Recipe tag | eYssen AI

32 Articles

Recipe ×

Document Processing and Accounting with AI

Krisztian Eyssen, Krisztian Eyssen

Document Processing and Accounting with AI

One question decides whether invoice scanning is worth it: **does the tax authority give you back your supplier's data, or only your own.** In Hungary it does, in Spain it does not — and that makes the same software redundant in one place and the only input in the other. Here is the formula to work it out: at a fifteen-euro hourly cost the credit pays for itself at **29 seconds** saved — so price is not the question.

Odoo Intermediate Business owner Recipe Business, Law & Security

Sep 13, 2026

A Company Knowledge Chatbot: A Realistic Plan

Krisztian Eyssen, Krisztian Eyssen

A Company Knowledge Chatbot: A Realistic Plan

An airline's chatbot gave the wrong rules for its bereavement fare. The company's defence was that it is not responsible for "information provided by its agents, including a chatbot" — and it floated the idea that **the chatbot is a separate legal entity**. The court's answer is one sentence worth printing out: **"It makes no difference whether the information comes from a static page or from a chatbot."** The damages came to $812. The precedent is not so cheap.

Intermediate Business owner Recipe Data, RAG & Self-Hosted AI

Sep 13, 2026

Understanding a Legacy Codebase with AI

Krisztian Eyssen, Krisztian Eyssen

Understanding a Legacy Codebase with AI

Model performance **falls off sharply above 32,000 tokens** — while the context window they advertise runs from 128,000 to a million. And there is a measured bias: ask how the repo is put together, and the model would rather **rewrite** it than understand it. This post is about how to ask so that does not happen.

Intermediate Developer Recipe Coding & Development

Sep 13, 2026

Documentation You Will Actually Maintain

Krisztian Eyssen, Krisztian Eyssen

Documentation You Will Actually Maintain

Two independent measurements say the same thing, and both contradict received wisdom: **bad documentation is measurably worse than none**, and the machine-generated "repo tour" **lowered the success rate while making runs 20–23% more expensive**. This post is about what survives all that — not much, but it works.

Intermediate Developer Recipe Coding & Development

Sep 13, 2026

Meeting Notes People Actually Read

Krisztian Eyssen, Krisztian Eyssen

Meeting Notes People Actually Read

A good set of meeting notes is not a summary, it is three lists: decision, owner, deadline. Nobody reads whatever else is in there. The machine can write all three — but in the most recent measurement **more than 40% of the claims in generated summaries were not supported by the recording**, and outside the big languages even the transcript gets every seventh word wrong.

Beginner Office worker Recipe Work & Productivity

Sep 13, 2026

Inbox Triage with AI

Krisztian Eyssen, Krisztian Eyssen

Inbox Triage with AI

The inbox is the place where AI measurably helps — but not the way you expect. In a randomised 6,000-person experiment, the people who actually used it spent **18% less time** reading email, while the number of conversations **did not change**. And there is one risk that barely gets discussed: an incoming message is not only data to the machine, it **can also be an instruction**.

Beginner Office worker Recipe Work & Productivity

Sep 12, 2026

Spreadsheets and Data Cleanup with AI

Krisztian Eyssen, Krisztian Eyssen

Spreadsheets and Data Cleanup with AI

The spreadsheet is the ground on which AI is wrong with the most confidence. In the most recent measurement, run on real business workflows, the best model reached **34.9% task accuracy** — and on debugging, **12%**, which makes it least suited to precisely what most people would use it for. Its mistakes are also quiet: what comes back is not an error message but a plausible number.

Intermediate Office worker Recipe Work & Productivity

Sep 12, 2026

From Outline to Deck: Slides with AI

Krisztian Eyssen, Krisztian Eyssen

From Outline to Deck: Slides with AI

The expectation is that the structure is your job and the machine takes care of the design. The only independent measurement says **it works the other way round**: on the logical build of a deck the best system scores 81 points, but **on visual layout even the best manages only 62.8**, with most of the field under 50. And source fidelity is worse still.

Beginner Office worker Recipe Work & Productivity

Sep 12, 2026

Drafting Contracts, Offers and Reports with AI

Krisztian Eyssen, Krisztian Eyssen

Drafting Contracts, Offers and Reports with AI

There are two measurements, and at first glance they rule each other out. On drafting contracts, AI performed **on a par with human lawyers** — 57% versus 56.7% reliability, and in under a minute instead of thirteen. On legal research, though, even the market-leading specialist tools **hallucinated in 17–33% of cases**. The two are not a contradiction: writing a draft and finding a citation are not the same job.

Intermediate Office worker Recipe Work & Productivity

Sep 12, 2026

Reading and Researching Faster Without Losing Accuracy

Krisztian Eyssen, Krisztian Eyssen

Reading and Researching Faster Without Losing Accuracy

Two large experiments gave opposite answers to the same question. A series with 10,462 participants found that an AI summary leaves **shallower knowledge** behind — even when it contains exactly the same information. In another experiment it **improved** comprehension, and did so even on questions the summary never touched. The difference comes down to one thing: **does it replace the reading, or lead into it.**

Beginner Everyone Recipe Work & Productivity

Sep 12, 2026

Automating Recurring Reports

Krisztian Eyssen, Krisztian Eyssen

Automating Recurring Reports

On a 2026 benchmark for data-analysis agents, the best system answered correctly first time in **38%** of cases — and, worse than that, one model returned **a silently empty answer in 63.4%** of cases when it got stuck. Not an error message. An empty result. On a monthly report that is the most expensive kind of failure, because it stays hidden for months.

Intermediate Office worker Recipe Work & Productivity

Sep 12, 2026

Customer Support Replies with AI

Krisztian Eyssen, Krisztian Eyssen

Customer Support Replies with AI

In a peer-reviewed study of 5,172 agents, an AI assistant raised the number of issues resolved per hour by **15%** — over 30% among beginners, but with **a slight loss of quality** among the most experienced. A 2026 field experiment then showed the limit: where AI **takes over** the conversation, customer ratings **get significantly worse**. The difference is not in the technology, but in whether the machine assists or replaces.

Intermediate Business owner Recipe Work & Productivity

Sep 12, 2026

1
2
3

Follow Us

Archives

Tags

Audience

Everyone Office worker Marketer Creative Business owner Developer

Category

AI Basics Safety & Risk Regulation & Policy Infrastructure & Chips Business & Deals Prompting Models & Market Writing & Content Images & Design Video & Audio Marketing & SEO Coding & Development Work & Productivity Agents & Automation Data, RAG & Self-Hosted AI Business, Law & Security

Format

How-To Explainer Comparison Tool review Recipe Template Troubleshooting Case study Glossary News

Level

Beginner Intermediate Advanced

Tool

OpenRouter Perplexity HeyGen ElevenLabs Suno Obsidian MCP n8n Ollama EYAS

Others

Odoo