ContextDocumentsPrompting
AI for business
AI is not generic. It just knows nothing about your company
When people say AI “writes like a robot”, they are almost always describing a model that has never seen a single document of theirs.
Home
Blog: AI for business and custom software
Practical notes on AI inside a company, agents, process automation and custom software — written from the projects we delivered. No pitch, and no figures we cannot show you.
DocumentsStructureAgents
Company memory
Your whole company memory, in one folder
A useful agent is not built out of prompts. It is built on top of a folder that states what you sell, to whom, at what price, and what you already promised.
HabitsMethodPractical
Working with AI
Five habits of the people who actually work with AI
They do not use a different model and they do not write longer prompts. They work in a loop: give context, ask, verify, correct in place, and save what worked.
SalesPreparationFirst step
AI agents
Your first meeting-prep agent takes about an hour to build
Do not start with the agent that runs your sales. Start with the one that writes a page before each meeting: you can tell within a day whether it is any good.
ProjectsRoutineTime
Productivity
The new chat every morning erases your work
You are not losing time because your prompts are bad. You are losing it because every morning you re-explain what it already understood yesterday.
DocumentsGetting startedPractical
Company memory
Do not tidy up before AI. The mess is the context
The project that never starts is the one waiting for everything to be clean. A model does not need tidy files. It needs reachable ones.
CRMPreparationClients
Sales
The AI agent knows more about the client than your CRM does
Empty CRM fields are not a discipline problem. They are a sourcing problem: the information exists, nobody just has time to copy it across.
ProjectsInstructionsTime
AI for business
You pay for the AI subscription and still do the work
If every request starts with two paragraphs explaining who you are and what you do, you are paying a subscription to do by hand the exact work it was meant to take over.
MistakesFixesPractical
Working with AI
Four mistakes when you put AI to work on your company
The same four mistakes show up in almost every company, whatever the industry. None of them is about the model, and all four can be fixed on the spot.
AccountingReportingCash flow
Finance
An AI agent changes the conversation you have with your accountant
“How are we doing this month” should not mean a phone call and a two-day wait. You already sent the documents that answer it.
ContextPermissionsRoles
AI agents
The same AI can be a chatbot or a colleague
A chatbot asks you for context. A colleague already has it. Between the two there is no different technology — just a folder and some permissions.
Company memoryInternal processAgents
From experience
The best thing we did this year: writing down the company memory
The valuable part was not the summary. It was the contradictions it surfaced inside our own documents, which nobody had ever read side by side.
PrivacyPolicyData
Security
Your company data and AI: it is not all or nothing
The conversation about company data stalls because it is held at the extremes. Split into three levels, most documents turn out to be public anyway.
DecisionBuyingProcess
Strategy
Do not buy any AI tool before answering this question
If you can describe, step by step, a job you do every week, you already have the specification. If you cannot, no subscription is going to fix anything.
MCPIntegrationsStandards
AI for business
One standard so AI can work with the apps you already use
Until recently, every link between an AI assistant and an application was a hand-written integration. MCP turns it into a socket with the same shape everywhere.
QuotingBudgetProcurement
Custom software
How to read a software quote
The gap between two quotes is rarely the gap between two developers. It is the gap between what was included and what was left out.
IoTRunning costsMaintenance
Connected hardware
What connected hardware costs to keep running
The hard part of an IoT project is not getting the device to talk the first time. It is keeping it talking for months without anyone driving out to it.
ProcessIntegrationsOperations
Automation
What process automation actually replaces
The right question is not “what can we automate” but “where is the same data being typed twice”. The answer is almost always between two systems that do not talk.