A persona-led walkthrough video of a working product, narrated in the voice of the user it was built for, captured by a browser driving itself through the screens, narrated by a synthetic voice trained on no human in particular, exported as an MP4 fit for a board pack. Six-figure work, until recently. A specialist agency, a script, a recording session, an editor, several weeks of back and forth.
I built the tool to do it in an afternoon. Not the videos. The tool.
What I built and why
The skill is called persona-walkthrough-video. Point it at one of my products, and it works through the following sequence on its own. It reads the codebase to understand what the product is. It infers a plausible persona from what it finds. A social housing tenant frightened of complaining. A compliance manager burned by a missed deadline. It maps every screen in the application. It asks which screens matter most for the audience I have in mind. It plans a journey through the product the inferred user would plausibly take. It drives a browser through the journey, capturing the screen as it goes. It writes narration in the persona's voice. It utilises OpenAI TTS to provide narration. It cuts the video to match the audio. Simple and reusable.
The reason I built it is more interesting than what it does. I have two products ready for market. Both need sales material. Sales material for software in 2026 is a short, persona-led video landing the proposition in under two minutes, not a deck. Buyers expect it. And every time I needed one, I was looking at a quote, a calendar, and a delay.
So I stopped looking and built the tool instead.
The arithmetic has shifted
There is a particular reflex experienced technology leaders need to unlearn. The reflex is to evaluate a small piece of software by the historical cost of building it. A walkthrough-video tool, ten years ago, was a product. It would have warranted a team, a roadmap, a backlog, a release plan. Months and five or six figures of investment.
Today it is a skill file. A few hundred lines of instructions to an AI coding agent, supported by three reference scripts. Built in an afternoon by one person who knew what they wanted it to do and had the cross-domain context to specify it properly.
The arithmetic has inverted. The expensive part is no longer the building. The expensive part is knowing what to build, knowing why it matters, and knowing what good looks like when the agent shows you a draft. I have written elsewhere about the inversion of the generalist penalty. This is the same insight viewed from a different angle. When the cost of construction collapses, the value of judgement rises in proportion.
The disposable-tool pattern
The persona-walkthrough-video skill is, by design, narrow. It does one thing. It is not a video platform. It is not a product. It has no roadmap. If a better way of producing this kind of video emerges in six months, I will throw it away and build a new one.
This is the pattern senior leaders are slowest to absorb. Software has been, until now, expensive enough you built it to last. You designed for reuse, for extension, for ten years of maintenance. The economics demanded it. Throwing software away felt wasteful because the labour was so dear.
When the labour cost approaches zero, those instincts stop serving you. A tool doing exactly what you need, today, and discarded when your need changes, is no longer extravagant. It is rational. The fleet of small, specific, disposable tools a single competent person now maintains replaces the bloated general-purpose platforms which used to be the only economically defensible option.
I now have a personal toolkit of around twenty skills like this one. Each does a single job for one of my products or one of my workflows. None of them is a product. All of them save me time.
What this means for boards
If you sit on a board, the implication is uncomfortable but worth sitting with. The technology spend you signed off this year is, in significant part, paying for capability an internally-built disposable tool would now deliver for a fraction of the price. Not all of it. Some procurement is genuinely necessary, and some workloads need industrial-grade software. But the slice which is not is larger than most boards understand.
The barrier is no longer technical. The agents are good enough. The models are cheap enough. The barrier is whether the organisation has people who know what to build, why, and what good looks like, and whether those people have permission to build it.
The second clause is doing most of the work. The people exist. Most organisations have not yet given them permission to act, because the institutional muscle memory still treats software construction as an expensive, controlled, governance-laden activity warranting a project board.
The boards starting to ask whether the muscle memory still serves them are the ones worth watching. The persona-walkthrough-video skill cost me an afternoon. The next one will cost less. Boards absorbing this trajectory will spend differently. Boards which do not will keep paying yesterday's prices for tomorrow's capability.
The question worth asking at your next strategy session is not whether AI will change your industry. It already has. The question is whether your operating assumptions about what software costs, who builds it, and how disposable it is, have caught up.