Freelancing and consulting as an AI engineer
Clients pay for solved problems, not for AI. Start every conversation with "show me how you do this today and tell me what exactly you want to improve", scope the work around that process and a measurable outcome, write down what you need from the client's team and how the result will be checked, and show proof from previous work through write-ups and case studies. Clients come to you most often when you've already solved, and written about, the problem they have.
Why AI projects with clients fail
Clients often arrive with the technology already chosen: "we want an agent", "we want to fine-tune". The project then fails because nobody defined the problem, the expected output or how success would be measured. Saying no to unnecessary AI builds trust, and sometimes the right answer is a Python script.
A proposal has three parts
| Part | What it says |
|---|---|
| Scope | The problem, the expected output, and what's not covered |
| What's needed from the client's team | Data access, a domain expert's time, decisions and who makes them |
| How the result will be checked | The metric, the test set, the period |
Questions to ask before quoting
- How is this done today, by whom, and how long does it take?
- What does a wrong output cost?
- What data exists, where, and who can give access?
- Who will maintain this after delivery?
- If an existing product already solves it, why not use that?
Finding clients
Clients ask the same thing interviewers do: show me your work, have you solved a problem like ours? If you've already solved that problem and written about it, the client is much more likely to reach out or give you the project. Share useful work in public and in communities, consistently, rather than posting "I'll do your work". Outbound works better when you research a company's specific problem first and show something relevant to it.
Pricing basics
Price from the value of the outcome and the effort, not from hours alone, and separate the build from ongoing maintenance and model costs, which the client pays for as long as the system runs. Check market rates for your role and region before your first quote.
Case studies
Agree on a case study clause before the work starts, and get permission before naming the client, showing screenshots or quoting numbers. Keep estimated value separate from measured results.
Job or freelancing first
Focus on one. For someone early in their career, a job first teaches how the industry works, and freelancing can start as side work.
From the floor
A client once came with a request to integrate an LLM, and I said no, because the same thing could be done with a Python script. Not all problems need AI, and that answer is part of why clients trust the recommendations that do involve AI.
In a career webinar, a builder asked for freelancing tips, and my advice was that if a client is trying to solve a problem you've already solved and shown in your blog or portfolio, they will reach out or give you the project, so join communities and share valuable content instead of asking for work. I also said that running free sessions consistently brings clients over time, sometimes people you didn't know were clients when they joined.