Share

Consider the following pre-AI use case: searching for a recipe for hummus on Google. What it returns is a collection of food recipes, largely ranked by popularity, where both finding interesting or unusual variants and the core recipe to improvise on requires trawling through a mass of links.

To address this information overload, I looked for the recipes for recipes themselves and have build a recipe search engine at cookette.com (ask me for a code to try it) that does a thorough AI analysis of hundreds of recipes. It’s a more efficient UI for food search.

After analysing many recipes, Cookette extracts the archetype, in terms of both ingredients and preparation.

It then ranks variants based on frequency of use of an ingredient and its purpose; regional variants and variants by famous chefs of restaurants.

What this does it allow you to understand the core, to enable you to grok the essence of it an improvise more easily. It also allows you to assimilate hundreds of variations in the easiest way possible. I allows you to accumulate, process and deliver new knowledge more efficiently.

Subscribe now

More broadly, I am interested in recipes and variants on a theme as a new UX and design methodology particularly suitable for the AI era.

DNA, the foundational information component of life itself is not really a blueprint but a recipe. A blueprint for a fruitcake would specify the coordinates of each raisin, whereas a recipe just indicates the steps where each outcome is a variation of the theme in terms of the actual placement of the fruit in each cake delivered from the same recipe.

As the DNA example illustrates, recipes are not just about food, they are a completely different way of building and analysing things where process and archetypes are embedded in the instructions. I have been looking at applications for recipes in three domains: architecture; games and tools. In each, I have been building some AI driven analysis and possible applications.

Architecture

For architecture I am interested in what is called vernacular architecture, the background styles such as a Georgian terrace, where the actual architect is often unknown and where each is built as a slight variation on a pattern - a recipe.

Recipes for architecture are much more interesting, lasting and important than the outputs of star architects and the best attempt to encode architectural recipes was by Christopher Alexander and his ‘Pattern Language’. Ironically, a Pattern Language is more familiar to computer science students than architects as it describes a methodology for encoding modularity that is applicable to object oriented software development.

Architectural recipes produce variants based on a theme and this variation is what makes them more artisanal and organic. The example below shows a parametric design for chinese courtyard houses based on a common recipe for the courtyard but adapted in each case to follow the existing topography and water drainage for natural variation. A recipe approach should be able to create true modern architecture but with the same organic qualities of a mediaeval hilltop town. I have written exactly how and why this is the case, here.

Recipe based architecture creates the possibility to use AI to actually design things that can be built rather than just pretty prictures that act as concept designs, much like AI software development tools can produce actual applications.

Linking building software that actually understands components, so that you can potentially produce a model in Revit that can generate working drawings that can be built is much easier from recipes as they encode meaning rather than just ‘lines on paper’.

Not only should it be possible to develop architectural styles through recipes that are far more organic and human, but it should be possible actually build them via AI enabled design where this cannot be done for one offs.

Share

Games

To produce a recipe for recipes, an overall concept must underly them and science fiction author, Kurt Vonnegut perfectly illustrated this when he mapped story archetypes to a fixed set of emotional arcs for a finite set of them.

Like many seemingly non comprehensive categorization strategies, the results of Vonnegut’s archetypes when mapped to existing stories, shows real correlation.

I extended Vonnegut’s method via AI to categorize archetypes for humor, looking for patterns and cultural variations in jokes, to see if this was potentially useful - and once I had managed to get decent outputs by asking the right questions, I extended the same method of interrogation of AI to analyse games, game mechanics and game archetypes.

At some point, I will publish this analysis separately as I think it is genuinely useful not just for games design, but understanding human behavioral incentives and how to design businesses and services of any type.

Tools

The last category I have been exploring in terms of archetypes and variations on a theme is tools where tools are again equivalent to the genotype of human physical output and the output of tools are the phenotype.

Richard Dawkins used the term extended phenotype to refer to things like snail shells. Our own houses are the human extended phenotype and the tools that created them are something that can be analysed in terms of physical agency (ability to move and transform physical materials).

Analysis identified 11 tool primitives that deal with matter, energy and information:

• Transform (cut, shape, bend, grind)

• Assemble / Disassemble (join, fasten, weave, weld, dismantle)

• Measure / Guide (mark, align, calibrate)

• Transmit Energy / Motion (lever, wheel, gear, motor)

• Contain / Transport (store, carry, channel)

• Preserve / Protect (shield, insulate, conserve)

• Replicate / Copy (molds, printers, duplicators)

• Sort / Filter (sieve, algorithm, centrifuge)

• Destroy / Erase (shred, dissolve, bomb, delete)

• Sense / Detect (microscope, radar, voltmeter, biosensor)

• Inform / Simulate / Control (record, calculate, communicate, model, program)

Via asking the right questions of AI models, I have started to create a categorization and genealogy (the tools that make the tools) of human tools from a flint axe to a particle accelerator.

The reason I think this is important is that it potentially creates a framework of understanding of agency in the real world, for defining physical AI capabilities from general purpose, AI enabled robotics.

Summary

I believe AI is very good at sourcing archetypes and variations on a theme, and that this creates opportunities for new UX paradigms specifically adapted for the AI era. I believe that this can be described as a language of information retrieval and processing that centers around recipes and that there are several domains where it may be particularly useful, such as in games and tools.

Subscribe now

Its economic model relied on distribution-side network effects in sales and marketing. That required equity funding to cover the gap between growth and monetization. Without funding, early monetization would slow before the market was captured. This created a winner-takes-all arms race between other companies pursuing internet-leveraged growth.

Share

I also discussed how the internet is causing similar societal disruption to the printing press, since the structural changes to communication are greater and how they might be resolved in future with small-world network topologies.

Now we can look at AI.

This is the big one, the information technology innovation that shows we are not merely in some Carlota Perez-style, later phase of industrialization but in a transition from industrial to information age that is as big as the one from agrarian to industrial.

We are living through a civilizational paradigm shift and like all transitions it will bring good things and bad, but we cannot uninvent things and the net gain will represent progress and will depend on embracing it with alacrity.

With AI, the software itself is a network, there are no algorithms or code in the traditional sense of codified, rule based ‘if-this-then-that’ logic. And noise is added which paradoxically achieves greater chance of accuracy in outputs. With noise added, outputs vary with the same input and so AI systems are statistical whereas boolean logic based ones are deterministic, this is a completely different way of processing information.

Subscribe now

AI models based on modern, ‘attention’ based (broad, simultaneous knowledge) transformer machine learning are the skeleton framework of settings left behind when you remove the data that created them. This scaffold framework then becomes a sieve that understands the world and can generate a response as it sifts a question.

This understanding seems to be based on a hidden dimensionality (i.e. orthogonal sieves) where it may be that approaching a dimensionality that matches reality with 0% error would require massive or even infinite resources such that we are stuck in the world of statistical and partial hallucination when it comes to AI.

In such a scenario the architecture and interplay between deterministic if-this-then-that logic and AI systems will become critical.

Most modern technologies seem like magic to people who didn’t create them. AI seems like magic to those that did. To paraphrase the observation about the mystery of Quantum physics, ‘if you think you understand AI, then you don’t understand AI’.

Modern generative AI networks can output anything from a text summary to a video to an application itself. They can produce things that used to take a lot of time and effort and therefore lower the cost of production. In the extreme case where they create a better version of themselves they create a runaway process of self improvement that lowers production costs to near zero.

In other words, in AI the network effects are at the opposite end from internet ones - production rather than distribution and given this underlying structural difference it transforms the business ecosystem AI system operate in, from startups to incumbents and investors.

This also means that the real AI revolution will be driven by software generation not video, text or image content generation, since software than writes software will create a flywheel whereas software that creates words creates a spam fest.

When AI enabled agency is taken to the physical world through AI enabled automated manufacturing that in turn creates manufacturing automation (robots that build robots), this will also be a flywheel that will trigger a manufacturing runaway event.

