GUIDE / AI ARCHITECTURE AND AI ENGINEERING
AI architect or AI engineer?
Who does what.
An AI architect decides how AI fits into a company’s work: which tasks it takes on, which data it may use, how its output is checked, what it costs and where a person decides. An AI engineer builds that system: the code, the integrations, the tests and the production setup. Large teams hire the two separately; a small or mid-sized company is often better served by one person who does both.
What an AI architect does
The architect answers one question before anything is built: should AI do this work, and how does it fit? In practice, that means deciding:
- Which tasks AI takes on, and which it should not.
- Which data the system may use, and under which privacy obligations, such as Québec’s Law 25.
- How the output is checked: evaluations on real examples, verification, and the points where a person approves.
- Which models and providers fit, compared on your own examples for quality in your language, cost, privacy and where data is processed.
- What each use costs, and how that cost is tracked.
- Who runs the system after launch, and how it is monitored.
What an AI engineer does
The engineer answers the next question: how do we build it so it works every day? That means:
- Writing the code that connects models to your systems and data.
- Building search over your own documents, the agents and the tools they may call.
- Writing the evaluations and tests that measure quality on real examples.
- Setting up deployment, monitoring, logs and cost tracking.
- Fixing what fails in production, and improving the system over time.
Side by side
| Aspect | AI architect | AI engineer |
|---|---|---|
| Main question | Should AI do this work, and how does it fit? | How do we build it so it works every day? |
| Works on | The whole system and its constraints: tasks, data, people, law and budget | The parts: code, integrations, tests and deployment |
| Delivers | A design: what AI does, which data it uses, how its output is checked, where a person approves and what it costs | Working software in production, with its tests and monitoring |
| Typical skills | System design, evaluation, privacy and governance, cost, and working with the people who do the work | Programming, model APIs, data pipelines, testing, deployment and operations |
| Without it | A pilot that impresses in a demo but doesn’t fit the work, the law or the budget | A design that never leaves the slide deck |
Which one does your company need?
- Unsure where AI could help
- Architecture first. List the tasks, rank them by value, risk and effort, and check the privacy obligations before any data moves.
- One use case chosen, not yet proven
- Both. Design a small pilot with success criteria set in advance, then build it and measure it on your real data.
- A pilot that works and must run every day
- Mostly engineering: security, monitoring, cost control and the human checkpoints that remain, with the design checked again before launch.
These are the three stages I work in: explore, pilot and deploy.
See the three stages ↗In Québec, a word about the title “engineer”
In Québec, only members of the Ordre des ingénieurs du Québec may use the title “engineer”. Job ads often say “AI engineer” for the building work described above, but the title itself is regulated. That is why this site says AI architect, and calls the building part AI engineering, or développement en IA in French.
SourceOrdre des ingénieurs du Québec, on who may use the title ↗
Why I do both
I design the system and write the code myself, so the design is proven in working software rather than left in a slide deck. The person who decides where someone must approve also builds that approval step, tests it and runs it.
Projects where I did both:
- Mnemosyne: a bilingual learning product in production with real learners, with memory and a layer that verifies what the tutor says.
- HaulView · Iris: an operations product in development, where Iris prepares each offer and a dispatcher approves it. Client details stay confidential.
- Adra: a multi-agent review prototype built solo in five hours, top 4 of about 30 teams at Mila Ventures’ Build OS26.
Questions
-
Is an AI architect the same as a solution architect?
Not quite. A solution architect designs how software systems fit together. An AI architect does that for systems where a model produces part of the work, which adds questions classic software rarely asks: how the output is evaluated, where a person must approve, what each use costs and which data a model may see.
-
Does an AI architect write code?
Not always. Some design the system and hand it to a team. I write the code myself, so the design is tested in working software.
-
Is an AI engineer the same as a machine learning engineer?
They overlap. A machine learning engineer usually trains and tunes models. An AI engineer more often builds products around existing models: connecting them to data and tools, testing them and running them in production.
-
Do we need a data science team to start?
No. You need a problem worth solving, someone who knows the work well, and examples of the documents or data involved.
-
How much does an AI architect cost in Montréal?
It depends on the stage and the size of the problem, so I don’t publish fixed prices. After a 30-minute call, you get a quick written quote for the stage you need, before any commitment.
Not sure which one you need?
Let’s look at it together.
Book a 30-minute call, in French or English. You get a quick written quote, before any commitment.
Or write to me directly: elias.t@inquisitiveflow.ca