Think about your next board meeting. Somewhere in the fortnight before it, three or four departments will each build a bunch of PowerPoints (in the unique style the CEO likes). Finance will export the P&L and the balance sheet out of the finance software and into Excel, because that’s where the analysis actually happens. Someone will dig out last month’s minutes, and the reports that came in from other people will be collected into a folder somewhere, and we’ll all try to remember what was agreed. Every one of those things is a finite asset: generated for that meeting as an obligation, looked at once, then lost, or if you’re lucky, filed well enough to give the next meeting some semblance of continuity. It’s ‘task theatre’.
Now, let’s explore a weird concept, because it’s fun, and more plausible month on month. What if none of it existed, until the moment it was needed?
I don’t mean automated, and I don’t mean stuck in a dashboard. What if the business ran on interconnected sets of data that never compiled themselves into PowerPoints or P&Ls at all, until the point somebody needed to look at one - and at that point, it was generated in realtime, for that person, in the form that person best appreciates?
So, in this article, I want to make an argument I’ve been circling for a couple of years in content, and push it further than I’ve pushed it before: into concepts, code, and into the whole digital experience. The idea / narrative / concept is the asset. Everything else is a ‘rendering’ of it.
Content got there first - I know, because we’re there now!
If you hold a narrative, an idea or a story properly, it can be told as a video, an interactive game, a VR experience, a website, a brochure, a book, and a blog post, all at the same time. This isn’t science fiction - it’s becoming a reality each passing month these days.
The execution cost of a packshot, a lifestyle shot or a video is now essentially zero (don’t take that literally - the approach is the value, the rendering is zero). Each one is just the representation of a culmination of a rich dataset - the product, the light, the context, the sound, the geometry - all coming together and rendering at the point of need. It’s the bet we’ve built TungstenFLOW on: content and brand consistency as the dataset (reasoning layer), rendered when it’s needed rather than produced and stored.
But, let’s take that one step further. Next year’s figures can render as a chart in the boardroom, a spreadsheet for the finance department, a video series for the CEO, and a set of interactive learning sessions for the French marketing team. Same idea, four renderings, and nobody needed to “make” the execution of the four things.
A brochure site is a rendering. The product listing on a third-party retailer is a rendering. The demo that illustrates a product, an experience or a codebase is a rendering. Which means the code is one too. Code isn’t a different category of thing from content; it’s another execution of the same digital experience.
I’ll caveat this straight away, because I’d rather do it than have you do it for me. None of this is plausible as a whole system today. There are too many interdependent factors for a business to genuinely run like that right now. But the foundations are being built, and they’re recognisable when you see them: businesses run on interconnected datasets that don’t render anything out, and don’t compile anything into a finite deliverable, until it absolutely has to be represented as one. Something that never leaves the command line, runs across several departments, and never appears as a deck or a P&L until the point it’s reviewed.
Take the bugs, kill the complacency
I used to do a lot of charcoal drawings, because at the core of the concept, you’re using the negative space to reveal the image underneath. So, to start… you cover the canvas with charcoal - completely. Then you reveal the artwork by removing it. Choosing a route and exploring can be really difficult. We have decades of formality and structure built into how we approach things. But, this is where I want to pick the fight. A seasoned developer said to me, in the way seasoned developers say stuff, that ‘agentic engineering’ - Claude Code, Codex, whichever - isn’t fit for enterprise consumption. It’s messy, inaccurate and carries too many bugs. Give me code written properly to spec any day. It’s not the first time I’ve heard this, and it won’t be the last either.
My answer was, and is, that code written to spec is a benchmark of the old world. Not an inaccurate one: for manufacturing and infrastructure it’s exactly right. But it’s finite and linear. It doesn’t easily evolve. It doesn’t adapt to the changing conversation, or to a business whose direction is shifting in bi-monthly cycles now. So given the choice between a buggy agentic ‘proof of concept’ and something that cookie-cuts a product off an assembly line, I’ll take the bugs. Bugs can be fixed.
And before anyone shouts me down: I’m not talking about finance, medical or infrastructure, or anywhere else where compliance, governance and validation have to be 100% accurate and bulletproof day zero. That territory keeps its rock-solid code and should. What I’m talking about is the far larger territory where rock-solid construction has become a walled garden in its own right - the thing that prevents you experimenting with anything, or scaling anything, because it was produced in isolation and can’t be manipulated afterwards.
When the code or experience is temporary, the speed at which you reach working functionality outweighs the bugs sitting inside it.
