AI agents for business · custom AI agent development
An AI agent for your business
that does the actual job.
A chatbot answers. An agent acts: it reads your data, uses your tools, follows your rules and hands over to a person when it should. I build custom AI agents around how your business really runs, and I have several of them working in production today, not in a demo.
The work
Four agents businesses actually put to work.
An agent is worth building when it owns one job completely. These are the four shapes that keep proving themselves.
01
The customer-facing agent
On WhatsApp, your website, or both. It knows your catalogue, prices and policies, asks the follow-up questions a good employee would, takes the booking or the order details, and passes the conversation to a human with the full context when it reaches its limit.
Best if customers wait for answers your team retypes daily.
02
The internal agent over your data
Staff ask in plain language; the agent searches your documents, database or archive by meaning, answers with the source attached, and can take the next step: fill the form, draft the reply, open the ticket.
Best if the answer exists but nobody can find it.
03
The multi-agent pipeline
For work too large for one prompt: one agent plans, another drafts, another reviews, another checks the facts. Each has a narrow role and a structured contract with the next, which is what makes long, high-stakes output reliable.
Best if the output is long, expert and must be right.
04
The agent inside your product
If you are building software, the agent is a feature: tool calling, memory, streaming, cost controls, evaluation, and the product around it. I build the whole thing in Next.js and TypeScript, not only the model calls.
Best if AI is the product, not an add-on.
The receipts
Agents with real users, not demo videos.
Darija
A WhatsApp agent that interviews and matches
RESO Khdma: an agent talks to workers inside WhatsApp in the dialect they type, builds their profile from the conversation and matches them to jobs with semantic search.
Read the RESO Khdma case study →4 agents
A legal team in a pipeline
FASL: strategist, drafter, senior partner and auditor agents draft Moroccan legal appeals, grounded in the law and able to stop and ask the lawyer for a document.
Read the FASL case study →7 agents
From product photo to finished video ad
Laqta: seven agents write, voice and render a Darija video ad from a single product upload.
Read the Laqta case study →Paid
A consumer product run by a model pipeline
Magical Hekaya: a multi-model pipeline writes, illustrates and narrates a personalised children's book, with paying customers.
Read the Magical Hekaya case study →
How it runs
From a job description to an agent on duty.
Step 1
A call, free
Thirty minutes on the one job you want the agent to own. I tell you whether an agent is the right tool, or whether a simpler automation would do.
Step 2
A fixed scope
One page: what the agent does, what it must never do, which tools and data it touches, when it lands and what it costs. Agreed before any code is written.
Step 3
Build and test on real cases
The agent is tested against your real conversations and documents, including the awkward ones, and you watch it work before it meets a customer.
Step 4
Launch with a human in reach
It goes live with handover to a person, logs of every conversation and a clear cost per use. You own the code, the prompts and the accounts.
Written up properly
Three agent systems, written up in full.
The questions
Asked by every owner.
Want to know how these are built? LLM agent architecture explains the structure behind the agents on this page.
What is an AI agent for business, in plain words?
Software that is given a goal, your data and a set of tools, and works through the steps itself: it reads, decides, acts and reports. A chatbot only answers questions. An agent can also look up the order, book the slot, draft the document or update the record, inside limits you set.
Is an AI agent different from a chatbot?
Yes. A chatbot replies from a script or a knowledge base. An agent uses tools: your database, your calendar, your WhatsApp number, your CRM. If you only need answers to common questions, a well-built chatbot is cheaper and I will say so.
Can the agent work on WhatsApp?
Yes, through the official WhatsApp Business API. RESO Khdma runs entirely inside WhatsApp. The same agent can also sit on your website, so customers use whichever they prefer.
Should I buy an off-the-shelf agent tool instead?
Often, yes. If a ready-made product does your job well, buy it. A custom agent makes sense when your process, your language or your data is the part the generic tools get wrong, or when the agent is part of what you sell.
How do you keep an agent from making mistakes with customers?
It answers only from your own information and shows where the answer came from, it is given a short list of allowed actions, anything risky needs a human to approve, and it is built to say “let me get a person” when it is unsure. Every conversation is logged so you can review it.
What does a custom AI agent cost?
It depends on how many tools it uses and how much is at stake when it is wrong, so I quote after a free call: one fixed price in writing, plus an estimate of the monthly running cost (model usage and hosting).
Which languages can it speak?
English, French, Arabic and Moroccan Darija are the ones I have shipped and can test properly as a speaker. Other languages are possible; I will be clear about what I can verify myself.
Who builds it and who owns it?
I build it myself; there is no team the work is passed to. You own the code, the prompts and the accounts, and it runs on infrastructure you control.
Free call · no deck, no pitch
Describe the job.
I will tell you if an agent can hold it.
Thirty minutes, no charge, and an honest answer, including “a simpler tool will do.” I read and answer every email myself, within 24 hours.
Email [email protected] →