Artificial Intelligence
The systems behind a brand's presence online — agents, custom GPTs, automation and AI characters, built to sound like the brand rather than like a chatbot.
What this is
The automated layer of a brand: the agents that answer, the systems that publish, and the characters that speak. Built so that what comes out sounds like the brand and not like a generic assistant wearing its logo.
This is the newest of the six capabilities and the one most often bought badly. The failure is rarely technical. It is that nobody decided what the system was allowed to be.
Voice is the whole problem
An agent is a brand touchpoint that produces language at volume, without a human approving each sentence. That is a genuinely new risk. Every other channel gets reviewed before it ships; this one does not.
So the work is upstream of the technology: what the brand sounds like when it is being helpful, when it is being asked something awkward, and when it does not know. That last one matters most. A system confidently making something up is worse than a system that says it will find out.
Guardrails are a brand decision, not an engineering detail
What must this never claim? Prices it cannot honour. Delivery dates nobody agreed. Capabilities the business does not have. A single invented answer can create a real commercial obligation, and it is far cheaper to define the boundary than to argue about it afterwards.
Bilingual, tested in both
Persian is where most of these systems quietly fall apart. Models are commonly weaker in Persian than in English, mishandle RTL text in interfaces, and produce Persian that reads like translation.
For a brand operating in Iran, that is not a rounding error. Persian behaviour is tested first here, on the same principle applied to type and copy elsewhere on this site: Persian is written natively, never machine-translated.
Automation and characters
Automation covers the unglamorous half — publishing workflows, repetitive production steps, the routine that otherwise consumes the time that should go into the work itself.
AI characters are the opposite end: a persistent persona that can carry a brand across content. Useful when the brand genuinely has a character to extend, and a gimmick when it does not. We will say which one we think it is.
Status
An active capability, offered alongside the strategy and content work rather than as a separate technology product. Published case studies will appear here once clients confirm them.
Common questions
Is this just adding a chatbot to a website?
No. A generic chatbot with a logo on it damages a brand faster than having nothing, because every answer it gives is written in somebody else's voice. The work here is the tone, the boundaries, and what the system is not allowed to claim.
Does it work properly in Persian?
That is the first thing tested, not the last. Many models are noticeably weaker in Persian than in English, and a system that is fluent in one and awkward in the other is not bilingual — it is an English product with a translation layer.
What stops it from saying something wrong about our business?
Explicit guardrails and a defined refusal behaviour. An agent that invents a price, a delivery date or a capability creates a real obligation for the business. Deciding in advance what it must not answer is most of the work.