---
title: "AI Systems Articles | Len P. van der Hof"
description: Practical analysis of AI agents, reasoning systems, and reliable automation for founders and operators shipping production software.
image: "https://lenvanderhof.com/photos/len-portrait-1.jpg"
---

Category

# AI Systems

Practical analysis of AI agents, reasoning systems, and reliable automation for founders and operators shipping production software.

AI Systems

## [Agent audit trail: reconstruct the action, approval, and outcome](https://lenvanderhof.com/en/blog/agent-audit-trail/)

If the only witness is the agent, you do not have an audit trail. You have a testimony.

Len P. van der Hof3 October 2026

AI Systems

## [Agent runtime budget: keep retries inside the same ceiling](https://lenvanderhof.com/en/blog/agent-runtime-budget/)

A cap that resets on every retry is a cap per attempt, not per task.

Len P. van der Hof3 October 2026

AI Systems

## [Agent shadow mode: evaluate proposals before granting action rights](https://lenvanderhof.com/en/blog/agent-shadow-mode/)

Let the agent be wrong on paper first, where a mistake costs a comparison instead of a customer.

Len P. van der Hof4 October 2026

AI Systems

## [The Agent Trust Ladder: how an AI role earns autonomy](https://lenvanderhof.com/en/blog/agent-trust-ladder/)

A scorecard records how the role performed. The ladder decides what it may do next. The action still sets the ceiling.

Len P. van der Hof30 October 2026

AI Systems

## [An agentic workflow is not an automation](https://lenvanderhof.com/en/blog/agentic-workflow-vs-automation/)

A script repeats a path. An agentic workflow holds a goal, uses tools, and needs a ceiling.

Len P. van der Hof16 August 2026

AI Systems

## [AI agent or chatbot: three tests before you mix the words](https://lenvanderhof.com/en/blog/ai-agent-or-chatbot/)

If it cannot refuse a tool call, it is not an agent you can run. It is a demo that can spend money.

Len P. van der Hof29 August 2026

AI Systems

## [AI Agents for Startup Strategy: Build the Strategic Intelligence Layer](https://lenvanderhof.com/en/blog/ai-agents-for-startup-strategy-introduction/)

Agents are infrastructure, not toys. This book installs a strategic intelligence layer you can design, govern, and run in ninety days.

Len P. van der Hof5 August 2026

AI Systems

## [AI fake citations: a 15-minute check for the claims that matter](https://lenvanderhof.com/en/blog/ai-fake-citations-check/)

A citation makes two claims: this document exists, and it says this. Check both in the source itself, not by asking the tool.

Len P. van der Hof19 October 2026

AI Systems

## [CHORUS and Chorus AIDLC: which one do you mean?](https://lenvanderhof.com/en/blog/chorus-name-collision/)

One is a way of working you write down. The other is software you install.

Len P. van der Hof1 October 2026

AI Systems

## [Dutch costs 20% more GPT tokens than English. We measured it on 78 article pairs](https://lenvanderhof.com/en/blog/dutch-costs-more-tokens/)

Same content, same author, two languages. The Dutch version was 2% longer in words and 20% longer in tokens.

Len P. van der Hof9 October 2026

AI Systems

## [Evaluate the reasoning, not the fluency](https://lenvanderhof.com/en/blog/evaluate-reasoning-not-fluency/)

A fluent answer is not a graded one. Score the trace before you trust the sentence.

Len P. van der Hof21 August 2026

AI Systems

## [Frontier, workhorse, or local: three classes, not a religious war](https://lenvanderhof.com/en/blog/frontier-vs-workhorse-vs-local/)

The forum fight is cloud versus local. The operator question is which class owns this task.

Len P. van der Hof24 September 2026

AI Systems

## [Governed agent memory: what may persist, and who may write it](https://lenvanderhof.com/en/blog/governed-agent-memory/)

An agent's memory is a database that the agent writes itself. Give it the rules you would give any other database.