The biggest shift in AI development is that costs of production of applications drop to zero but the costs of running them doesn’t as apps often rely on connection to large AI models who charge on a per token use basis. This changes flows of capital upstream to large balance sheets and infrastructure providers: large frontier models developed by large tech companies; chips; data centers and eventually energy. AI costs are ultimately measured in token charges which in turn are a function of energy costs as fixed capital expenditure standardizes.

There is currently an arms race between China and the US in terms of infrastructure build out, on the assumption that for whoever gets to the AGI flywheel of automated self improvement, first, wins.

Because AI is general purpose technology with huge efficiencies not just in the marginal cost of production (shrink wrapped software on CD) or its distribution (internet SaaS) but in the actual cost of development of an application (AI coding), it will require huge customer bases for the productivity benefits to show. In this scenario the West may have to partner with India to create a sufficiently large rival bloc to China and South East Asia and we are already seeing OpenAI developing with the India market as a growth priority.

The structural differences in AI that create production side network effects and upstream value creation will create a barbell opportunity that may disrupt Venture Capital.

On one side of the barbell sits small agile teams leveraging a customer base for multiple products, monetized early and requiring less rounds of financing and operating in non winner-takes-all markets. These will look somewhat like creative agencies but producing applications not campaigns and if they do take venture financing it may only be one or two rounds.

On the other side of the barbell will be deep tech (real science backed by protected IP) infrastructure opportunities. These range from large language model innovations to novel chips or energy innovation or in-house development from large organizations based on models from their own data at volume. All will involve relationships with large balance sheet incumbents or governments and require large ticket rounds out of the reach of most VCs.

Subscribe now

The death zone will be the bit in the middle - VC backed AI SaaS applications which are a house of cards in two aspects:
1. new features baked into an LLM can wipe them out in an instant.
2. applying the internet model of funding growth in advance of monetization onto an industry where all software costs money to run, you get the situation best captured in a satirical post on Twitter.

Unfortunately the death zone is where many VCs are investing in AI, because they are mapping the internet investment playbook onto companies that say ‘AI for X’ or ‘AI for Y’ where X and Y and everything else are ultimately owned, or partially owned through token spend, by the LLMs as an LLM is an ‘everything app’.

Cash drying up from either this pyramid scheme of venture backed, AI inference token spend or new innovations that create step changes in efficiency, could trigger a major correction that could spread into AI infrastructure, just as the disappearance of VC backed runway aka ‘burn rate’ caused the dotcom one.

However, no matter how harsh, that correction would not be permanent as the AI revolution itself is not going away. It could end up being much like the gap between the dotcom bubble bursting due to overinvestment while internet users were relatively low and the rebirth as mobile internet delivered them.

Share

Share

Over the next few weeks, I’ll release the principal components of it with some explanation. Below is the description of the Internet component, what the principal innovation the internet brought and how it was different.

The internet created a meta network that allowed computers, computer networks and their users to join a single global one, without permission and the principal thing this delivered was a communication system where there were infinite channels.

In the broadcast era, few people owned the principal communication channels, from newspapers to TV or radio stations, and the dissemination of information was one-to-many. What the internet did was allow for many-to-many communication - anyone could publish a website and anyone could visit it.

In business terms, this transformed distribution or sales & marketing, replacing it with something completely new and far more trackable. The single biggest misconception about successful internet businesses is that there were about new technology - they were mostly about marketing, but marketing that required internet native understanding.

One of the more ironic things I’ve witnessed in my career is that non technical investors in internet businesses tended to over invest in technology, precisely because they were non digital natives and didn’t see the difference between ‘deep technology’ (the kind of thing that comes out of university research and has patents) and tech savvy distribution with commodity technology.

Subscribe now

Most internet platforms were build on deep understanding of internet era distribution and leveraged commodity technology such as open source operating systems, server and database software. On the other hand, most genuinely deep tech companies in the early internet era were based on innovations that were too early to go to market.

In summary, the internet created distribution-side network effects or, more accurately, network benefits. Network benefits comprise network effects and virality, where each has a specific meaning. A network effect is where the value for each user increases with total users (a social network with a couple of users is less valuable for both than one with many) and virality is where each new user brings more than one new user, creating non linear growth.

Since greater value and more users create more awareness and attractiveness, network effects create more virality which creates increased network effects, and so on, for a flywheel of growth. Because it was a communication network innovation, the internet made scientific business growth a reality.

Internet growth metrics could be mapped directly to business model unit economics since increased virality meant that Customer Acquisition Costs dropped the bigger you grew while potential Customer Lifetime Value increased.

I deliberately say ‘potential’ customer lifetime value as realising that value by forcing people to pay up front or early, would increase Customer Acquisition Costs and slow growth, the optimal solution was to delay monetizing users and grow faster than all competitors to capture a market.

The internet created winner takes all markets for those that could delay monetization and focus on growth alone.

The problem with this growth-before-revenue model was that it required subsidy and the problem with that subsidy being in the form of a traditional bank loan was that internet businesses were a race with one winner and lots of losers. This made them too high risk for debt financing as there was too much downside and a fixed upside.

Venture Capital had existed for decades but only really took off as an asset class with the advent of the web. After a false start (dotcom crash) based on insufficient users to match supply, the rise of internet enabled smartphones created ubiquitous internet connectivity and giant businesses. By taking equity, upside was uncapped and this mitigated the downside risk, allowing for the winners, who had spectacular returns by capturing entire markets, to subsidise the losers.

Subscribe now

Just as less tech savvy investors ironically invested in technology, less tech savvy investors often invested in companies that looked like they would at least make some money even if they didn’t aim to capture entire markets with the accompanying significant risk of failure. Because network benefits were at play, Venture Capital was based on returns with Pareto distributions (a form of power law) and successful VC funds only worked if EVERY investment was a potential unicorn.

A back of envelope calculation shows this: a winning investment, which will end up being a less than 10% stake has to ‘return the fund’ (say $100M) so it has to be worth $1 billion+.

Even more dramatic than this is the fact that distribution amongst VC funds themselves follows a similar power law where, as a whole, VC doesn’t really outperform public market indexes where investment isn’t locked in for a decade. The Venture backed investment ecosystem is fractal - the winning companies in the winning funds account for the majority of returns.

Despite these odds, Venture Capital backed internet platforms replaced industrial era firms to became the world’s largest companies, all because of the structural dynamics of network distribution.

As with all progress, there are benefits and downsides, and the early days of the web were full of utopian optimism that was later to be dampened as cyber vulnerabilities, misinformation and tribal political battles were waged on the internet.

This should have been no surprise. The structural change in human communication from one-to-many to more-to-many that the invention of the printing press brought about is far less than a shift to many-to-many. But the printing press’ challenge to the Christian priesthood’s dissemination of information in Europe triggered the Reformation and more than a century of conflict, resulting in the birth of the modern nation state once things reach a stable power equilibrium again.

What we are currently going through, from the rise of populism, culture wars and cyber attacks is a product of the structural changes to the flow of information brought about by the internet and it would be naive to think that the effects will be less dramatic than the Reformation.

However, just like solving for internet growth was based on structural principles of networks, the solution to internet societal disruption is also based on network science.

Over connected networks have a particular characteristic - things spread wherever there is a link. In networks such as this, disease spreads instantaneously and the same is true for memetics to genetics. Ideas in overconnected networks spread quickly and since there are an infinite number of lies and a finite number of truths, there are more compelling lies than truths and lies win.

In terms of network software vulnerabilities there is also a factor of ‘technical debt’ ie lazy coding delivers quick results with long term vulnerable spaghetti code (debt), which when coupled with network interconnectivity means that cyber vulnerability grows over time as the attack surface becomes enormous.

The solution to both problems is in network topology, specifically, a goldilocks zone where networks are neither over connected ot not connected enough, called a small world network. What this looks like is clusters of connection with only a few links between the clusters.

This, more balkanized but still linked, version of the internet than the vast, over connected plains of much of social media is what is necessarily required for a stable societal equilibrium in the network era. It is the network equivalent of the physical topology of the post Reformation city state and hopefully will not follow decades of religious wars.