The people building the tools have already placed this bet. The engineer who created Claude Code told Fortune in June that it is now “100% written by Claude Code”, and that he hasn’t written a line by hand in eight months. Marketing changed dramatically in four or five years. Enterprise code is changing in two, maybe less. When does this lead to operations, commercial, finance, or the sales floor?
Don’t ‘AI’ the old workflow
Every launched a project this autumn called Thesis Statements - short essays from people building at the frontier, which get debated live at their Thesis 2027 event in Brooklyn on 5 November - and one of them, from Sumeet Singh, is titled “Founders who AI-ify existing workflows will lose”. I’d stand with most of that collection, but that one got me the most, because it’s specifically talking about long lead times and old approaches to production, project and development methodology. I have seen this soooo much in my domain, that one of my content pillars is literally ‘compression economics’. We’ve seen the change in the past 3 years - only fools would say we haven’t. And those that continue to approach business in the same way can ONLY expect the same output. Old school will build. It will just build the wrong thing, and by the time you get to the finish line, the result you needed is going to be different. There used to be a startup term called ‘failing fast’ - designed to illustrate the need to build something quick and generate an interest or ROI even quicker to see whether the idea had legs. I think this is not the same as what I am saying here, but it’s close. The idea here is keep the friction in what you choose to make, but don’t render it out as one thing yet. Work with the concept across multiple domains, explore how you could tell this story as a book, a video, an audiobook, a mobile game, or an interactive DOOH display.
I’ve watched exactly that conversation arrive from the content side over the last few years. If you produce the thing in isolation, you can’t manipulate it afterwards. So the work isn’t producing the thing. The work is creating fluidity in experiences, ideas and concepts, and letting them render or execute themselves however they need to in order to support, advise, inform or entertain whoever’s in front of them.
The sea of sameness
There’s a second essay in that collection I’d stand behind, from Emily Vernon: “AI will spur a resistance to mediocre ideas”. Her line is that AI makes mediocre ideas irresistible - why spend six months on a brand when you can generate something passable in a few prompts for free? I call the same thing ‘LLM averages’. You’re leaning on the model to do the full thinking for you, and you end up with an average of an idea. A design by community mess.
You can see this fatigue setting in already. Ask a coding agent to design a flyer, a brochure or a website and look at what you get, then look at the trends running against AI-generated banners, posters, digital influencers and UGC. The backlash isn’t really about the output being bad; it’s the volume, and the ‘sea of sameness’ around it, generating the moniker ‘AI slop’, which has created visual fatigue a lot this year. It’s a huge problem across software as much as social media.
I recognise it from the other side of the desk. When I used to hire for my teams, you could spot the showreel built from a tutorial or a stock file from a mile away! It sat vanilla in the reel - no individuality in how the asset was used, just a copy. It’s the mark of taking the easy road. What’s changed is that the general public now has that same eye, because the volume of ‘Canva-approached’ content has trained it, and the appetite sours very, very quickly.
Which is where the two halves of this argument meet. A rendering from your own rich dataset - your product, your figures, your voice, your idea held properly - is the opposite of a rendering from the model’s average. Fluidity and sameness look similar from a distance, which is exactly why the distinction is worth making out loud.
Accessibility is a render target
Here’s the part that usually gets missed, and it’s the part I lean on a lot in L&D. Accessibility, localisation and translation all typically happen after the experience, event, or content is created. The dyslexia-friendly version, the dyspraxia-friendly training series, the version that works for someone hard of hearing, the French edition - all edited, or considered afterwards… if there’s budget left.
Once ‘everything is a rendering’, there is no master version. The dyspraxic training series gets the same weight as the original-language one, because they’re both just the idea rendering for the person in front of it. Accessibility, regionality and cultural relevance stop being retrofits and become equal render targets - as much about equalising them as about offering them.
The day it all ships at once
Imagine the day that a film, the game, the website and the marketing campaign all come out at the same time, built at the same time, in the same way. Bonkers. I know how that sounds. But it’s exactly how AI video sounded when I started talking about it four years ago, and that turned out to be a comment on the timeline rather than on the idea. That’s the pace of change. It is bonkers - but it’s far more plausible this year than ever before.
If you’re deciding where next year’s budget goes, the decision underneath all of this is simpler than it looks: build the interconnected datasets rather than the deliverables. The deliverables will render themselves when somebody needs to look at them. The PowerPoint can wait till Thursday, focus on the idea for a few more days.