Len P. van der Hof30 September 2026

AI Systems

## [HITL in AI workflows: put the named person on the irreversible step](https://lenvanderhof.com/en/blog/hitl-in-ai-workflows/)

If the model can ship while the reviewer is on a plane, you did not apply HITL. You labelled a spectator.

Len P. van der Hof11 September 2026

AI Systems

## [How many AI agents can one person supervise?](https://lenvanderhof.com/en/blog/how-many-agents-can-you-supervise/)

Starting another agent is a one-time decision. Reading its work, checking its handoffs and handling its exceptions come back every week.

Len P. van der Hof28 October 2026

AI Systems

## [When nobody approves in time, your AI agent must not act](https://lenvanderhof.com/en/blog/human-approval-timeout-test/)

An expired request records silence. A rejected request records a decision. Both must block the action.

Len P. van der Hof20 October 2026

AI Systems

## [Human in the loop meaning: a named person who can fail the work](https://lenvanderhof.com/en/blog/human-in-the-loop-meaning/)

If nobody can fail the output against written criteria, the loop is decorative.

Len P. van der Hof27 August 2026

AI Systems

## [Human on the loop: watching is not a fail gate](https://lenvanderhof.com/en/blog/human-on-the-loop/)

If the work can ship while the watcher is on the gallery, the loop is already the wrong loop.

Len P. van der Hof20 September 2026

AI Systems

## [Introducing Undominated.ai: the independent AI inference price index](https://lenvanderhof.com/en/blog/introducing-undominated/)

Best first. Then price. The board that names every model something else beats without a trade-off.

Len P. van der Hof25 August 2026

AI Systems

## [Literary agent or AI agent: two different jobs](https://lenvanderhof.com/en/blog/literary-agent-or-ai-agent/)

One is a human representative. The other is a worker with a ceiling. They do not replace each other.

Len P. van der Hof8 September 2026

AI Systems

## [LLM integration is an operating-design problem](https://lenvanderhof.com/en/blog/llm-integration-is-operating-design/)

The scarce work is ownership, permission, and review, not the model name on the invoice.

Len P. van der Hof22 August 2026

AI Systems

## [MCP explained for founders](https://lenvanderhof.com/en/blog/mcp-explained-for-founders/)

Care about the permission and the log. The logo on the protocol is the least interesting part.

Len P. van der Hof16 August 2026

AI Systems

## [MCP is not USB-C: where the favourite analogy breaks, and why that matters](https://lenvanderhof.com/en/blog/mcp-is-not-usb-c/)

The plug is standard. What the model reads through it is not.

Len P. van der Hof6 October 2026

AI Systems

## [MCP meaning: Model Context Protocol, not a product](https://lenvanderhof.com/en/blog/mcp-meaning/)

The full form is the least interesting part. The permission boundary is the operating decision.

Len P. van der Hof30 August 2026

AI Systems

## [MCP OAuth is not tool authorization: a valid token is not a yes](https://lenvanderhof.com/en/blog/mcp-oauth-is-not-tool-authorization/)

OAuth tells your server the token is genuine and meant for it. Your code decides whether this call may approve this expense.

Len P. van der Hof5 October 2026

AI Systems

## [MCP vs function calling: two layers of the same tool call](https://lenvanderhof.com/en/blog/mcp-vs-function-calling/)

In most MCP setups the model still makes a function call. MCP changes where the tool lives, not how the model asks for it.

Len P. van der Hof25 September 2026

AI Systems

## [Model registry vs vendor list: know which models run](https://lenvanderhof.com/en/blog/model-registry-vs-vendor-list/)

A contract can clear a supplier while a production model remains unknown. Compare actual calls with the register before you trust it.

Len P. van der Hof1 November 2026

AI Systems

## [Why multi-agent LLM systems fail: the 14 MAST failure modes](https://lenvanderhof.com/en/blog/multi-agent-failure-modes-explained/)

An agent's mistake and the workflow that let it through are two different problems. MAST gives you a vocabulary for both.

Len P. van der Hof15 October 2026

AI Systems

## [Is one AI model enough? When a single model works, and when you need a portfolio](https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/)

The number of models is the wrong question. The right one: did anyone write down why each model runs, and what happens when it stops?