Subscribe now

Share

If something remains in a state of mystery, we sometimes say that it ‘raises more questions than it answers’. Asking good questions and being comfortable with a never ending production of new ones are at the core of human creative expression and I believe that in the age of AI, the role of humans will increasingly be to ask great questions rather than deliver perfectly formed answers.

This will be the domain of creatives, art rather than craft and ultimately architects rather than engineers. I use the term architect rather that artist in order to differentiate between someone primarily creative and someone primarily execution focused, working in the same project. The difference between an engineer and an architect is that an architect’s purpose is to question the brief itself, not just to make a building stand up but to design something new and original. It’s not always a good idea for a heart surgeon to operate on you in an interesting and unusual way, but it’s good if an architect does this when designing a new space.

The shift away from engineering will face resistance and already is, since for decades engineers and engineering thinking has dominated information technology, with engineers receiving the largest salaries and craft rather than art also dominating supposedly creative design roles such as UX.

Nothing shows this more than the absurd level of geekery and obfuscation that design languages and tools from CSS to Figma have brought to web design, making it harder to produce a web page today than 30 years ago. Today’s UX designers are a masonic guild of journeymen, protecting their craft against art.

Subscribe now

At its most basic level, a prompt is a question and the art of prompting is to come up with interesting questions and use cases that unlock untapped capability. But most of the focus until now has not been around coming up with interesting questions to ask AI models, but has gravitated towards ‘prompt engineering’, which is about how to phrase a question not what question to ask in the first place. I would argue that prompt engineering was the last gasp of engineering culture, desperately trying to find an engineering or craft role in between the creative role of coming up with questions and the automated AI role of answering them.

Prompt engineering was dead as soon as it was born as AI systems themselves have moved towards inference side reasoning, enabling better prompt engineering within AI chat interfaces themselves, from 'single shot’ questions. In response to this, the focus has recently shifted to structured ‘if this then that’ flows for agents, where the deterministic, branching logic of these flows is the bread-and-butter staple of codified logic for programmers. Except that even with agentic flows, we have seen the explosive growth of tools like n8n due to AI produced workflows, directly exported in a format for import.

In summary, engineers are being squeezed out between AI automation and tools that replace the hard-graft, craft element for generating anything new. In some ways this is a continuation of the replacement of painting with photography, where a machine automated the craft element of documenting things, leaving the only paintings with any credibility being those that were perceived of artistic rather than figurative, documentary value. Ironically, the large number of art schools in places like the UK was a product of the army of painters required to document the feats of the British Empire in the age before photography, rather than a particularly aesthetic culture, yet as this became obsolete the schools remained and shifted towards abstract artistic expression rather than craft.

Generative AI is a camera of humanity, a window into the collective memory of human output that is able to remix human culture in any medium. Like a DJ asking herself which tracks to mix rather than producing music, in the age of generative AI all that remains in the question.

Questions are the most profoundly human quest of all and without them AI models sit as inert, lifeless sieves of trillions of ones and zeros with no purpose and no goal. In the age of AI, you may not need to know how to paint or draw, take pictures, write or even how to prompt, but the pinnacle of human existence will reside in being able to ask great questions.

Subscribe now

The first screen based interfaces combined the input and output into the same UI element, with a line at the bottom with a specific character set that prompted you for a command. And there were a bewildering array of command prompts, which reflected the kind of heterogenous design environment which is a hallmark of any early period of technology, from pre-smartphone cellphones to commercial aircraft before the jet age:

Early Operating Systems:

A> – CP/M

B> – CP/M (drive B)

C:\> – MS-DOS / PC-DOS

] – Apple DOS / ProDOS

> – TRS-DOS

1> – AmigaDOS

Classic Unix-Like Shells:

$ – sh (Bourne shell), ksh, bash (regular user)

% – csh, tcsh, zsh (default for some users)

hostname% – tcsh (custom)

# – bash, zsh (root user)

> – fish shell

Microsoft Systems:

C:\> – MS-DOS / FreeDOS / CMD.exe

C:\Windows\System32> – Windows CMD.exe (example)

PS C:\> – Windows PowerShell

> – Windows Terminal (varies by shell)

$, %, etc. – WSL (inherits from Linux shell)

Apple Systems:

$ – macOS Terminal with bash (pre-Catalina)

% – macOS Terminal with zsh (Catalina+)

> – fish shell on macOS

Linux / BSD Systems:

$ – bash shell (regular user)

# – bash shell (root user)

% – zsh shell

> – fish shell

[user@host dir]$ – typical bash prompt (customizable)

Mainframes & Others:

$ – VMS / OpenVMS

READY – IBM TSO (z/OS)

> – RSTS/E

@ – TOPS-10 / TOPS-20

Scripting / Development Shells:

>>> – Python REPL

irb(main):001:0> – Ruby IRB

> – Node.js REPL

*> – Lisp REPL (CLISP)

In [1]: – Jupyter console

Embedded / Networking Devices:

Router> – Cisco IOS (user exec mode)

Router# – Cisco IOS (privileged exec mode)

=> – U-Boot (bootloader)

~ # – BusyBox shell (root in embedded Linux)

This interface paradigm can be summed up in the following schematic.

Apple introduced the first commercial version of Xerox Parc’s mouse & icons, event driven, Graphical User Interface (GUI) Operating system interface in 1983 and it became the paradigm for operating systems, as Microsoft and Unix layered GUIs on top of their command line OS’. The following schematic shows a GUI.

With a GUI, you didn’t need to memorise what applications or what commands were possible as they were displayed in a menu. A button’s text or icon described the capability.

This solved an important UX problem known as discoverability, where messing around led to finding out. In terms of being able to use a computer, it democratised operation, leaving only computer programming to the domain of experts (who persisted with command line interfaces).

When the web was developed on top of the internet it created a graphical application called the browser which allowed documents on other computers on the internet to be used with the same ease as a GUI based operating system allowed access to local ones. The key UI components were the url bar and hypertext links, which allowed any document, stored anywhere in the world to be retrieved directly or from a pointer within another document.

The early web had neither images or forms, but when forms were introduced, desktop applications could be simulated using them and this led to all sorts of software being delivered from anywhere in the ‘cloud’ via a browser. This included browser based access to other internet protocols such as email.

These software applications were dlivered as services within a browser and created a UX paradigm like the below, where the URL bar eventually merged with search allowing for app discovery and signup, where application menus would be rendered within the browser window using the GUI paradigm.

Although the very first web browser allowed for publishing, commercial browsers were only for form filling and delivery of web pages, not publishing them. Another web protocol, FTP was used for loading files onto servers.

An online diary program called blogger was the first to fully merge publishing into the browser, turning the web into a two way, read-write medium. The two-way, read-write web became known as Web 2.0.

An important aspect of blogging was that it consisted of a series of individual posts that could be displayed reverse chronologically into the same web page and this became know as a feed. At the same time, Moreover ( which I co-founded) was know as the web-feed company as it turned updates from thousands or news sources into a single feed, like a weblog, I co-created the standard, RSS, to formalise the structure of these feeds.

With the blogger team (who later created Twitter) we created a project called Newsblogger, with combined searching news links and posting them into a single feed and this became the UX paradigm for social media. Blogging and RSS combined to create the news feed, which when combined with many-to-many user subscriptions created the social media UX paradigm shown in the schematic below.

The next UX paradigm evolved when smartphones changed mouse movement to direct touch and swipe and Apple introduced a new language for online applications which were not browser based and had to be installed via an app store. Although smart phones drove online use from the office and home to truly personal and ubiquitous, everywhere, they were a retrograde step in may ways, gatekeeping and restructuring development and extracting rents for the app store, in Apple’s case.

Even some of the most iconic mobile app inface developments such as swipe left and right for dating apps were actually just copies of existing desktop UX. Hot or Not had developed online dating with the same UX intent (where the choice was combined with the submit action), before mobile.

AIO chat is the next major paradigm UX shift. I believe is still evolving and that it creates an opportunity for a new OS, which I am collaborating on.

