Field Notes
Self-hosted AI, written down.
CPLT builds AI assistance for business owners and their teams. Stéphane Lepain’s notes carry dates, assumptions and limits.
One email when a new essay ships. No other use, unsubscribe by replying, or take the RSS feed.
Prefer video? Business video and transcript · All CPLT videos
What are you deciding?
How to compare supplier quotes with AI before buying
A purchase-quote checklist for comparing the same scope, checking the figures and keeping the supplier decision with a person.
How to make your unstructured data AI-ready
Much of what a small company knows sits in documents, emails and scans. AI answers go wrong when it cannot read the file, picks an old version or ignores who may see it. These are the four steps I use.
Can AI Process Your Invoices? What to Check First
AI can read an invoice and fill in the supplier, dates, totals and line items. I would check each result against your orders and records before anything is paid.
AI agent vs chatbot: what a small business gets
A chatbot answers in a conversation. An AI agent carries out the steps of a task in your tools. How to tell them apart, where each fits and what to ask a supplier.
What to do when your AI assistant gives a wrong answer
Find out what the assistant was working from, fix the document or the instruction, keep the question for re-testing and decide who approves that kind of answer.
Can Copilot access all your company files?
Check the sharing rules your team already uses, the visibility of connected sources and what happens when someone’s access changes.
Can I see what my team types into ChatGPT?
What an owner can and cannot see of staff AI chats depends on the plan and its admin settings. Decide what you want to see, then tell your team.
How to train AI on your company documents
Most small businesses get what they want without training a model. The assistant looks up the right document when someone asks and shows the page it used.
How to get employees to actually use AI
Many staff try AI on their own, quietly and without training. I would start from the work they repeat, approve the tools and agree who checks each result.
Can AI answer customer emails for a small business?
Yes, for routine questions your documents already answer. I would let AI prepare the reply and keep a person on every reply that commits the business.
Which tasks should a small business automate with AI first?
I would shortlist repeated work with clear sources and a named reviewer, then compare the full cost of completing it to the same standard.
Charte IA en entreprise : un modèle pour les PME
Neuf règles à copier et à adapter à votre PME, des exemples d’usages à écarter et un déroulement proposé sur une semaine.
AI acceptable use policy for a small business
Nine rules to copy and adapt, with everyday examples of unacceptable use and a suggested rollout over a working week.
Is It Safe to Put Company Data in ChatGPT?
It depends on the plan your people log in with and on what they paste. What OpenAI says it does with each plan, what to keep out of any chat, and the rules I would give a small team.
Peut-on mettre des données d’entreprise dans ChatGPT ?
Tout dépend de l’abonnement utilisé par votre équipe et de ce qu’elle colle dans la conversation. Ce qu’OpenAI dit faire selon l’offre, ce qu’il vaut mieux garder pour soi, et les règles que je donnerais à une petite équipe.
AI automation agency or independent consultant: which suits a small business?
Both can build AI that does real work for a small business. They differ in who you deal with, how much they take on at once and who runs the result afterwards. The questions I would ask either one.
Agence d’automatisation IA ou consultant indépendant : que choisir pour une PME ?
Les deux peuvent construire une IA qui fait un vrai travail pour une PME. Ils diffèrent par l’interlocuteur, la quantité de travail menée en parallèle et la personne qui fait tourner le résultat ensuite. Les questions que je poserais à l’un comme à l’autre.
Why Enterprise AI Needs More Than a Smarter Model
Can you trust AI with your enterprise decisions? For some of them, yes, with six controls around the model. For the rest, no. A smarter model gives you none of the six.
What Is a Private AI Stack vs ChatGPT Enterprise?
A private AI stack runs in a workspace your business owns: the interface, documents, indexes and logs stay in systems you control. Here is how that differs from ChatGPT Enterprise.
I Tested AI on GitLab’s Public Security Questionnaire
I took GitLab’s published security questionnaire, hid its answers, and tested whether an automated workflow could answer the same questions from GitLab’s public handbook alone, with every answer quoting its source.
Eight Questions to Ask Before Approving Any AI Spend
Before the budget moves, eight practical questions about an AI project’s process, outcomes, data and costs. A recommended checklist, not a validated predictor.
Why AI Use Doesn’t Turn Into AI Value
McKinsey finds 88% of surveyed organisations use AI somewhere, but 39% report enterprise-level EBIT impact. Organisational barriers recur in the research; technical limits matter too.