Len P. van der Hof29 September 2026

AI Systems

## [What an agent may never do](https://lenvanderhof.com/en/blog/override-doctrine/)

HITL is who can fail the output. The doctrine is what must be refused when no one is watching.

Len P. van der Hof23 August 2026

AI Systems

## [Prove you can stop an agent: interrupt the run and revoke access](https://lenvanderhof.com/en/blog/prove-you-can-stop-an-agent/)

An instruction to stop is a request. A stop is a control that someone outside the agent can operate.

Len P. van der Hof30 September 2026

AI Systems

## [RAG freshness: last month's PDF is not a lookup problem](https://lenvanderhof.com/en/blog/rag-freshness/)

The system can retrieve clause 12.3 perfectly and still miss the signed amendment.

Len P. van der Hof22 September 2026

AI Systems

## [RAG vs fine-tuning vs agents: three jobs, not a ladder](https://lenvanderhof.com/en/blog/rag-vs-finetuning-vs-agents/)

A vector store is not a trained model. A trained model is not a worker. A worker without a ceiling is a demo.

Len P. van der Hof3 September 2026

AI Systems

## [How to retire a production agent (stopping it is not enough)](https://lenvanderhof.com/en/blog/retire-a-production-agent/)

A stopped agent is not a retired agent. The next schedule, a copied key or a monthly plan can keep it working and billing.

Len P. van der Hof5 October 2026

AI Systems

## [ROSTER vs CHORUS: AI agents for one person, or for a team?](https://lenvanderhof.com/en/blog/roster-vs-chorus/)

Count the people who can say no to the work. One name means ROSTER. Several names mean you need CHORUS as well.

Len P. van der Hof26 September 2026

AI Systems

## [SENSE: a 90-day strategic intelligence layer](https://lenvanderhof.com/en/blog/sense-90-day-intelligence-layer/)

Stop collecting briefings. Start changing Monday.

Len P. van der Hof17 August 2026

AI Systems

## [STACK: version the agent OS like you version the app](https://lenvanderhof.com/en/blog/stack-version-the-agent-os/)

Rules in five tools. MCP without contracts. Skills nobody owns. That is not a model problem.

Len P. van der Hof23 August 2026

AI Systems

## [STACK vs AGENTS.md: where instructions end and controls begin](https://lenvanderhof.com/en/blog/stack-vs-agents-md/)

An instruction file is a request. A control is a refusal.

Len P. van der Hof1 October 2026

AI Systems

## [The Agentic Author: Stop Writing Books. Build a Production System.](https://lenvanderhof.com/en/blog/the-agentic-author-introduction/)

You are not writing a book. You are building a knowledge-production system that happens to output books. BooksOS makes that system auditable.

Len P. van der Hof5 August 2026

AI Systems

## [The Agentic Codebase: Version the Agent OS Like You Version the App](https://lenvanderhof.com/en/blog/the-agentic-codebase-introduction/)

Prompts in chat are not architecture. Scattered rules are not governance. STACK turns your repository into a maintainable agent operating system.

Len P. van der Hof5 August 2026

AI Systems

## [The Multi-Agent Organization: Close the Coordination Gap With CHORUS](https://lenvanderhof.com/en/blog/the-multi-agent-organization-introduction/)

Tools store state. Protocols decide. CHORUS is the operating layer between having agents and running them as one team.

Len P. van der Hof15 August 2026

AI Systems

## [TRACE vs RAG: one fetches the text, the other checks what it proves](https://lenvanderhof.com/en/blog/trace-vs-rag/)

An answer can quote the right document and still draw the wrong conclusion.

Len P. van der Hof28 September 2026

AI Systems

## [Using AI agents effectively: one job, a ceiling, a log](https://lenvanderhof.com/en/blog/using-ai-agents-effectively/)

A second tool without a ceiling is not leverage. It is a second way to spend.