The current state of AI chat is in many ways another retrograde step, this time all the way back to the command line, where a text prompt (we even use the same terminology for the text entry as command line interfaces) returns information that is mostly just text.

This is surely temporary. We are starting to see the prompt results become much richer, with image, chart and application generation inline, but the major impediments I believe are not technical but design.

With prompts themselves, prompt engineering veered down a path towards how to ask a question rather than how to discover interesting questions to ask. Not only is prompt engineering an engineers rather than a creative way of looking at the problem of AI chat interaction, it is tackling a problem that AI is already solving itself.

The best prompts now are quite simple, leaving AI to handle how to answer a question. Meanwhile AI chat suffers from the same problem from command lines to Alexa - how can i remember what to ask? Only this time the problem is exacerbated by the fact that AI is capable of practically anything making the task less one of remembering commands but a creative one of coming up with great questions or delivering an interface to discover them and then wrapping the resulting prompts as buttons.

AI buttons are different from, say Photoshop menu commands in that they can just be a description of the desired outcome rather than a sequence of steps (incidentally why I think a lot of agents’ complexity disappears). For example Photoshop used to require a complex sequence of tasks (drawing around elements with a lasso etc.) to remove clouds from an image. With AI you can just say ‘remove clouds’ and then create a remove clouds button. An AI interface is a ‘semantic interface’.

If you then created a marketplace for semantic tools for image editing you would be overwhelmed with options, so there needs to be a way to narrow the options. One way is to scope prompt options based on the active element in a conversation - so if its an image, only image editing buttons (prompts). Another narrowing would be to only show remove clouds if there are clouds in the image, but that still wouldn’t be enough - so there needs to be a core set of actions per object, with the ability to add favorite buttons and to have more appear as you type in the chat bar. So the chat bar shows menu options dynamically as you type.

Some prompts are persistent ones that apply to other prompts, such as what language to reply in and what writing style or format. These define the character of the responses and can broadly be defined as AI personas. The interfaces defining these characteristics (such as GPTs) feel embryonic or following a dead-end UX path.

With agents, which are essentially complex branching prompt tasks, much of the focus has been on creating these branching flows. Again, focus on creation of these structured, if-this-then-that, flows is engineering thinking and can be largely automated by AI itself as witnessed by the rise in popularity of the n8n tool which is in part driven by importing agent flows generated by AI.

I see agents interfaces as moving from the current ones that look like n8n flow charts to something completely different - stored prompts such as ‘book me a flight’. Because there is a branching flow and variables, these agent prompts are special because you might get asked some further questions to have the task fulfilled. The interface for this type of structured conversation, even where the conversation is a one-off one to build and configure the agent, is what is important.

All of these point to prompt discovery and UI as being a core component of an AI OS. For now, I can’t say much more except to point to a specific document type that I believe provides a clue for the way forward.

For decades, computers ran applications than often created documents. But then many academics wanted to run code within their documents rather than use static charts or equations. These ‘notebooks’ were popularised by Jupyter a decade ago, but they can be traced back to Mathematica’s development on code within documents, much earlier. A schematic of the ‘notebook’ UX paradigm is shown below.

I believe that the inversion of the docs-in-apps to apps-in-docs model is another component that is key to unlocking part of what forms an AI OS from a UX perspective and am working on bringing it to reality.

Despite all this, I find Keynote is the quickest and most efficient way to tell visual stories and presentation decks are on the one hand responsible for some of the most artless media in the world, but on the other, they codify storytelling into something that democratises it, for everyone.

I am a big fan of simple structures and formulas for communication, from minimalist Philip Glass’ music, to John Cooper Clarke’s, punk poetry to Ken Burns’ stills based documentaries, Adam Curtis’ text plus archive footage and music overlay, Jonathan Meades’ beautiful architectural criticism or Lawrence Lessig’s keynote talks, punctuated with words on screen in a distressed typewriter font. All of these are formulaic masterpieces.

The formulaic content model I have been obsessed with for decades has been lists and the perfect list can be profound, such as the best of Desert Island Disks. I have spent my life building lists and list tools, from Oobject to RSS, and created the visual bookmarking concept behind Pinterest, as a wishlist tool, WIsts, and now I’ve taken some of my insights from lists tools to presentations.

I like simple and obvious things that look trivial in hindsight, but which everyone needs. And this is why I have built SWYSIWYG.com, a web page building tool in an era where surely web building tools are a thing of the past and nobody needs another one. But I do, and I have built for myself what I always wanted so that I can merge the world of presentations and PDFs with something that can be hosted on the web and indexed by search engines and parameterized by LLMs.

SWYSIWYG stands for Simple What You See Is What You Get. It is a web page building tool that has the ease of creating presentation slides but for HTML. It breaks two dogmas of responsive web page design that I believe have done incredible damage to web design in the name of supposedly good ‘design’.

Instead of breakpoints and responsive elements of a web page that automatically adjust, there are two versions of every page, one for landscape (desktop) and one for portrait (mobile). You lay out elements on a slide, exactly where you want them and they look exactly the same (WYSIWYG) no matter what. This means that you can actually do professional graphic design where text and images are placed in pixel perfect relationships rather than adjusting and rewrapping unpredictably.

This is much like designing a slide in a deck, except that because this is a web page the slide can be infinitely high and its height automatically adjusts as you drag things onto a canvas.

For the moment, instead of building a publishing back end (I may do at some point), you can export the pages as a zip with separate images or as a single file where images are directly encoded in the html.

To get this to work requires some tricky coordinate mapping as I am breaking the way that things have been designed for more than a decade and it is very much a beta. But I produce a lot of decks for my insights work and with SWYZIWYG I will eat my own dog food till it’s fit for human consumption. If you would like to try it, contact me for a password.

The future is bringing the past closer.

1. LiDAR

LiDAR (Light Detection and Ranging) works by emitting rapid laser pulses from an airborne sensor which bounce off the earth’s surface. By calculating the time it takes for each pulse to return, LiDAR generates high-resolution 3D models of the terrain, even through dense vegetation or jungle canopy. When paired with AI algorithms trained to recognize anthropogenic patterns, such as rectilinear geometry, alignments, or elevation anomalies, LiDAR can automate the detection of archaeological features across vast, previously inaccessible landscapes. This fusion of remote sensing and machine learning is enabling archaeologists to shift from exploratory fieldwork to targeted verification, dramatically accelerating site discovery and interpretation.

LiDAR has profoundly reshaped our understanding of the Maya civilization, challenging earlier views that depicted it as a collection of loosely connected city-states in a sparsely populated rainforest. Before LiDAR, archaeologists relied on ground surveys that were limited by dense jungle cover. However, airborne LiDAR scans over the Petén Basin in Guatemala revealed more than 60,000 previously unknown structures, including temples, causeways, defensive walls, and complex agricultural terraces. These findings pointed to a highly organized and urbanized society with dense population centers and a vast transportation network connecting cities. The scale and coordination implied by these features suggest that the Maya were not only more populous (possibly 10–11 million people) but also more politically unified and militarized than scholars had previously thought.

In the Amazon Basin, LiDAR has overturned the long-standing myth that it was an untouched wilderness inhabited by small, nomadic tribes. Previously, academics believed that poor soils and a lack of infrastructure precluded the development of complex societies. But recent scans have revealed geometric earthworks, ceremonial platforms, causeways, and raised fields that extend across Acre, Brazil, and the Llanos de Mojos in Bolivia. These settlements, some dating back over 2,000 years, demonstrate large-scale landscape engineering and organized planning. Combined with discoveries of biochar-like manmade soil capable of supporting intensive agriculture, the evidence indicates that the Amazon was once home to populous, sophisticated cultures whose legacy had been largely erased by depopulation and forest regrowth.

