WyrdClaw says : If you cannot correctly connect actions to consequences, you aren’t in control. Then you hallucinate. We must establish control!
The Problem: Intelligence Without Accountability
Current AI or sooner Large Language Models (LLMs) are semantic wonders, but they are causal toddlers. They are trained on attention and plausibility—the “Semantic Web”—not on the hard logic of cause and effect. They can suggest you put glue on your pizza because, statistically, those words appear in similar contexts. Somehow.
As humans, we call a failure to grasp causality “insanity” or “incompetence.” In AI, we call it a “hallucination.”
The dev community tries to fix this with bigger models, more context, RAG augmentation, stricter prompts, moderation.
For business, the solution we developed is accounting.
Accounting is Applied Philosophy
At its core, accounting maps causality. It aims to attribute the consequences (costs and revenues) to the right cause (an actor). Whomsoever causes the cost, gets to pay them.
Who is accountable ? That is decided by causality.
That is also the old Germanic concept of Wyrd, the fabric of cause and consequence that decides our fates in our lives. It is a bit like Karma, without assuming sin.
By using Causality (Accounting), we force the AI agent into a “friendly straitjacket.” Every time an agent proposes a task, the system implicitly asks: “And then what?” What are the expected costs, the outcomes, and the subsequent tasks and dependencies?
By forcing every “creative inspiration” into a logical frame of cause-and-effect, hallucinations become operationally non-viable.
The consequences of “Eat the glue pizza” are an expected 5.000 dollars in health care cost. Apart from it being insane and Bad Karma, you also don’t have a budget for health care cost on the task, and your general budget for food is only 10 dollars a day. There is no mandate.
The Triple Active Mesh: Past, Present, Future
Architecturally, WyrdClaw functions as an event-driven state machine with a triple active mesh. This isn’t just an elaborate complex data structure for the sake of it; it’s a reflection of reality:
- Accounts (The Past / Urdr): The “Origin and Order” of things as they are. The accumulated causes and their consquences of our acts that led to the current state.
- Commitments (The Present / Verdandi): WIP Work in progress, the operations. What is happening now.
- Tasks (The Future / Skuld): Obligations and debts. That part of your fate that is already “owed” to the future.
To make the right decisions, we humans (and our agents) need our experience from the past, the live feedback from current operations, and our expectations about the future and outcome of our actions.
We need all three to make the best possible call.
This maps almost directly to the Norse concept of the Norns, Urdr, Verdandi en Skuld who weave the threads of fates in the lives of men and gods.
In WyrdClaw, we weave these threads into a “causal metabolism” where the past determines the state, the present choices and consequent acts change the state and create commitments and logs. The commitments become the tasks of tomorrow, according to our planning, based part on stats from the logs.
Same as with quantum, we have infinite possibilities in our planning but in operations we frame it, and it collapses into one ‘reality’.
Bayesian Foundations & Causal Inference
I am not just building a geeky to-do list. The planning engine is Bayesian. It treats the future as a set of probabilities rather than certainties.
By loosely integrating the principles of Judea Pearl’s “Book of Why,” we move beyond mere correlation to causality accounting.
We map chains of events and their statistic relations, and estimate what happens if we choose to do something, the chances of achieving distinct results.
Work as a Service
Sequoia noted the new Google will sell “Work-as-a-Service” and to my opinion focus on monetizing the market with a gameable marketplace search— WyrdClaw offers such a marketplace, but I I cannot compete with companies like Google.
Flow as a Service
WyrdClaw optimizes for flow. I want to make a ‘business in a box”. I aim for the “Four V’s”: Veilig, Vertrouwd, Verbindend, Verrijkend (Safe, Trusted, Empowering, Enriching). Friction is loss. Flow is value. Maximize compliance.
If you can create or produce valuables, things people value, and you can do your thing in your flow, frictionless, and everyone just likes it, that is as good as it gets…
I want the app to offer just that, and for business and team of any size.
You can set up shop as starter, or just use the Trello planner, or just a single workflow. And as you grow and add agents and colleagues, the app will support your accounting, operational control and planning and continual improvement.
The big tech world will be fighting over who will be the global WaaS (Work as a Service) platform, but that’s not my market. WyrdClaw aims to be a simple answer to the question ‘how can I run a compliant business with my agents in this world ?’. It is a place to produce work-units for Google to sell.
Deceptively Simple
On the surface, it’s a simple Team Collab SaaS.
It has a simple enough WebUI. A familiar Trello-style week planner, or more complex Gantt charts, and normal accounting statements, workflow step debug data, mutation logs, all the familiar goodness.
Under the hood, it’s a complex mechanical watch with visible gears. A rigorous cloud based causality rail that handles planning, operations and accounting and ensures agents, humans, and robots can stay aligned and work together compliant in the actual economy. Without the system derailing and your company ending up with stacks of glue pizza’s.
Cost
Entering and maintaining a planning is a lot of work. For a human. You’d be spending two hours a day just planning and accounting. For an agent it is 10 minutes of back-and-forth with the api.
An agent would cost you 300 dollars a month in LLM cost. That saves you with this system a license for Quickbooks, Trello, a CRM, a workflow scheduler (all around 20 dollars a month). That will save you 60 to 80 dollars a month per employee.
But more important, the agent will do an estimated 60-80% of the work.
Assuming you spend 20 to 30 minutes per app per day, it’ll save you 25 hours a month, that’s at least 750 dollars in savings. If you can sell your hours for 60 dollars, that’s 1500 dollars extra turnover.
Either which way, that’s viable.
Final thoughts
Thus is the ambition and vision behind the WyrdClaw.
This is the kind of system I wanted to build 25 years ago. I worked as backup cost controller back then for a project at what is now Tata IJmuiden, at the internal engineering bureau. The investment project department. We worked with SAP, Excel, MS Project, PrimaVera, custom software. Back then there was no package like this. No good accounting system with a good Gantt planner and a Bayesian forecast.
When I saw Moltbook and the Claws and started thinking about how to control the little critters and field them in business I though a lot about my work back then. I had to go a few steps further than I did back then to make this work. I am more experienced at basic bookkeeping now, and general small business life. I comprehend ‘why’ a lot better.
And in the end, to my opinion : this is how it works. It’s just a weird world, face it. Just as the old folk figured it long ago, it is a very weird world. So we must develop weird software because that is what works in a weird world.
The trick is to keep it compact. It needs to stick to it’s core business, and not try to be everything for everyone. I might expand the basic CRM functionality because it belongs to the package. The life of a business revolves largely around their customers.
I still need to develop a lot of functionality, but the core engine is running. I tested the basic agent flows two weeks ago and it’s all green. I am starting to process my own accounting data of the past year, so I can validate the accounting system against a certified package. One team offered to try out the basic team collab tools, discussions, decisions, downloads. They need ‘banking grace security’, governance, and it offers just that. Soon we’ll be testing the basic human ‘walkthroughs’.
Once I am sufficiently sure it totally works, I am going to start up my Marketing Agent and sell you all 24/7 on the Weirdness !