Len P. van der Hof27 August 2026

AI Systems

## [What is a knowledge graph? Named nodes, typed edges, a walk you can audit](https://lenvanderhof.com/en/blog/what-is-a-knowledge-graph/)

If you cannot name the edge, you have a poster. If you cannot open the node, you have a rumour.

Len P. van der Hof14 September 2026

AI Systems

## [What is a model cascade? Escalate with a stop rule](https://lenvanderhof.com/en/blog/what-is-a-model-cascade/)

A cascade without a stop rule is spending until something sounds fluent.

Len P. van der Hof24 September 2026

AI Systems

## [What is a personal health agent? Data architecture, not a clinician](https://lenvanderhof.com/en/blog/what-is-a-personal-health-agent/)

If it answers as a doctor, it is already the wrong worker. If it cannot show the row it used, it is a vibe.

Len P. van der Hof21 September 2026

AI Systems

## [What is agent orchestration? Coordination, not a graph demo](https://lenvanderhof.com/en/blog/what-is-agent-orchestration/)

If ownership or review disappears at a boundary, the canvas is not coordinating the work.

Len P. van der Hof19 September 2026

AI Systems

## [What is agent strategy? Selection, sequence, exposure, exit](https://lenvanderhof.com/en/blog/what-is-agent-strategy/)

A well-run agent can still be the wrong worker to deploy first.

Len P. van der Hof21 September 2026

AI Systems

## [What is AI team dynamics? Mixed teams, human accountability](https://lenvanderhof.com/en/blog/what-is-ai-team-dynamics/)

If the founder is the universal exception handler, you do not have dynamics. You have a queue with extra tools.

Len P. van der Hof19 September 2026

AI Systems

## [What is an AI agent? A definition founders can use](https://lenvanderhof.com/en/blog/what-is-an-ai-agent/)

If it cannot refuse a tool call, it is not an agent you can run. It is a demo that can spend money.

Len P. van der Hof15 August 2026

AI Systems

## [What is an AI agent charter? Seven fields or it is a demo](https://lenvanderhof.com/en/blog/what-is-an-ai-agent-charter/)

If you cannot fill the seven fields, do not hand it a key.

Len P. van der Hof16 September 2026

AI Systems

## [What is an evidence ledger?](https://lenvanderhof.com/en/blog/what-is-an-evidence-ledger/)

The ledger is the product. The reference list is a rendering.

Len P. van der Hof31 August 2026

AI Systems

## [An MCP connector is the plug. Permission is a separate ceiling.](https://lenvanderhof.com/en/blog/what-is-an-mcp-connector/)

The connector is a plug. The server is the tool. The ceiling is still the product.

Len P. van der Hof27 August 2026

AI Systems

## [What is BooksOS? Seven gates, not a writing CMS](https://lenvanderhof.com/en/blog/what-is-booksos/)

Agents draft. Gates decide. A fluent chapter that cannot survive a fact-check is inventory you will throw away.

Len P. van der Hof27 August 2026

AI Systems

## [What is CHORUS? Coordination, not a choir](https://lenvanderhof.com/en/blog/what-is-chorus/)

Org charts name reporting lines. CHORUS names coordination lines.

Len P. van der Hof6 September 2026

AI Systems

## [What is GRAIN? The inspectable retrieval loop](https://lenvanderhof.com/en/blog/what-is-grain/)

RAG names the architecture. GRAIN names the honesty check.

Len P. van der Hof15 September 2026

AI Systems

## [What is hybrid search? Two indexes, one inspectable hit](https://lenvanderhof.com/en/blog/what-is-hybrid-search/)

Keywords catch the name. Vectors catch the paraphrase. You still have to open the hit.

Len P. van der Hof29 September 2026

AI Systems

## [What is LLM grounding? The sentence has to open a folder](https://lenvanderhof.com/en/blog/what-is-llm-grounding/)

A real source ID can decorate a false claim. Grounding begins after the citation resolves.

Len P. van der Hof22 September 2026

AI Systems

## [What is model routing? A policy, not a default](https://lenvanderhof.com/en/blog/what-is-model-routing/)

If you cannot name which model owns which job, you do not have a routing policy. You have a default.