Which AI Model Should Check Your AI’s Work?
I sealed four AI models in a sandbox with no view of the answer key, then had each one review the same frozen code for planted defects. Two providers’ own safety filters stopped their model from finishing the job before it reached my score sheet.
How Jev Speeds Up My Production AI Review Workflow
Jev is a small decision model I call from code for four narrow jobs on my production AI workflow: picking a model, ranking reference pages, catching bad inputs and suggesting where a file belongs. A person decides every result.
What Is an Automation Engineer, and Do You Need One?
An automation engineer embeds inside a company to turn AI into a working part of a real process, not a demo.
One Agent on My Phone, Five Tools Behind It
A one-day experiment on my laptop: one assistant on my phone takes the request, Jev sizes it, cheap models do the research, frontier models get the hard work, and every task ends with a written hand-off. What moves into my production workspace first is the discipline, not the router.
How to Scope an AI Trial Before Buying Tools
Choose the work, define the data boundary and human checks, and write down when to continue or stop before committing to tools.
Acceptance Checks for an Internal Knowledge Assistant
Write down what a usable answer must do, which sources it may use and when a person must take over. A fictional handbook shows how to specify the checks.
Cheaper AI Tokens. A Bigger Bill.
Kimi K3 charges less per token than GPT-5.6 Sol. Yet one published benchmark reports a higher cost per task. I checked the prices, the token counts and what the comparison actually proves.
AI Pilot Cost: Measure Cost per Accepted Result
Define an accepted result, count failed attempts and human review, and compare the full cost of the work before committing to a build.
Automatic Model Routing: What to Test Before Using It
I assess automatic model routing by task quality, full cost, fallback behaviour and data permissions. A complexity estimate must not decide where data may go.
The LLM Writes Data. A Script Writes the Document.
I separate AI findings from document rendering so the source data, review decisions and output can be checked without regenerating the reasoning.
Agent Finds, Human Checks: Using Gated Vendor Knowledge
I separate finding a reference from permission to retrieve or share it. Human access does not automatically authorise copying gated content into an AI system.
Prompt Caching: Verify Configuration Against Actual Usage
I verify cache settings through accepted configuration and provider usage. A misspelled or unsupported option can look correct in a file and never take effect.
How I Evaluate a Cheaper Model for a Second Opinion
A cheaper reviewer is useful only if it catches the errors that matter. Here is how I would test one without confusing agreement with correctness.
When an AI Validator Invents a Rule
I check AI rule citations against an agreed source. A plausible identifier is not evidence, and a correct reference does not prove the conclusion.
Why an Agent Can Use Less Than Its Model’s Context Window
I check the effective context limit, framework configuration and actual requests before relying on a model’s advertised window for long tasks.
An AI Coding App Indexed 627,652 Files On My Laptop
OpenAI’s Codex desktop app registered my home directory as a workspace and kept scanning a project folder deleted three months earlier. The interesting part isn’t the bug. It’s that I could find it at all.
Named or Assumed: The First Question in Architecture Review
The weak point of most architecture dossiers is the sentence that says security is “handled by the platform.” Here’s the one distinction that separates auditable inheritance from a gap in disguise.
AI Router Health Checks: Count Cost and Detection Delay
I count scheduled API calls and compare their cost with the required failure-detection time before changing a model-router health-check interval.
Buying an AI Server vs Paying for OpenAI API Tokens: A TCO Comparison
Is buying an AI server cheaper than paying for OpenAI API tokens? Compare hardware, subscriptions and API usage with setup, power, operator time and result quality included.
3 Hidden Failure Modes of Self-Hosted LLMs Past 25 Users
KV cache fragmentation, the context-window I/O trap, and kernel scheduler thrashing, the three failure modes that crash self-hosted LLM stacks at scale, and how to fix them.
Use AI on Your Documents Without a Third Party
A local model alone does not keep documents in house. What has to run on your own systems, from OCR to backups, and how I check each step.
vLLM vs llama.cpp vs Ollama at 25 Concurrent Users
Single-user local inference is a solved problem. Production LLM serving at 25 concurrent users is a different engineering domain, and the published benchmarks all agree on the shape of the failure.
OpenAI DPA and GDPR Article 28: What to Check
A practical review of the contract, API retention, regional processing and the decisions your team still owns before sending personal data to an AI service.
Talk it through with me. Request a free call