AI agency vs freelancer vs IT services firm (ESN): which to choose
AI agency, freelancer or IT services firm: a decision guide by project type, with sourced day rates, total cost, the risks and when each option wins.
The short answer. AI agency, freelancer or IT services firm: it depends on the project. Hire a freelancer for a precise AI task your team can take over: a connector, an extraction pipeline, an audit. Hire an IT services firm (what the French call an ESN) to add people to a team that already knows what it's building, over the long run. Hire an AI agency when you need a complete product or agent in production, fast, without the team in-house. The right choice depends less on the day rate than on three questions: who owns the outcome, who manages the work day to day, and who takes the system over in a year.
Most comparisons stop at the day rate. Yet it's the number that tells you the least. A €700-a-day freelancer who ships alone in six weeks can cost less than a team billing €1,000 a day. The opposite happens too, when someone has to manage the work and then add the interface, security and infrastructure around it. So we start from your project and get to the vendor after.
The comparison in one table
| Criterion | AI freelancer | IT services firm (ESN) | AI agency |
|---|---|---|---|
| What you buy | an expert's time | people, often on time and materials | an outcome delivered by a team |
| Who manages day to day | you | you, or the firm's project lead | the agency, with a contact on your side |
| Start | fast if the person is available | a few weeks (sourcing) | fast |
| Scope covered | one topic, in depth | broad, depending on who is staffed | full product: front end, back end, AI, infra |
| Scaling up | limited to one person | strong | medium |
| Usual pricing | day rate | day rate, sometimes fixed price | fixed price per phase or capped time and materials |
| Main risk | depending on one person | interchangeable profiles, heavy management | lock-in if nothing is handed over |
| Wins when | the need fits in one sentence | volume and duration matter most | you need to reach production fast |
These cells describe tendencies. They aren't rules. Some IT services firms are very strong in AI, and some freelancers work in collectives. Look at shipped projects rather than the label.
What each option really costs
Day rates, with their sources
The most solid public numbers are for freelancers in France:
- Experienced freelance data scientists: €681 a day on average, per Malt's data scientist rate barometer (profiles active over the last three months).
- Experienced freelance developers: €576 a day on average, per Malt's developer rate barometer.
- AI / LLM freelance specialists: a median of €810 a day, with half of reported rates between €680 and €940, per TJMètre's AI barometer. The sample is small (41 observations), so read it as an order of magnitude.
For IT services firms, La Fabrique du Net reports day rates from €320 to €2,000 depending on the profile, and estimates that a mid-level developer for three months comes to roughly €25,000 to €35,000.
For AI agencies, there's no reliable public benchmark. The ranges you see on blogs come from the agencies themselves. Ask for a price per phase with a deliverable, and compare it with the total cost of the other options.
Total cost, not the day rate
For an honest comparison, add up:
- The service: days × rate, or the price of each phase.
- Your management time: specifying, deciding, testing. On time and materials, that time sits with you, and it's often underestimated.
- What is missing around it: a freelancer who knows models won't necessarily build the interface, authentication, hosting or security. Someone will have to.
- Running costs: model calls, hosting, monitoring.
- What comes next: maintenance, updates when models change, handover.
An AI project never stays frozen. Models change every few months, and so do your data and processes. The cheapest option on day one is rarely the cheapest after a year.
When each option wins
The freelancer wins when the need fits in one sentence
"Extract amounts and dates from our PDF contracts into our ERP, with an error rate measured on 200 documents." That's a good freelance brief. The scope is clear, so is the success criterion, and someone on your side can run the result.
Freelancers also win on rare expertise: an audit of your RAG architecture, a model evaluation, help on a hard problem for a team that already exists.
It breaks down when the project needs interface, back end, infrastructure and security all at once, or when the person leaves halfway. Require the code in your repository, documentation as you go, and access in your company's name from day one.
The IT services firm wins when volume and duration matter most
You have a product team and a roadmap, and you're short of hands for eighteen months? An IT services firm can supply several profiles, replace them, and get through your procurement and security questionnaires. That's its job.
