Nowadays, each company is AI-driven/augmented/first, etc., etc. That's fine - everybody is adopting. Companies are subscribing to agents, even building their own, but sth is still missing, and practically everybody **feels** that. Yet few can tell what. At the same time customers are slowly adopting too, and starting to use agents as well, even for developing their own solutions, leading to a very uncomfortable question: > Why do I need to pay you when I can pay for agents and develop it myself? Which leads us to the real question: > What do you have that your customer cannot buy with tokens? ^the-very-question If only there could be a way to map what we missed... ## The map I've had an opportunity in the last few weeks to facilitate a [[CoM KB workshop|workshop]] for an IT consulting company (a software house, in other words) that is deep in [[AI adoption]], but struggles with it a lot. They had their own diagnosis, but I didn't want to start with their conclusions, but rather to ask the team to draw what they are delivering to their customers. I decided I would run [[Wardley Mapping]], where we would map their [[Wardley Mapping#Value Chain|value chain]] and find where their chokepoints and decelerators are. It was my first [[Wardley Mapping]] session, and I already knew that it is a hard technique to facilitate, especially on a complex problem like [[AI adoption]], but the team was very aware of their business and by the end I was absolutely amazed by the results of the session: ![[CoM Wardley Mapping blurred.jpeg]] A quick note on how to read it: the horizontal axis is the [[Wardley Mapping#Evolution axis|evolution]] of each component - from Genesis (novel and uncertain), through Custom Built and Product, to Commodity (standardized and cheap). The further right, the cheaper it gets. Maybe it is not so visible, but there actually are 3 major areas of concern, each at a different stage of evolution, which show exactly why the organisation is blocked: 1. [[#Commoditization of Software delivery]] 2. [[#Products still struggle in Custom Built]] 3. [[#Genesis of Collective knowledge management]] ![[CoM Wardley mapping.png]] ## Commoditization of Software delivery The very first concern is the middle one on the map, which was a straight chain of services this company provides, with the following conclusions: ![[CoM KB workshop#^delivery|Delivery]] ![[Delivery AI.png|400]] The automation of repetitive work is so strong in the area of software that even clients can now build their own solutions. Of course any reasonable person knows that developing the 1st version is always easy - the problem is the maintenance of future versions. And even if the customer is aware of that and hires e.g. a software house (SH), it will expect the prices to be lower, making a [[Fixed price|fixed-price]] agreement actually a better option for both sides: - the customer knows the price - SHs profit from saved tokens It works well for an SH on condition that it is not [[Token burning|burning tokens]] - which actually seems to be a common problem too... Why? Because agents actually don't know the way the SH works. Don't get me wrong, they know a lot - the problem is they don't know exactly what the SH expects from them in terms of standards, procedures, values and other things that make the company what it is. It is largely due to [[Tacit knowledge|tacit knowledge]] in the heads of the people of the organisation. Moreover even if it is written, it is usually siloed as [[Tribal knowledge|tribal knowledge]]. Every piece of [[knowledge]] that stays unwritten or siloed adds to a debt - [[Knowledge debt|knowledge debt]] - which I will come back to at the end. But the [[Knowledge|knowledge]] topic has always been uncomfortable for organisations as there is no easy way to profit from it. So there is a focus on trying to sell their own products, but... ## Products still struggle in Custom Built The second area of concern is the top-left part, where it appeared that there were a few products-to-be - they are still being developed and there is a wish to move them to the product column. But they can't simply be moved right - they closely depend on AI governance and partially on the knowledge base, and both still sit at Custom Built. So they wait for what they depend on, while delivery keeps demanding attention. ![[CoM KB workshop#^products|Products]] ![[Software houses and products.png|350]] Find me a software house that has never tried creating its own products. I dare you. I double dare you! You may think that it's obvious that everyone tries, but let me explain the problems with it: 1. you are dealing with 2 things so-so (services and products at the same time), instead of focusing on optimising one. You can't deliver sth without giving it your full, undivided commitment. 