In Cambodia, the city of Angkor, capital of the Khmer Empire, was also radically reinterpreted thanks to LiDAR. Previously, Angkor was thought to be primarily a ceremonial complex surrounded by scattered settlements, and its decline attributed to sudden ecological or military collapse. However, LiDAR surveys conducted between 2012 and 2015 revealed a sprawling urban infrastructure covering more than 1,000 square kilometers, with structured road grids, canals, reservoirs, and habitation mounds. These findings show that Angkor was not just a religious center but a fully-fledged low-density metropolis, supported by sophisticated water management systems that regulated floodwaters and enabled agriculture. Rather than a sudden collapse, the new data supports a gradual decline driven by environmental, economic, and social change.

Elsewhere in mainland Southeast Asia, LiDAR has revealed extensive early urbanism that had been overlooked due to forest cover and later development. Sites like Mahendraparvata and Longvek in Cambodia, once thought to be isolated or peripheral, were shown to have dense grids of roads, embankments, and habitation platforms, dating back several centuries earlier than previously believed. These challenged the traditional narrative that Southeast Asian urbanism was late to develop and heavily derivative of Indian or Chinese models. Instead, LiDAR evidence points to independently developed forms of urban planning, hydraulic engineering, and political organization. These discoveries affirm that early Southeast Asian civilizations were capable of large-scale, indigenous complexity.

LiDAR’s biggest impact lies not just in the discovery of individual sites but since it operates over large areas all at once, its major rewriting of history has been in the reinterpretation of entire civilizational narratives, showing that many ancient societies were larger, more organized, and more resilient than traditional archaeology ever suggested.

2. AI Decryption

AI decryption leverages machine learning trained on large corpora of historical scripts, in order to interpret ancient texts that were previously illegible, damaged, or inaccessible. At its core, it combines advanced imaging with pattern recognition algorithms that can identify ink traces, character shapes and writing structures. By comparing features across known and unknown scripts, AI can infer missing parts, segment text into readable lines, and translate them with increasing accuracy. Crucially, AI can operate on fragile or unopened artifacts such as carbonized scrolls, to virtually unroll them, without disintegrating.

The most famous examples of AI decryption are the Herculaneum papyri, a library of over 1,800 carbonized scrolls buried by the eruption of Mount Vesuvius in 79 AD. For centuries, these scrolls were thought unreadable due to their extreme fragility. But last year, using a combination of CT scans and AI segmentation algorithms, researchers managed to decode previously invisible Greek letters from inside unopened scrolls. They revealed substantial portions of works by the Epicurean philosopher Philodemus, including his treatise On Vices. Fully decoding the Herculaneum library alone could potentially double the number of Ancient Greek or Roman texts.

Beyond Herculaneum, AI is beginning to revolutionize our understanding of ancient scripts and languages worldwide. AI models have been applied to partially decipher Linear B and Mayan glyphs, helping accelerate interpretations that once took decades. In Mesopotamia, researchers are developing tools to scan and classify thousands of clay tablets, some still untranslated, using natural language processing tailored to Akkadian and Sumerian syntax. Similar efforts are underway with Aramaic inscriptions, Dead Sea Scroll fragments, and even undeciphered scripts like the Indus Valley seals, where AI might detect repeated structures or semantic groupings beyond human perception. As training data expands and models become more sophisticated, they could help resolve longstanding mysteries about the structure and operation of early civilizations.

3. Ancient DNA

Ancient DNA (aDNA) is one of the most disruptive scientific tools to emerge in the last two decades, offering window into the deep past that has overturned, revised, and sometimes outright invalidated centuries of archaeological and historical consensus, overnight. Ancient DNA analysis only became viable in the early 2000s with breakthroughs in high-throughput sequencing and contamination control, allowing researchers to extract and analyze fragments of DNA from ancient bones, teeth, and sediments, sometimes more than 50,000 years old. By comparing ancient genomes to modern populations, scientists can reconstruct not just individual lineages but entire patterns of migration, disease and ethnicity. Unlike traditional archaeology, which often relies on material culture or linguistic inference, aDNA offers direct, incontrovertible biological evidence which is frequently a narrative violation.

AI is now essential in ancient DNA research, used to clean and reconstruct degraded genetic sequences, identify contamination, and correct damage typical of ancient samples. Machine learning models detect patterns of ancestry, migration, and interbreeding by clustering genomes and modeling admixture events, even from low-quality data. AI also helps identify functional gene variants—like those affecting immunity or lactose tolerance—and tracks how they evolved.

Ancient DNA is redefining identity. In places like Britain, aDNA has shown that the Beaker people, who arrived around 2400 BCE, replaced about 90% of the island’s existing population (the Stonehenge builders) within a few centuries, suggesting that continuity between prehistoric and modern Britons is far weaker than once believed. In Egypt, aDNA studies from mummies revealed dominant ancestry from the near East (e.g. Mesopotamia) during the New Kingdom. In India, ancient genomes have added fuel to debates around the Aryan migration, suggesting substantial population shifts during the Bronze Age that coincide with the spread of Indo-European languages, challenging both colonial-era and nationalist interpretations.

One of the most profound impacts of aDNA has been the rewriting of European prehistory. For decades, the dominant archaeological model held that the spread of agriculture into Europe was a cultural diffusion: ideas moving from the Near East, not people. Ancient DNA shattered this notion. Starting with studies around 2010, genome-wide analysis showed that early European farmers were genetically distinct from local hunter-gatherers, having migrated en masse from Anatolia. Later, around 2500 BCE, another massive genetic shift occurred with the arrival of Yamnaya steppe pastoralists (from around modern day Ukraine), bringing Indo-European languages, horses, and a patrilineal social order. This genetic turnover was so profound that in many parts of Europe, over 90% of the male lineages were replaced. These findings overturned more than a century of archaeological orthodoxy and introduced a new, more dynamic and more violent view of Europe’s formation.

Even more profound, aDNA has also begun to challenge the model of global human history, challenging the Out of Africa model of human origins in its traditional form and showing it to be much more complex and not a single event in either direction. While the genetic consensus still supports a primary African origin for Homo sapiens, recent sequencing of archaic human DNA suggest a far more complex and messy genealogy than previously thought, more like a web than a family tree. Findings show possible gene flow back into Africa from Eurasian populations and evidence that archaic admixture may have contributed crucial immune and environmental adaptations. The story of human origins involves multiple migrations, interbreeding, and feedback loops.

Most dramatic of all, Ancient DNA has literally reshaped how we understand what it means to be human. The discovery of Neanderthal interbreeding with Homo sapiens was one of the earliest and most paradigm-shifting breakthroughs of ancient DNA research. Until 2010, the dominant view in paleoanthropology was that Neanderthals were a separate, extinct hominin species that died out without contributing to the modern human gene pool. But when a team at the Max Planck Institute sequenced the first draft of the Neanderthal genome, they found that all non-African humans today share approximately 1–4% of their DNA with Neanderthals. This conclusively proved that interbreeding occurred after modern humans left Africa and encountered Neanderthals in Europe and western Asia around 50,000–60,000 years ago. Further analysis also uncovered interbreeding with Denisovans, a previously unknown hominin group identified entirely through DNA, into populations in Oceania and parts of Asia. These findings fundamentally rewrote the story of human evolution, replacing the clean “replacement” model of Homo sapiens’ rise with a complex one of interbreeding and interaction between multiple hominid groups.

The Future of History

AI decryption, LiDAR, and ancient DNA are rewriting history by revealing hard data where only speculation existed. AI decryption is likely to double our knowledge of Roman philosophy and literature. LiDAR is uncovering vast pre-Columbian cities in the Amazon and previously unknown infrastructure around Angkor, forcing a re-evaluation of how complex ancient societies were. Ancient DNA is settling debates on migration and ancestry, showing massive population replacements in Europe and interbreeding with Neanderthals and Denisovans.

The AI future will uncover what is in our landscape, our myths and in our bones. It will bring the past closer, discover lost civilizations and texts, while ancient DNA will create a comprehensive prehistory of the world before writing existed.

Despite having worked in technology for decades, in order to keep my UK architects license, I have to do Continued Professional Development projects, which are now mandatory after a fire in a London tower block in 2017 that killed 72 people eventually led to an overhaul of the fire regulations.

