Notes from practice
Blog
We write about what we build ourselves: where document search systems break, how a knowledge graph differs from a vector database, when several agents do worse than one. No trend round-ups — only mechanisms and what we have checked by hand.
AI Engineering
RAG system: what it is, when you need it and where it breaks
RAG is a way to make a language model answer from your documents rather than from its memory of the internet. The idea is simple and the failure is predictable: almost every problem in these systems arises not in the model but in how the documents are chunked, retrieved and passed to it.
Knowledge Systems
Knowledge graph or vector search: which one fits your task
Both approaches answer the request “find me what is relevant”, but they understand relevance differently: vector search looks for what is similar in meaning, a graph for what is connected by structure. Confusing the two costs projects months of work.
Agentic Systems
When a multi-agent system loses to a single agent
A multi-agent design looks convincing on a slide: a team of specialists instead of a lone worker. In production it regularly turns out slower, more expensive and less predictable than one agent with good tools. Here is where exactly the line runs.
AI Engineering
LLM hallucinations: the mechanism and the engineering measures against them
A hallucination is not a model failure but the model’s normal mode of operation, carried through to a result that is inconvenient for us. Understanding the mechanism matters more than any list of “prompts against hallucinations”: it shows which measures work and which only create a feeling of control.
Agentic Systems
AI agent: what it is and what turns a system into an agent
Almost anything with a language model in it gets called an agent — from a script with a prompt to a system that runs a process on its own. The difference is fundamental: it decides which tasks the system can be trusted with and what control it needs.
AI & Business
How to calculate the ROI of AI implementation so that the numbers survive scrutiny
Most AI payback calculations fail a single question from the finance director: “What exactly will change in our costs?” Here is a method that survives that question, and the mistakes that inflate the result several times over.
AI & Business
Business process automation with AI: what to take on and what to leave alone
AI does not belong everywhere there is manual work. Half the processes people tackle “with a neural network” are covered more cheaply and more reliably by ordinary rules. Here is how to tell one from the other before the budget is spent.
Metaintelligence Architecture
Metaintelligence: why the next level is not the model but the environment
We are used to measuring progress by the size of the model. But a system that carries work for months does not consist of a model: it has relationships between participants, limited resources, a way of reaching agreement and a trail from which any decision can be taken apart. This series is about how such a layer is built and why it sets the ceiling of what is possible more than the next generation of weights does.
Metaintelligence Architecture
The agent as a bounded local world
In most systems an agent is a function with a prompt. Such an agent cannot refuse work, cannot come to an end and cannot run out of resources. Yet it is precisely being bounded that makes a participant fit for an environment where nobody controls everyone.
Metaintelligence Architecture
A link as a first-class object, not a message queue
In most systems the link between participants is invisible: there are calls, there is a queue, there is a shared channel. Make the relationship an object in its own right, with its own state and history, and things appear that do not exist otherwise: the ability to reach an agreement, revise its terms and part cleanly.
Metaintelligence Architecture
Living graph: when topology is a consequence of work, not a design
An architecture diagram goes out of date the moment it is drawn. A living graph is an attempt to close the gap between the diagram and the system: links here are not designed in advance but arise from the participants’ local decisions, and disappear when they stop being useful.
Metaintelligence Architecture
Consensus without a centre: deciding together without taking anyone’s word for it
As long as a system has a master node, the question of a shared decision does not arise: the master decides, the rest carry it out. As soon as there is no master, or it cannot be trusted unconditionally, a problem appears that ordinary agent set-ups simply do not deal with.
Metaintelligence Architecture
Agent economics: why autonomy needs a budget
The most reliable limit on behaviour is not an instruction in the prompt but a resource that runs out. When every action and every thought is debited from an account, “don’t do anything unnecessary” stops being a request and becomes the physics of the environment.
Want your task looked at?
Describe what is not working or what you want to build. We will answer on substance — and say so if the task has a simpler solution.
Write to us