Well-written technical posts have one problem that’s awkward for the author: readers rarely read them in full. Thirty seconds of attention, a bounce, a return a week later - that’s the statistic you have to make peace with. It isn’t about laziness. It’s about the fact that attention in a LinkedIn feed, or over a morning coffee, rarely has room for fifteen hundred words.

In my publication pipeline a second format sits alongside the text. I call it HyperFrames: thirty- to forty-five-second video explainers, embedded on the blog next to the full article, distributed further across social, and reused in training. They don’t replace the text. They’re its „trailer” - a first layer, after which the reader decides whether to go deeper.

In this post I cover when it’s worth turning text into video, how to structure the script, which visualisations reliably work for QA and technical topics, how to embed video sensibly on a blog, and how to reuse the same material across several channels at once.

When a post earns a video

Not every post qualifies. In practice I apply three filters.

The first is whether the topic can be condensed into a single sentence without losing its essence. A post on „How to write AGENTS.md for a test repo” - yes. „A complete map of the concepts skill / tool / agent / MCP” - probably not, because there’s too much context for this format. From a cross-cutting map like that, a video can carry out at most one pair of concepts; the rest stays in the text.

The second filter concerns the visual element. QA has plenty of things that show well on screen - flows, directory structures, decision diagrams, report fragments, before/after comparisons. If a topic holds up only as prose, the video comes out flat. If there’s something concrete to show, it usually works.

The third filter is the scope of the topic. A cross-cutting post with a broad audience and an introductory character distributes well in this format. Niche depth aimed at someone who already understands the topic is too narrow a target for video - production cost doesn’t pay off against the reach.

In practice about half of my posts pass those filters and get a video. The other half stay plain text, and that’s fine. Not every topic deserves every form.

How to cut an article down to thirty seconds

This is editorial work, not technical. Calm-voice narration sits in the range of one hundred fifty to one hundred seventy words per minute. That puts thirty seconds at seventy-five to eighty-five words, and forty-five seconds at one hundred ten to one hundred twenty-five. Set against the fifteen hundred words of a full article, that’s a dramatic constraint.

Fitting into it requires three operations. You need to find the single sentence that says what the video is about - not a description along the lines of „an introduction to the topic”, but a concrete claim. Instead of „introduction to AGENTS.md”, „if you use an AI agent in your test repo without AGENTS.md, you’re wasting its time and yours” reads better.

The second operation is choosing one thing the viewer will learn. Not three, not ten - one. The rest stays in the article. Trying to cram several bullets into thirty seconds leaves none of them memorable.

The third operation is a single call to action. „Read the full post to see the file skeleton”, or „Save this prompt and try it this afternoon”. A video without a next step is like a well-written paragraph that trails off mid-sentence.

Script structure

Structurally I split the video into four phases of seven to ten seconds each.

Hook (0-7 s) is a statement, question or observation that stops the scroll. It has to be specific and slightly provocative - „Today we’ll talk about AGENTS.md in testing” is too bland. „Your AI agent gets a test repo and spends the first minute guessing - that costs you money” works better.

Problem (7-15 s) justifies the hook. It refers concretely to a familiar frustration - for example, „on every PR you explain to the agent that you use data-testid, not CSS classes”. The viewer has to recognise their own situation in it.

Demo (15-30 s) shows the concrete. A file fragment, a diagram, a quick before/after comparison. Visually, the most important phase - this is where the viewer decides whether the topic applies to them.

Takeaway (30-45 s) leaves one thing to remember plus one call to action. „File structure: setup, commands, conventions, don’t-do-this section. Full post linked below”.

Each phase has its own role. Hook stops. Problem justifies. Demo convinces. Takeaway routes onward. Removing any of them breaks the whole piece.

Visualisations that work for QA

In a thirty-second format the viewer doesn’t have time to decode complex imagery. In practice four or five types of visualisation work for me.

The first is a file fragment with a highlighted section. AGENTS.md on screen, the highlight following the narration. The viewer sees the text but doesn’t need to read it - the eye tracks the highlighted region. For „document structure” topics it’s one of the strongest formats.

The second is a „without” vs „with” comparison. Agent output without AGENTS.md on the left, with the file on the right. The difference is visible instantly, without explanation - the format largely sells itself.