Some of the updates are sensible but others were clearly an over-reaction and even the UK architect’s institute, the RIBA, thought so. Over prescriptive legislation has led to unintended consequences through complex rules that impacted people’s ability to get mortgages, home builders to build to solve the UK housing crisis or the use of timber in construction. The rules had become political, but to challenge them was to face the accusation that you didn't care about lost lives.

As current temperatures in Northern Europe are a massive 50% higher than July average and climate change makes periods of extreme heat the new normal, twice as many people have died in the UK in a single weekend than die in building fires, yearly. But unlike Grenfell, the political pushback is against the architectural thing that could solve it - air conditioning, where building regulations throughout Europe discourage A/C use and in some places effectively outlaw it.

You can’t argue about the fire regulations or you are accused of not caring about people’s lives, but you aren’t allowed to argue about people’s lives when it comes to energy regulations or you are accused of not caring about the planet. As I’ll show here, air conditioning in buildings in Europe should now be a mandatory component of infrastructure, their use can have minimal climate impact and they are actually already in place anyway.

Aside from such extreme cases where elderly or vulnerable people have died during heat waves, there is an estimated 2% productivity drop per degree above 23C. Current regulations throughout Europe limit commercial cooling below 26-28C, which potentially means a 6%+ plus drop in GDP output during hot summer months, even with air conditioning, where it is temperature restricted, since GDP is a direct function of productivity. Even if this is just for a short period, if you consider that peak to trough, the global financial crisis and the extreme recession it caused resulted in a 4.3% drop in US GDP, the impact on economic wellbeing in the continent that is warming most rapidly from limiting A/C, is significant.

If this sounds like an over reaction, consider the case of Singapore. It went from being one of the poorest to one of the richest countries in the world, per capita, and its founding father, Lee Kuan Yew attributed this primarily to one thing: "Air conditioning was a most important invention for us, perhaps one of the signal inventions of history. It changed the nature of civilization by making development possible in the tropics. Without air conditioning you can work only in the cool early-morning hours or at dusk. The first thing I did upon becoming prime minister was to install air conditioners in buildings where the civil service worked. This was key to public efficiency."

As it happens, air conditioning is de facto mandated in new buildings in Europe, as heat pumps are a recommended alternative to fossil fuel sources for heating and a heat pump is just a reverse air conditioner. Most modern heat pumps have a switch for dual use and many systems sold as air conditioning, globally, are actually just rebranded heat pumps.

But environmental regulations for new buildings such as those in the UK specify overall energy use not just climate emissions, so if a heat pump is explicitly intended for cooling then its use could violate the building regulations. The solution is often one of ‘don’t ask, don’t tell’, where heat pumps are installed and calculations are based purely on heating.

For a more extreme example, Switzerland imposes some of the most restrictive, bureaucratic, and arguably Orwellian limitations in Europe on using heat pumps for air conditioning. In cantons like Vaud and Geneva, authorities legally prohibit homeowners from activating the cooling function of reversible heat pumps—even though the hardware is fully capable—unless they obtain a rarely granted, highly bureaucratic “cooling permit” that often requires proof of medical necessity and environmental justification. Installers are sometimes mandated to disable cold mode at installation and cantonal inspectors may audit usage logs, scrutinize smart systems remotely, or claw back subsidies if cooling is detected. Even neighbors can trigger investigations by reporting the hum of an outdoor unit during a heatwave.

The unintended consequence of Swiss Romande hysteria about air conditioning is that people just go out and buy portable A/C units which are far less efficient, instantly negating the whole point of the regulation. The recent spike in portable A/C unit purchases in Geneva—233% in just one month—is especially dramatic, showing a shift from occasional to essential use.

More illogical than all this is the fact that Switzerland’s electricity grid is anywhere between 91% and 97% zero emissions, so running A/C has almost no impact on the climate, as can be seen from the graph below.

The logical fallacy of focusing on energy saving vs emissions reduction, stems from confusing energy with physical resources. Energy is not a finite resource like materials, it constantly shines down from the sun and is necessarily wasted back into space. Consuming energy doesn’t heat the planet, consuming the wrong type does. The environmental argument against consumerism does not hold for zero-emission energy.

The issue is, of course, that an environmentalism based on lowered consumption of everything (even if that consumption, such as Swiss electricity, is environmentally benign), fits a certain world view where the potential for growth is an uncomfortable truth even if it can be proven to be sustainable.

But the problem with this unilateral de-growth stance is that there is only one planet and we all share the same planets weather system, so lowering productivity and reducing energy consumption in the continent that has the cleanest electricity in the world means that markets with far higher electric energy emissions are made more competitive, resulting in a net increase in carbon emissions, globally.

Lowered emissions themselves are still a great idea and the increase in competitive energy use elsewhere (leakage) only reduces their effect by 25%, but when it comes to zero emissions energy, Europe should increase overall consumption of it.

In summary, we need to continue the switch to heat pumps in Europe but support their use as air conditioning in building regulations and permits (with a focus on emissions not energy use) and remove restrictions where the grid is ultra low emissions (such as Sweden, Finland, Switzerland and France).

This will help people, the economy and the planet.

In this newsletter I’ll identify new trends outline future scenarios and propose product, business and investment opportunities based on the macro environment of unprecedented societal and technological change. The insights are aimed at an audience of both investors and operators.

We also offer a service where we combine and reframe them around your business and deliver that directly in an AI model with interfaces to extract ongoing value.

That in turn opens up the opportunity for a deeper engagement around AI era business transformation and the design and prototyping of new products and services as an alternative to traditional corporate venturing.

Every Business Will Have a Website.

In the early days of the web I spoke to a potential investor, suggesting that one day all businesses would have a website and there was a business to be made building them. He didn’t believe this - websites were for geeks.

Despite the need for websites being correct, website building businesses became the domain of agencies and integrators, a largely artisanal rather than scaleable activity. Startup opportunities only flourished later, when the introduction of forms into websites meant you could build browser based applications and capture internet network effects to build end user, platform scale businesses around a website that was an application itself.

Every Business Will Have an AI Knowledge Base

A similar thing is happening now with AI. Every organization and many individuals need an AI model augmented with their own internal documentation, a fancy version of uploading documents and chatting about them with an AI chatbot - an AI knowledge base for interrogation by anyone.

The AI knowledge base is the AI era equivalent of the website, the core around which future AI end user use cases will evolve, but much of their delivery will be an artisanal activity unless it is delivered as a true application.

Here I’ll argue that although much of the AI opportunity is upstream in deep tech and infrastructure and that most venture backed startups will be eaten by ‘big AI’ at the front end, downstream, the value add that makes an AI knowledge base an application is not in the back-end software to process documents and create it, but in the processes and interface around it. This is the one AI wrapper business that really matters.

The AI Era is About Asking Great Questions

Most People Have Faced the Alexa Problem, which is not being sure what you can ask it. This problem is an even bigger one with AI and is often overlooked.

Whereas prompt engineering focuses more on how to ask questions (something which itself is becoming obsolete as AI systems self tune prompts) asking great questions in the first place is where much AI era value-add can be created. Great questions are the simplest, atomic-unit of creativity. As AI shifts the focus of human value-add away from execution to ideas, where questions are the entry points for human creativity, a knowledge base application will center around questions packaged as sophisticated menus and interfaces.

From Answers to Online Agents to Real World Automation

AI’s provide answers based on human questions as prompts. As they evolve from providing single answers, to connecting these to actions in the online word (agents) to actions in the real world (from self driving to general purpose robotics), asking questions becomes more about defining use cases and workflows.

These can all be seen as expanding on the core idea of asking the right questions, and a are a creative challenge rather than an engineering one. Even a workflow can be seen as responses to a sequence of the right questions (is the user in need of x - then y).

In the AI world, everything humans do starts to revolve around asking questions rather than providing answers and in the broadest sense, questioning everything is the domain of creatives and designers rather than engineers and surgeons.