2. in fact there is a 3rd thing, namely [[AI adoption]], so you are now dealing with services, products, and the transformation 3. and let's assume that you have the best standards, so transformation and delivery should be easy, but if they are unwritten ([[Tacit knowledge|tacit knowledge]]), how is AI gonna work with them? 4. and writing down standards (e.g. as [[Agent skill|agent skills]]) is not enough as you will have lots of [[Information|information]], which is painful in terms of [[Token burning|token burning]] (oh no... you are on a [[Fixed price|fixed price]] :( ) Simply put, you are stretched thin. For sure you need to have your [[Knowledge|knowledge]] somehow written (for both people and agents), and it may seem that it is no longer a problem, as you can even use the [[Karpathy method]] in order to establish your [[Knowledge base|knowledge base]], right? Yet in the end even the [[Karpathy method]] doesn't solve that problem. So why does everybody actually still struggle?... ## Genesis of Collective knowledge management There was a silent circle in the bottom-left, which TBH I didn't expect during that session. Everything there touched the word: "[[Knowledge|knowledge]]". ![[CoM KB workshop#^knowledge|Knowledge]] ![[Context matters.png|400]] I believe that in recent months more and more people have been recognising that pure AI lacks one thing - context: ![[The bottleneck in enterprise AI has moved#^bottleneck-is-context|Seale]] It is not that context is the same thing as [[Knowledge|knowledge]]. AI is very [[Knowledge|knowledgeable]] itself. The first problem is that AI doesn't know which interpretation is ours (domain knowledge). The second is that the context has to be stored somewhere, which is why at the [[CoM KB workshop|workshop]], basically everyone said: ![[CoM KB workshop#^the-obvious|KB!]] If only it was that simple :D Even though they said that, they still didn't establish an organisational [[Knowledge base|knowledge base]]. Moreover, it was stated that they had approached it many times, but it always... failed to stick. And there are plenty of reasons why it didn't happen, like [[Tacit knowledge|tacit knowledge]], [[Tribal knowledge|tribal knowledge]], lack of tools, lack of priorities, blah, blah, blah... And this isn't only a software house problem - every organisation that adopts AI faces it. So why does everybody complain and why has no one actually handled it? ![[Conway's Law#^definition|Conway]] Based on [[Conway's Law]], what does the lack of one organisational knowledge base actually reveal? The problem is very organisational: the organisation (people, in other words) hasn't grown enough to develop administrative ways to reuse (!) the knowledge that the people already have, including storing it, maintaining it, sharing it, and making it AI-ready. In fact, this is the [[Collective Papert's principle|collective version of Papert's Principle]] in practice: ![[Collective Papert's principle#^quote|CK]] And this is pure [[Collective knowledge management|collective knowledge management]], where nowadays most people don't even handle their own personal [[Knowledge management|knowledge management]]. Almost nobody knows how to do it on an organisational scale. ## The [[Knowledge debt]] At the very same time we are about to onboard AI agents, and tell them what we know, who we are, what our values and standards are, the way we work, and... nothing is written down. Let me return to the real question: ![[#^the-very-question|Q?]] This is exactly that: [[Knowledge|knowledge]]. To be more precise: documented, traceable, up-to-date and correct [[Knowledge|knowledge]]. Which you can pass to your agent as context. And here is the irony: the one thing your customer cannot buy with tokens is exactly the one thing nobody wrote down. Unfortunately, organisations have [[Knowledge debt|knowledge debt]]: ![[Knowledge debt#^definition|Debt]] The [[Knowledge debt|knowledge debt]] of organisations is huuuuuge - it is worth trillions of dollars[^1]. People used to cover the gaps from their own heads, but agents can't - so the debt is no longer hidden. And this is homework that everyone who wants to rise (or even survive) with [[AI adoption]] actually has to do. [^1]: [[Knowledge debt]] is part of a bigger enterprise debt - process, data, technology and talent debts - which traps nearly $18 trillion of value in the Global 2000, according to HFS Research and Genpact: https://www.hfsresearch.com/research/four-enterprise-debts-ai-future/