How to Start an AI Career in 2026 Without Learning to Code

You do not need to code to win in AI

The most reassuring line in this conversation between Liam Ottley and AI engineer Dave Ebbelaar is that you do not need to learn to code to succeed in AI, and it comes from someone whose entire background is code. There are many levels to play the game at. You can simply use the tools, you can train others to use them, you can consult, or you can go all the way to full AI engineering. The level you pick depends on the kind of success you want, so the first job is defining what success means for you: a job, quitting a job, or building an agency.

The structural shift matters. The old route into AI was code first, then consulting on top. Now you can flip it: start at the top with education and workshops, move into consulting, and drop into development only if and when you want to.

The non-technical paths that actually work

For people who are not going to write code, the conversation lays out concrete entry points:

• Go into businesses and train staff to use ChatGPT, described as the lowest-hanging fruit because the productivity data sells itself
• Learn automation on make.com or n8n, with make.com recommended first as the more beginner-friendly base skill
• Set up MCP servers inside the ChatGPT and Claude tools teams already use

- Go into businesses and train staff to use ChatGPT, described as the lowest-hanging fruit because the productivity data sells itself - Learn automation on make.com or n8n, with make.com recommended first as the more beginner-friendly base skill - Set up MCP servers inside the ChatGPT and Claude tools teams already use

The MCP idea is the standout offer. Instead of a big top-down build, you embed AI where people already work, which automatically keeps a human in the loop. A sales rep can pull calendar and CRM data inside ChatGPT, and the pitch becomes I can make your team five percent more productive in two weeks with custom MCP servers. It is small, standardizable, and it spirals into consulting and larger builds.

The safest bet I have seen in entrepreneurship is this space, if you just apply yourself, keep your ears open, and participate in communities.

If you do go the developer route

For those who want the technical path, Ebbelaar names six components to take a Genai project to production: LLM fundamentals like tokenization and prompt engineering, system design and cognitive architecture, packaging a production-ready app with Docker, RAG with vector databases and reranking, monitoring with guardrails and evaluations, and deployment. He is candid that the last ten percent is brutal, where getting from eighty to eighty-five percent accuracy can take months and still not be enough.

His freelance model is worth copying. Rather than juggling many small clients, stack one long secure contract at engineer rates with smaller higher-paid fixed-price projects on top, which avoids the feast-or-famine cycle.

On university and the platform shift

The verdict on university is blunt: the credential can still open certain doors, but the AI curriculum is outdated before you graduate. To learn fast, watch what creators are building on YouTube and build your own projects to solve real problems. Experience talks, and four to five years in the trenches beats a piece of paper for most paths.

Finally, watch the platform consolidation. OpenAI's agent builder is GenAI-native with guardrails and evals baked in, echoing the assistants API that nuked a wave of startups. Expect an Apple-style closed ecosystem with ChatGPT at the center. The agency upsell that follows is selling a private AI platform that manages shadow AI risk, runs every model, and reveals who the power users are, which in turn shows you the next thing to build. The window is still wide open, and the people who start building now are the ones who benefit.