Design Rather Than Build

Just as producing a piece of writing can now be based on delivering the right prompt, building a new application will be about instructing an agent with a novel and well defined use case and creating something in the real word will be about specifying a task rather than executing it. This is why creative industries will prosper. An architect specifies what to build, while the contractor builds it; a product designer specifies a use case and functionality while a software engineer codes it and an interviewer asks the right question while a reporter transcribes answers and events.

The future belongs to great interviewers, designers and architects.

What A Design and Analytics AI Agency Will Do

I previously wrote about how I plan to deliver a new type of strategy consulting based on identifying and combining a spectrum of different narratives to derive insights. This was based on my career as a technology entrepreneur and venture capital investor, where my focus was on business models, trend spotting and financial services knowledge.

Now I’ll describe how that consulting business feeds into a new type of creative design agency, with a core product around AI knowledge bases in order for it to scale. This is based on my previous career as a designer, having originally qualified as an architect and then applying that design training to software products and eventually businesses, through systems design.

Tying the Product Design & Insights Opportunity Together

There are three components to what we’ll do at F2N:

1. A newsletter where I identify new trends, future scenarios and product, business and investment opportunities based on the macro environment of societal and technological changes.

2. A consulting service where we (myself and a collaborators) recenter these insights and opportunities around your business and combine with your own, then deliver in a way you can extract your own insights via custom ‘lenses’, for investors, corporates or deep-tech startups.

3. A service where we use 2. as a starting point for AI transformation to optimise internal operations or design and prototype new products.

Over time, the idea is to scale and productise these offerings, de facto, based on solving real needs. In terms of team, the idea is to draw from a pool of highly creative and experienced collaborators on a project by project basis.

1. F2N Insights: AI Era Scenarios Newsletter

Unique Market Insights from a Technology and Investment Perspective

Our analytics business works by identifying market and technology trends which I will outline in this newsletter and synthesising them into insights, investment theses and future scenarios, for funds, startups and corporations. Together with a broad range of collaborators, our (and that is why I say our) unique value proposition is based on me having sat on both sides of the table as both an entrepreneur and investor, working in internet technology since before the web and with a broad network of collaborators on connections to be able to describe current opportunities for deploying resources or capital.

2. F2N Insights: AI Era Business & Investment Strategies

(a) Insights from your Frame of Reference

To do this means conversations with stakeholders and asking the right questions so that our insights are positioned from your own businesses frame of reference and combined with your own insights. In the past this was done by manually capturing that information in formal workshops and meetings which made the process less natural and serendipitous. We have all experienced where real insights come from: the lobby area of a conference, over drinks after a meeting or on a hike. With today’s AI tools, transcriptions and understanding of these types of interaction can be automatic and seamless.

We will orchestrate natural interactions, automatically capture transcripts and embed them in a dedicated AI knowledge base.

This process is already far more efficient than today’s investment analyst or strategy consulting services, but I believe that in future this will become as standard as having a website and will not be where our real value is delivered. Where we deliver value is in our ability to ask the right questions to garner ‘investable insights’ from the external market or your own organization, internally.

(b) Reporting Through Your Lens

Strategy consultants and analysts typically deliver a beautifully presented report, but this has several problems that make it obsolete. Firstly, reports are often created by more junior people with less experience and so key insights can get diluted. Secondly, the report is often a static document with a finite shelf-life that delivers no ongoing value. Lastly, the report delivers information in one format, delivering the same message to an entire audience.

We will deliver custom interfaces with menus of prompts, personas or agents that deliver real-time updated insights from the knowledge base.

Instead of a single static report you get updated information, delivered through your own lens, where that lens can be at a company, department or individual level.

Again, this software and process is not where the real value add lies, which is in the creative process of crafting what questions to ask to extract the best insights and packaging them as interfaces.

3. F2N Design: AI era Transformation & Venture Design

The process of delivering strategy services as a custom interface to transcripts uploaded to an AI model acts as a go to market to deliver more widespread AI era operational practices within an organization - think of this as the AI equivalent of ‘digital transformation’, where AI transformation revolves around more efficient operational practices around the core AI knowledge base, delivered in conjunction with design and prototypes of software tools around it.

In practice, what this looks like is creating tools and processes for automatic transcriptions of conversations and menus of prompts and agents to extract custom reports.

This allows people to work in an environment where they don’t need to write formal documents as all documents are produced on the fly upon interaction with an LLM and where form filling is replaced by asking questions in natural language from unstructured data, where useful questions to ask are selectable from custom menus.

We will help you combine your own information with an LLM, with the assumption that the best way to do this will change over time and eventually commoditize into a standard offering that we will enable.

We will then help you build bespoke menus of the right questions you can ask of your information tailored to individuals and departments.

And from this starting point we can enable an organization to work around the core knowledge base in an environment largely free of documents, forms and excessive bureaucratic process, to become much more human, enjoyable, creative and efficient through a process of AI era transformation.

From this starting point our offering can extend to designing and prototyping new products for internal efficiency or new customers.

As AI makes software development much easier, applications that weren’t previously viable become possible and more and more niche applications can be built, the focus switches to finding creative use cases and knowing how to design and interact with rather than build them.

Winning in this space will not be a small opportunity, and the gap exists for something like an IDEO in terms of creative design meeting McKinsey in terms of intellectual analysis. That’s what I hope to help build at F2N.

Last week, my friend Nicolas Colin produced an excellent essay, titled “Late Stage Investment Theory” which delivers a thesis for the structural changes to technology investment from Private Equity to government funding, over the coming decade. I have recently written an oped on the changes to Venture Capital for the asset management publication, Citywire, and I was interviewed about them in last week’s Financial Times piece on the same subject.

These things are very important as they will affect global competitiveness in the age of AI.

I agree with a lot of what Nicolas says, but I have a different view of the underlying framework, and this changes how I think it will play out and how capital should be deployed. This post outlines a different framework at a high level.

Carlota Perez’s Phases of Innovation

The framework that Nicolas uses is based on Carlota Perez’ work which she described in the book “Technological Revolutions and Financial Capital”. This became highly influential in Silicon Valley and underpinned several VC firms’ theses. I’m pretty familiar with Perez and her ideas as when I was at the VC firm, Anthemis, we sponsored her ongoing work.

To briefly summarize the Perez model and where it came from, in the 1920s, economist Nikolai Kondratiev identified regular cycles (K-waves) of roughly half a century, in periods of industrial era innovation.

At the turn of the millennium, when the internet revolution was just starting to go mainstream, Perez took Kondratiev’s waves and broke them into two periods of two phases each, allowing for some overlap. This was based on integrating some of Schumpeter’s ideas of the dynamics of innovation & creative disruption and their financing into Kondratiev’s model.

Perez then mapped these onto historical phases (although there is a lot of shoehorning or selective categorization to make them fit, in my opinion):

Colin’s Late Cycle Investment Theory.

Based on Perez’ framework, Nicolas Colin argues that we are now in the latter phase of the digital revolution - the deployment phase, and that is why structural rifts are starting to show in both Venture Capital and Private Equity and why money is increasingly flowing to deep tech and information technology infrastructure projects (AI models, chips, data centers and electricity) that require much larger pools of capital and have different risk profiles.

This shift in capital flows is real and provable and therefore Nicolas’ theory is important, but I believe the reasons for it are different. Although he acknowledges that the deployment phase of a Perez cycle is merely a lens to extract insights from and doesn’t need to be based on hard ground truths, I also believe the lens is flawed and therefore the theory can be refined.

An Adjusted Framework

The framework I propose has five differences:

1. The AI wave is not an extension of the internet one, it makes more sense to look at it as a new one and the overall AI, internet and pre-internet phases of information technology innovation are different from the industrial era cycles.
2. The reason for the shift upstream towards infrastructure and hardware financing is to do with the underlying differences of AI (specifically a shift in network effects from distribution to production).
3. These shifts towards digital era manufacturing mean that the Chinese manufacturing model, which is post industrial and coupled with software, wins.
4. The regular cycles of k-waves are fundamentally wrong, the cycles vary. In the digital era the cycles shortened compared to the industrial age and in the AI era the rate of shortening will accelerate.
5. Overlapping cycles are combined to create something that does not have any overall regularity, but decomposing it to all sorts of different cycles based on different trends and narratives, proves you have the right overall model.