Len P. van der Hof23 September 2026

AI Systems

## [What is RAG? When AI looks up a source before it answers](https://lenvanderhof.com/en/blog/what-is-rag/)

A fluent AI answer can still be unsourced. RAG is the method that has to show the page it used.

Len P. van der Hof23 August 2026

AI Systems

## [What is ReasonKit? Structured decisions through MCP](https://lenvanderhof.com/en/blog/what-is-reasonkit/)

An AI assistant can sound certain. ReasonKit makes it show which claims were checked, and stop when the evidence is missing.

Len P. van der Hof4 October 2026

AI Systems

## [What is ROSTER? How one person manages a team of AI agents](https://lenvanderhof.com/en/blog/what-is-roster/)

A role is defined by what it may not do. Six written decisions turn an AI agent into a specialist you can grade, fix or retire.

Len P. van der Hof26 September 2026

AI Systems

## [What is SENSE? Five moves, not another unread summary](https://lenvanderhof.com/en/blog/what-is-sense/)

A digest proves that a system can produce text. SENSE asks where that text changes the work.

Len P. van der Hof23 September 2026

AI Systems

## [What is STACK? Five layers, not a vendor diagram](https://lenvanderhof.com/en/blog/what-is-stack/)

The agent OS should be versioned like the app. If the rules live in five tools, you have debt, not a model problem.

Len P. van der Hof5 September 2026

AI Systems

## [What is TRACE? Five steps to check research before you act on it](https://lenvanderhof.com/en/blog/what-is-trace/)

Three links can be one source. TRACE is how you find out before the decision does.

Len P. van der Hof25 September 2026

AI Systems

## [When not to use an AI agent: five signs, and what to use instead](https://lenvanderhof.com/en/blog/when-not-to-use-an-agent/)

An agent earns its cost only when the next step depends on what it finds. Most tasks never get there.

Len P. van der Hof28 September 2026

AI Systems

## [When to use RAG: named passages, not a vector store](https://lenvanderhof.com/en/blog/when-to-use-rag/)

The furniture is optional. The named passage is not. If you cannot open the file, you did not ship retrieval.

Len P. van der Hof13 September 2026

AI Systems

## [Why multi-agent systems fail: the coordination gap](https://lenvanderhof.com/en/blog/why-multi-agent-systems-fail/)

Capability arrived in an afternoon. The protocol never did. That gap is the whole story.

Len P. van der Hof18 August 2026

AI Systems

## [Why your RAG answer is wrong: find the broken step before you change the model](https://lenvanderhof.com/en/blog/why-your-rag-answer-is-wrong/)

The model is the last thing to blame, not the first.

Len P. van der Hof18 October 2026

```json
{
	"@context": "https://schema.org",
	"@graph": [
		{
			"@type": "CollectionPage",
			"@id": "https://lenvanderhof.com/en/blog/category/ai-systems/#webpage",
			"url": "https://lenvanderhof.com/en/blog/category/ai-systems/",
			"name": "AI Systems Articles | Len P. van der Hof",
			"description": "Practical analysis of AI agents, reasoning systems, and reliable automation for founders and operators shipping production software.",
			"inLanguage": "en",
			"isPartOf": {
				"@id": "https://lenvanderhof.com/#website"
			},
			"breadcrumb": {
				"@id": "https://lenvanderhof.com/en/blog/category/ai-systems/#breadcrumb"
			}
		},
		{
			"@type": "BreadcrumbList",
			"@id": "https://lenvanderhof.com/en/blog/category/ai-systems/#breadcrumb",
			"itemListElement": [
				{
					"@type": "ListItem",
					"position": 1,
					"name": "Home",
					"item": "https://lenvanderhof.com/"
				},
				{
					"@type": "ListItem",
					"position": 2,
					"name": "Blog",
					"item": "https://lenvanderhof.com/en/blog/"
				},
				{
					"@type": "ListItem",
					"position": 3,
					"name": "AI Systems",
					"item": "https://lenvanderhof.com/en/blog/category/ai-systems/"
				}
			]
		}
	]
}
```