Offers
Packaged formats show their price. The others are quoted after a first call. We confirm scope and a written quote before any work.
01Four formats
We don’t resell AI tools or subscriptions (ChatGPT, Claude, Gemini or others) and earn no commission on them. What we sell is our own help: the services below.
-
Discovery workshop
€490
About 3.5 hours, up to 5 people from one company, under €100 per person with five.
Your needs, constraints and budget reviewed together. A written plan with a prioritized first tool.
-
Training
€90 per seat
Open group, about 3 hours, up to 8 people. Next session on request. Also private or one to one.
A working setup on your own computer. A first exercise, built by you.
-
Build with you
Quoted after a first call
One agreed tool and its checks.
A working tool you know how to change, with its code and documentation. A change you make yourself. A handover: who maintains what.
-
Monthly guidance
Quoted after a first call
One 90-minute session a month and questions by email.
Your tools keep improving, at your pace. Advice when you get stuck.
Prices exclude VAT, which is added. AI subscriptions, API usage, hosting and domains are separate, paid directly to the providers. Remote by default; on-site with travel and accommodation agreed in the quote.
02Your first AI agent
Creating an agent is easy. Making it safe and economical is not.
Copilot, ChatGPT, Gemini and Claude all let you create your own agent: an assistant that follows your instructions, reads your documents and can act in your tools. Available features depend on your plan and permissions. Setting one up takes a few steps. The hard part comes after:
- What it can see and do. Give it only the files and actions it needs. In some tools, asking you before it acts is off by default.
- What it reads. A document, a web page or an email can hide instructions that turn your agent against you. We test it against them before your team relies on it.
- What it costs. An agent works in loops and can use far more than a chat. The right model, short instructions and a spending cap keep the bill predictable; caps are often off unless you set them.
It works. But what happens when it gets something wrong?
An agent can pass your first tests and still reach more information or actions than its task needs. A convincing answer can hide a mistake. An email or a document can contain instructions that divert it from your request.
Imagine an agent built to summarise customer emails. If its connection also lets it send messages, a malicious email could trick it into forwarding confidential information. Depending on its permissions, other mistakes could change records, delete files or trigger repeated paid actions.
That is why we check more than the result. We limit what the agent can access and do, require your approval for sensitive actions, set usage limits and test it with misleading inputs. Telling an agent to “be careful” is not enough: the limits must be enforced by the tools it uses. No check makes an agent risk-free; these reduce what can go wrong.
These risks are described in the OWASP Top 10 for LLM and generative AI applications.
How we help
- Training: you build your first agent on your own account, in the assistant you already use. €90 per seat, excl. VAT; private or one to one on request.
- Build with you: an agent connected to your own systems, with its checks and a handover. Quoted after a first call.
Not sure you need an agent? Start the Discovery Chat: it tells you whether a standard subscription, a cowork tool or your own tool fits.
03Which tools and subscriptions?
Instead of bending your work to generic software, get a tool shaped around the way you work. After Software as a Service (SaaS) comes AI as a Service (AIaaS). You need to learn how to use it, but don't be afraid of building: with the right method and the right checks, it has never been easier to start.
It depends on the task, and it is usually more than one: a chat assistant, a coding assistant, an editor such as VS Code, and version history with Git (often hosted on GitHub). We favour open-source and widely used tools, and choices change as the tools evolve.
For the tools we build with you and your business data, we don't use free plans. We pick plans whose data terms you have read, and explain what each service receives and what it costs. A free plan is fine for discovering a tool, as long as you don't give it sensitive data.
04How it works
Start with the guidance you need. After a first conversation, we agree the right next step. Ongoing guidance is optional.
- Identify. Pick one task that costs time or causes mistakes. (Discovery workshop)
- Plan. Agree what the tool does, which data it uses, how you will know it works. (Discovery workshop)
- Do. Build it with AI, with our guidance. You do the work; we guide and unblock. (Training, build with you, monthly guidance)
- Check. Test on sample data. Compare saved versions. Use it for real only once checked.
- Improve. Use it, note what to change, adjust one small step at a time. Then back to Identify for the next improvement or the next tool.
The same plan, do, check, improve cycle behind our method, adapted to a small business.
05Questions
We're not developers. Can we really do this?
Yes, with a well-chosen first tool. You describe what you need, the AI writes the code, and you learn to test and adjust it. We pick a starting project that fits your level.
Is the first call free?
Yes. The 30-minute call is free, with no commitment: we talk about your task and tell you whether a custom tool is the right answer.
Why do we need help if ChatGPT can do it?
Chat gets you started, but building a real tool in a chat window means endless copy and paste, and you end up testing the AI's code instead of driving your need. With the right setup the AI works on the files directly and checks its own work. You stay focused on your business, and it goes much faster.
Why not hire a developer?
For a large or critical system, you should. For a small internal tool you'll want to adjust yourself, learning to build it keeps you in control. We tell you which case you're in.
How much time does it take?
A workshop is half a day, a training session about three hours. Building together takes a few sessions, with some practice in between. We agree the time before starting.
What if the result doesn't suit us?
Before starting, we agree in writing on the scope and on how to check that the tool works. After the discovery workshop, the written plan is yours: you can carry it out with us, on your own or with someone else.
Which subscription and tool should we use?
Rarely just one, and not the same forever. It depends on the task and your data. For the tools we build with you, we don't use free plans, and we explain the cost and data terms of each choice.
Is our data safe?
Use sample or anonymized information while learning. Before using business data, agree which services, account settings and access permissions are appropriate. Never put passwords or API keys in prompts or published code.
What if something breaks?
You learn to compare saved versions and go back to an earlier one. That recovers your tool's files. It cannot undo an email already sent, a payment already made or data already shared, so those steps get checked first.
Who maintains the tool afterwards?
Someone on your team, named at handover, with written instructions. Monthly guidance is there if you want ongoing help.
Will we depend on you?
No. What you keep (code, documents, accounts) is agreed up front and stays with you. AI and hosting services are subscriptions in your own name.
Doesn't all this AI use waste energy?
We choose the simplest approach that meets the need. Some tools use AI only while being built, others while running. We explain the difference and avoid unnecessary calls.
Ready to talk about your first agent?
Already know what you need? Book a call (opens a new tab). Or write to contact@avapmo.com.