transcripts and captions

    On every video. Check product names. Note the hours saved.

  2. Week 2

    Rough cuts

    For talks and long recordings. A person still picks the moments.

  3. Week 3

    Write the brief

    From the log of what you changed: voice, names, what to show.

  4. Week 4

    Count

    Hours per stage, edits per draft, and videos used.

05

How to tell it's working

  • Hours saved

    Per stage

    Measured per stage, so you can see which one pays.

  • Edits per draft

    Trending down

    As the brief improves, drafts should need fewer changes.

  • Videos used

    Share published

    More of what gets made should reach the audience.

If hours go down and the share of videos you use goes down with them, you haven't saved anything. You've made less of what works.

FAQ

Questions people ask

Where can AI help most in video production?

With the repetitive work: transcripts, captions, translations, resizing for each platform, silence removal and rough cuts. These have a known answer and a person can check them quickly.

Should AI write the script?

It can draft one, but a person decides the idea first and rewrites the draft, because the script has to be accurate about your product and about what the viewer should do next. Reject the first version by default.

Can AI edit a video for me?

It can assemble a rough cut and suggest candidate moments. The hook, the order and what to cut are decisions, and a rough cut tends to pick moments by sound and pace, not by how useful they are.

How do I keep AI output on brand?

Give it a one-page brief with your voice rules, words to avoid and what to show, name one person who approves every output, and keep a log of what you change so the brief improves.

How do I measure whether AI is saving time?

Track hours per stage for a few videos before and after, and count edits per draft and the share of videos you publish and use. Hours saved only count if the videos still get used.

DevReel

How we do it at DevReel

The plan comes first: who it's for, where they are in their journey, and what we'll check afterward. At each stage after that, AI does the repetitive part and a person who has used the product makes the call.

Cloudinary's videos usually average around 400 views. Ours passed 9,000 in the first week. For ngrok, watch hours tripled and subscribers went from about 450 to over 800 in the first week.

Cloudinary views

9,000+

in the first week, against a usual ~400

ngrok watch hours

3×

subscribers from ~450 to 800+ in the first week

Now the hard part

Tell us where the hours go in one of your videos. We'll show you which stage to hand to AI first.

Kevin Blanco founded DevReel. He has made 500+ developer-tool videos and spoken at 80+ developer conferences.

Keep reading