The flip side: on time and materials, you manage the work. If the AI need is still fuzzy, you pay for days spent finding out. In France, time and materials also has a legal frame: the labour code bans any for-profit operation whose sole purpose is lending staff. A sound engagement relies on real expertise and supervision by the vendor. Consultants directed day to day like employees are a different thing. Your legal and procurement teams know the topic. Ask them before you sign.
The AI agency wins when you need production, fast
The typical case: an AI product or agent that has to run inside your tools, with real data, access rights and usage metrics, within a few weeks. Nobody in-house has the time or the full skill set. The agency owns the outcome with a small team that covers the whole path.
The risk: ending up dependent on a system nobody on your side understands. Put it in the contract from the start: ownership of the code, prompts and test sets, and a handover with documentation and runbooks.
Eight situations, one recommendation
| Your situation | Best-fit option | Why |
|---|---|---|
| A single connector or automation | Freelancer | clear scope, easy takeover |
| An audit or expert opinion on your AI stack | Freelancer | one-off expertise |
| Reinforcing an existing product team for 12+ months | IT services firm | volume, replaceable profiles |
| A program with dozens of developers | IT services firm | capacity, procurement, framework |
| A first AI agent in production in your tools | AI agency | full product, short timeline |
| A new AI product, from mockup to launch | AI agency | cross-functional team |
| Validating a use case before hiring | AI agency, then bring it in-house | learn before you hire |
| Industrializing AI across your own dev teams | AI agency embedded in your teams | standardize tools and practices |
Combine instead of choosing
The setups that work best often mix options. An agency builds the first system and trains the team while you hire. A freelancer then takes over maintenance. An IT services firm reinforces the teams once the method is in place.
The Theodo case is one example. Theodo delivers critical projects for its clients and brings in The Vibe Company as a technical expert embedded in its teams. Together, we build software factories for hundreds of developers: standardized tools, agents and skills, plus the support and training so every team adopts them. In this setup, the agency doesn't replace anyone: it gives the existing teams a shared base so they move faster.
A software factory also helps harden the product. A security rule forgotten in a single repository is enough to open a door. When agents and skills are shared, a secrets or dependency check applies everywhere at once. We cover this approach in AI security agents to harden a SaaS and in our guide to the enterprise agent harness.
The questions that decide it
Before you sign, whatever the option, get written answers to these:
- Who owns the outcome if the criteria aren't met?
- Who, by name, works on the project, and for how much time?
- What is the first usable deliverable, and when?
- How is the AI's quality measured, and on what data?
- Who owns the code, prompts, tests and documentation?
- What happens if the key person leaves?
- What recurring costs come after delivery?
To go further on judging a proposal and a POC, see how to choose an AI agency. For what comes next, see the stages of a project with an AI development agency and the POC-to-production checklist.
FAQ
AI agency or freelancer: which is cheaper?
Per day, the freelancer almost always is. Per project, it depends on what's left to do around them: management, interface, infrastructure, security, maintenance. Compare the total cost over a year, not the day rate.
How much does an AI freelancer cost in France?
Per Malt's barometer, an experienced freelance data scientist charges €681 a day on average. TJMètre gives a median of €810 for AI / LLM specialists, on a small sample. Rare profiles with production experience sit above that.
What's the difference between an IT services firm and an AI agency?
An IT services firm mostly supplies people who reinforce your teams, usually on time and materials, under your management. An AI agency takes a project end to end with its own team and a deliverable. The line is blurry: check who owns the outcome in the contract.
Can you start with a freelancer and move to an agency later?
Yes, if the code lives in your repository and is documented. The reverse works too: an agency ships and hands over, a freelancer maintains.
Which option fits an AI software factory?
You need people who have already standardized tools, agents and skills across several teams, and who train your developers. That's rarely a job for a lone freelancer. An agency embedded in your teams, or a dedicated in-house team, fits better.
Not sure which option fits?
The Vibe Company is an AI agency in Paris. We build AI products and agents that run in production, and software factories for development teams. In 30 minutes, we look at your project and tell you which option makes sense, even if it isn't us.