1. AI creates a new major cycle

AI software is different, there aren’t really any ‘algorithms’ at all, in the traditional sense. There is no codified ‘IF This Then That’ logic and the outputs of AI models are statistical, emergent qualities of a network, not deterministic calculus. This fundamental software paradigm shift is why we are in a new wave of innovation as different from the age of steam vs Fordian assembly lines in industrial eras.

2. The shift towards infrastructure and hardware financing is because of AI not a digital era deployment phase.

With AI, the software itself is a network where the software writes software, to the point where this is potentially a runaway effect when AI models build their own successors (the singularity). What this means is the network effects are in the production side since as software writes more software there is a flywheel effect where the cost of production drops (NB, the cost of production drops but the cost of running software, the AI token cost, is significant and this disrupts things like startup freemium models).

The business models of AI are fundamentally different from the internet which has distribution network effects, so the opportunities are at the opposite end of the market. It means that value will flow to deep-tech startups (things based on real science and IP vs distribution advantage), incumbent tech platforms, hardware and infrastructure providers. We may see Corporate Venturing arms of large technbology companies eclipse traditional VC and that strategic financing and possibly pre internet style corporate labs (a new era of Xerox Parc and Bell Labs style entities) will return.

AI companies that are wrappers around LLMs, financed by Venture Capital firms that are applying the internet investment model to AI, will perform badly and cause the number of VC firms to shrink.

In public markets, we have already seen that most of the recent gains in the S&P have been accrued by the tech platforms that are also AI infrastructure providers. A dotcom style correction with continued long term growth could repeat if there is infrastructure over-build, but unlike the dotcom era which collapsed as there weren’t enough users, AI growth seems more robust there. A correction is more likely to be driven by geopolitics and the transition from AI to robotics, where China will have an unfair advantage.

3. The Chinese manufacturing model wins.

This component of the framework is the same as Nicolas’, however I am including it as an item in the amended framework, to illustrate that it appears for different reasons.

China dominates manufacturing, but specifically it dominates manufacturing of the types of hardware that have embedded software as a core component - these include EVs, drones, smartphones etc. and they form the principal components of the commanding heights of the post industrial economy.

Because manufacturing can never be near zero cost (as it requires physical materials) but software creation in the AI era potentially can, a new model has emerged where software is given away to sell hardware. For example, BYD gives away its self drive software, potentially undermining Tesla’s ability to sell self drive subscriptions and DeepSeek open sources LLMs, copying Meta’s playbook to undermine OpenAI (which is turn was copied from Microsoft giving away the browser to undermine the threat from Netscape in the early days of the web). For Apple, a supply chain management and product design company that is struggling to do either and who have charged a bigger margin than the manufacturer, this model is potentially very threatening - if AI removes their OS moat.

The digital era manufacturing model undermines the digital era services one which in turn created venture backed companies that replaced industrial era ones as the world’s biggest (except for in Europe). In other words, US software was ‘eating the world’, now Chinese hardware is. And just like US digital era software services ate European industrial era services, Chinese, digital era manufacturing, will eat both US and European industrial era manufacturing unless they become highly automated with significant software sophistication.

When this software enabled hardware extends to AI, you get self driving cars and ultimately, general purpose robotics which extend AI agents from the online to the real world. And humanoid, AI robots can be trained on videos of people to do what people do, so they have a bigger addressable market and training network effects. As such, AI robotics is predicted by many (such as Nvidia’s CEO) to be the biggest industrial sector ever, with a market greater than $50 Trillion. The AI arms race will therefore expand from software to hardware and to geopolitics.

4. The regular cycles of K-waves are fundamentally wrong.

Not only is AI a new wave of digital innovation within the prost industrial period, the previous digital era waves were fundamentally different from the industrial era ones.

The PC era lowered the marginal cost of production to zero (an extra Windows CD cost Microsoft pennies) and the Internet era lowered distribution (sales and marketing) costs to zero. Network effects (value to all users increases with number of users) and virality (new users bring more new users) meant that things grew fast and captured entire markets.

Unlike in the industrial era, growth was not linear and so the K-waves were shortened - the deployment phase of the internet isn’t coming only now with AI, it’s been here since the major platforms (GAAFA or Mag 7) came to dominance at the top of the internet food chain, many years ago.

5. Cycles are invisible till they are decomposed.

Music waves and stock market charts in the real world, rarely look totally regular and looking for patterns in the latter is often a fool’s errand. Imposing those patterns on real trends is similarly dangerous.

But there is a magical property of messy waves - that they are based on overlapping simple waves that can be mathematically extracted through a Fourier transform. In other words, in the reverse process, if you have an economic model of complex overlapping cycles you can in principal know that your model is correct if their overlap produces cycles that approximate to what is observed in real world messy things like markets.

The idea of “letting the data tell you” what cyclical components live in economic or financial time‐series via Fourier (or related) transforms actually goes back almost a century, where econometricians developed spectral analysis to see which frequencies show up most strongly in the data. Jan Tinbergen was among the first to apply these ideas to business‐cycle data in the 1930s, winning a Nobel for dynamic models of the analysis of economic processes.

Today, many hedge funds use a signal processing approach to quantitative analysis and these emerged from AHL’s development of systematic trading in the 1980s.

Rather than impose “clean” 50 year K-cycles or fixed Perez waves, these firms let the data reveal which frequencies carry the most variance—and then design filters or strategies around those empirically observed cycles.

The difference, however, is that this data is either quantitative data for quantitative analysis or qualitative data for quantitative analysis

There aren’t, to my knowledge, any purely discretionary or thematic hedge funds that take Perez’s narrative-driven waves (e.g. “Internet deployment,” “biotech boom,” etc.) and run a straight Fourier-or-wavelet decomposition on those qualitative time-series. Instead, what you do see is quant shops turning qualitative proxies into numbers.

Firms like Man AHL, Two Sigma, D.E. Shaw or Renaissance will pull in alternative data—Google Trends on “quantum computing,” Wikipedia page-views for “AI” and then treat that as a time-series. Once it’s numeric, they apply exactly the same spectral tools (DFT, band-pass filters, wavelets) to hunt for dominant cycles or regime shifts.

No major discretionary thematic fund publishes a “tech-cycle spectrum”, we plan to. Managers like ARK Invest, Global X, or thematic private-equity firms still base their cycle calls on fundamentals, valuations, adoption curves and expert calls—not on spectral analysis of cycles.

Conclusion: ‘Narrative Spectrum Investment.

The current combination of geopolitical and technological shifts are on a scale that we haven’t seen in centuries, so the underlying thematic trends dominate everything in terms of investment. There is a shift from quantitative to qualitative corporate strategy or investment advantage.

At the same time, fundamentals don’t matter as much or don’t provide opportunities for ‘alpha’ (investment advantage). Sell side analyst reports are often low quality even if packed with data, high frequency trading is a race to the bottom and narratives both real and imagined (meme trading) dominate both public and private market investment.

Rather than imposing a single frequency as a lens to extract a narrative as Perez does, analysing frequencies to extract narratives and their degree of influence, potentially allows for a degree of objectivity in thematic investing.

Making sense of this through signal processing approaches to qualitative thematic information by spotting trends and opportunities for everyone from Private Equity to hedge funds would be a core component of AI era rather than late stage, investment theory based on a narrative spectrum rather than k-waves.

In other words, if you find the overlapping narratives that constitute a model that is backtestable to some ground truths, you can predict future scenarios and invest in them. If you understand the spectrum of both fictional and real narratives you can predict things.

We will attempt to deliver the ‘tech-cycle spectrum’ and what will follow on this substack will be narratives that describe current trends and future scenarios that we will attempt to piece together into a combined model, over time.

For the moment the substack will be free but access to the overall data, repurposed and combined with your own, in an AI model with interfaces designed by us, can be provided on a retainer basis. If you are interested, ping me: david@f2n.co