The third is a decision diagram built progressively. A branch appears as the narration mentions it. That maintains tempo and lets the viewer follow the logic alongside the video, rather than trying to absorb the whole thing at once.

The fourth is a screen recording of a real workflow. Prompt, click, result. A dozen or so seconds. The viewer sees this isn’t an abstraction, it’s something running in a concrete tool. Works especially well for MCP demonstrations, AI workflows and agent-output review.

The fifth is an animated role or flow diagram - a few icons (tester, agent, repo, MCP, evidence) with arrows showing what goes where. It works for conceptual topics.

What I avoid: a talking head with no illustration (dull, viewers bounce), long text slides (nobody has time to read them), effects for effects’ sake (fade-ins on every cut are distracting), and stock video with „techy” motifs (looks random and cheap).

Rule of thumb: if the viewer pauses at a random moment, what they see should stand on its own. Every frame is a potential thumbnail.

Embedding on the blog

Technically embedding is simple, but a few things are worth getting right, otherwise the format doesn’t land.

Video goes above the fold, but under the headline and a single introductory sentence. The reader immediately sees there’s an option to watch instead of reading - without being hit with it in the first second. Hiding the film in the third half of the article means only someone who has already read to the end finds it.

Autoplay with sound is off the table. Browsers block it and the rest of the users feel attacked. Autoplay without sound (muted) with a clearly visible unmute button is acceptable. The safest variant is click-to-play.

Subtitles should be available from the start. Ninety percent of social feed views happen without sound; on the blog most views are muted too. Burned into the video or delivered as SRT - both options are fine.

I put one descriptive sentence under the video („a thirty-second introduction to AGENTS.md structure”). That’s SEO and accessibility at once, not an ornament.

Conversion is worth measuring: how many viewers press play, how many finish, how many click through to the article. Without those numbers you can’t tell whether the format is working or just taking up space above the fold.

Reusing the same material

The same thirty-second piece typically serves five channels - and it’s from that perspective that the economics of production start to balance out.

On the blog the video works as a trailer. Embedded under the title, with the in-video link leading back to the full article.

On LinkedIn the same film - but uploaded natively, not as a YouTube embed. LinkedIn rewards native video. A post with a short comment underneath and a link to the article collects noticeably more reach than a bare link to the blog.

On X, Mastodon or internal Slack I publish a shortened version, fifteen to twenty seconds - the strongest fragment pulled out as a „trailer of the trailer”. It carries a single insight, with a link to the full post.

In a presentation or internal training the video goes in as an illustration on a slide. In a presentation you can pause, comment, expand - and the slide gains thirty seconds of finished material that would otherwise require building an animation from scratch.

In onboarding a list like „first, watch these eight thirty-second films” cuts half an hour of reading down to half an hour of viewing, with a full mental skeleton on the other side. Details stay in documents, which can be long and don’t need to be read immediately.

One production, five uses. The economics of content marketing in QA start to make sense only when you count reuse, not single use.

Pre-publication checklist

Before every video I run through eight checks that block publication when something isn’t wired up. Is the single sentence that says what the film is about written down and confirmed. Has the single thing to remember been chosen deliberately. Is the hook in the first three seconds non-generic - „today I’ll talk about…” doesn’t qualify. Are the subtitles burned in and proofed for typos. Does every frame stand on its own - I check with three random pauses. Is the link to the full post in the video description. Does the thumbnail (first frame) work on a mobile screen. Are the versions prepared for each channel: blog embed, native social, trailer of the trailer, training slide.

Eight points, a few minutes of review. Protects against publishing something that only lives on the blog and then fails to scale anywhere else.

Summing up

Not every post deserves a video - three filters (single sentence, visual element, cross-cutting scope) filter out most. In thirty to forty-five seconds there’s room for one opening sentence, one thing to remember, and one call to action. The structure is hook → problem → demo → takeaway, each phase seven to ten seconds. Visualisations that work: file with highlights, before/after comparison, progressive diagram, real screen recording. Embedding on the blog without audio autoplay, with subtitles and a text alternative. Reuse across five channels in parallel - blog, native social, shortened version, slide in training, onboarding.

This post closes the first set of six. In the rounds that follow I go deeper into specific AI workflows, examples of good and bad AGENTS.md files, and experiments with evaluating agent output - this time with concrete project case studies rather than just conceptual frames.