Skip to content
jCode

Freelancer, agency or hire: who should build your AI feature?

Choose a freelancer for a defined feature, an agency for several profiles at once, and a hire when AI is your product's core. I am a freelancer myself, so the guide applies its checks to my own offer too.

Published 13 October 2026 · Jean-Baptiste Thery · EURL jCode

001
Context

LLM API integration and orchestration make up 45% of AI projects, search across documents 30% and agents 15%, according to Malt's Tech Trends 2026 report, published in April 2026.

What it means for you

Most AI features are integration work in an existing product: choosing a model, connecting your data, building the interface and checking the answers. That is software engineering as much as data science.

Back to the options

002
Option 1

A freelancer fits a defined feature in an existing product, when you need senior judgement quickly. On Upwork, earnings from AI integration work grew 178% in 2025.

The signs

The scope is clear, your team will own the feature afterwards, and working remotely suits you.

What to check

AI features they have put into production, how they test them, and how they handle your data. Plan documentation and a handover from the start.

In my case

Between 2022 and 2025, I built the frontends of AI products for Stellia.ai (incubated at École Polytechnique), Sweez and Leadbay (Y Combinator F25). I also co-founded and built WorkoutGen, a personalised workout generator whose AI coach suggests weights and adjusts volume and rest after every session.

WorkoutGen had 547 automated test files in October 2026. On client code, AI tools follow the client's rules and use only business or API accounts that do not train on it. Documentation is part of the delivery, as on my three Sciences Po projects.

Back to the options

003
Option 2

An agency fits when the feature needs several profiles at once.

The signs

The project needs design, mobile, data and backend work in parallel, or you want one supplier with an account manager.

What to check

Who will actually do the work, their experience with AI features in production, and the agency's written policy on AI use.

Back to the options

004
Option 3

Hiring fits when AI sits at the core of your product and needs care every week. Expect competition: in June 2026, Malt reported that demand for agentic AI expertise had grown 5,800% in twelve months, more than twice as fast as supply.

The signs

The feature drives your product, its models and data change often, and you need someone accountable for it over years.

What to check

Experience shipping AI features rather than only prototypes, and the ability to explain and test them without the tool.

Back to the options

005
In every case

Ask a freelancer, an agency or a candidate the same five questions. Rules for AI-generated code are still rare: in a survey of 307 technology leaders published in June 2026, only 4% had governance covering it.

Five questions

Who reviews the code, what agents may do without approval, where your code goes, whether a speed-up is promised and whether the person can explain what they deliver. A separate guide takes each one in turn, with its sources.

Back to the options

006
Going further

If a freelancer fits

ServiceAI features for your web applicationGuideFive questions to ask a developer who uses AICommitmentsHow I use AI in client work
007
Contact

Let's talk about your project.

Tell me what you want to build or improve, your budget and your target date.

  • Fully remote
  • English or French
  • France · UK · US
Describe your project on MaltAsk a question on LinkedInDownload my CV