From Laid Off to AI Engineer in 90 Days: The Realistic Plan

From the archive. Originally published 2026-08-28. Tool details, examples and offers reflect that publication date. See our current guides for newer material.

What this 90-day plan actually is

If you have been laid off and you are trying to move into an AI role, you have probably read a lot of advice that sounds good and helps nobody. One camp tells you to spend six months learning to code from first principles. The other camp tells you that prompting ChatGPT is a career now. Neither of those gets you hired in a quarter.

This plan is different because it works backwards from what hiring managers are actually saying yes to in 2026. They are hiring people who can take a language model, wire it into a real workflow, handle the retrieval and the evals, and ship something that does not fall over in front of a user. You do not need a computer science degree for that. You need a focused 90 days and a portfolio that proves you can do the work.

The structure below is four phases across twelve weeks. Each phase has one job. Do not skip ahead because a later phase looks more exciting. The reason people stall at week seven is almost always that they rushed weeks one and two.

Why most 90-day advice fails

The common plans fail for three reasons. First, they front-load theory you will never use on the job, so you spend a month on gradient descent and never build anything a recruiter can click on. Second, they treat "AI engineer" as one job when it is really a spread from prompt-and-glue application work to full retrieval and evaluation systems, and the plan never tells you which end to aim at. Third, they ignore that you are job hunting the whole time, not just at the end, so the portfolio, the network, and the interview prep all get crammed into the last week.

This plan fixes all three. You build in public from week one, you pick a lane early, and you start talking to people while you are still learning.

The week-by-week plan

Weeks 1 to 2: Foundations you will actually use

Goal: be able to call a model from code, get structured output back, and explain what a token, a context window, and an embedding are without notes.

• Get comfortable in Python at a working level: functions, dictionaries, list comprehensions, virtual environments, reading a stack trace. You are not becoming a software engineer this fortnight, you are becoming dangerous enough to build.
• Make your first API calls to a hosted model. Send a prompt, parse the response, force a JSON schema, handle an error and a retry.
• Learn the vocabulary that comes up in every interview: tokens and pricing, context windows, temperature, system vs user messages, embeddings, vector similarity, hallucination, and why chunking matters.
• Set up your public workspace now: a GitHub account with a real profile, and a short build log somewhere public. Every project from here goes in both.

Weeks 3 to 6: Build three portfolio projects

Goal: three things a hiring manager can open, run, and understand in under five minutes each. Increasing difficulty. Real READMEs with a screenshot, the problem, the approach, and what you would do next.

• Project 1, a structured extraction tool. Take messy input (emails, PDFs, support tickets) and return clean structured data with validation. This proves you can get reliable output out of a model, which is half of real AI engineering.
• Project 2, a retrieval-augmented assistant over a real corpus. Pick a body of documents you actually care about. Chunk it, embed it, store the vectors, retrieve on a query, and answer with citations. Handle the case where the answer is not in the documents.
• Project 3, a small agent that does a multi-step task. Something that plans, calls two or three tools, and produces a result a person would pay for. Keep the scope tight. A working narrow agent beats an ambitious broken one.

Project brief that works

"Build a tool that ingests my last 200 saved articles, lets me ask questions across all of them, and always answers with links to the specific articles it used. If the answer is not in my library, it says so instead of guessing."

Weeks 7 to 10: Make it production-shaped

Goal: turn one of the three projects into something that looks like it belongs in a company, and learn the parts that separate a demo from a system.

• Evals. Write a small test set of inputs and expected behaviour for your best project. Measure accuracy before and after a change. Being able to say "I moved retrieval precision from 71 to 88 percent" is worth more than any certificate.
• Deployment. Put it behind a real endpoint or a deployed app with a URL. Add basic logging, cost tracking, and rate limit handling.
• Guardrails. Handle prompt injection, bad input, and the empty-answer case. Show you have thought about what happens when a user is hostile or the model is wrong.
• Write one technical post explaining a decision you made and the tradeoff. This is your interview talking points, drafted in advance.

Weeks 11 to 12: Job hunt at full intensity

Goal: applications out, conversations booked, interview loop practised. Note that the networking started weeks ago, this is where it compounds.

• Rewrite your CV around the three projects and the metrics. Lead with what you built and shipped, not the courses you took.
• Apply narrow and deep: 5 to 10 well-researched applications a week with a tailored note, not 100 blind ones.
• Message people doing the job you want. Ask what their week actually looks like. Most will answer, and some will refer you.
• Practise the loop: a take-home in the style of your projects, a system design conversation about RAG and evals, and a behavioural round about the layoff told as a forward-looking story.

Pick your lane early

"AI engineer" spans a wide range. Decide in week two which end you are aiming at, because it changes how you spend weeks 7 to 10.

Application-focused

You wire models into products: prompt engineering, tool calling, integrations, UX around AI features. Lighter on infrastructure, faster to a first offer if you come from a product or front-end background.

Systems-focused

You own retrieval quality, evals, latency, cost, and reliability. More depth required, higher ceiling, better fit if you come from data or back-end work.

What makes this work

1. Build in public from day one. A hiring manager who can watch your last eight weeks of commits and posts does not need to take a risk on you.
2. Ship narrow, not ambitious. Three small projects that run beat one big project that is 70 percent done. Scope is the skill.
3. Measure something. One real metric on one project changes every interview. "I improved it" is a claim. "I moved it from 71 to 88" is evidence.
4. Job hunt in parallel, not at the end. Networking has a two to four week lag. If you start it in week eleven, it pays off after your runway ends.
5. Use the layoff as a story, not an apology. "The role was cut, I used the time to go deep on AI engineering, here is what I built" is a strong answer. Practise saying it plainly.

Want to walk through this plan live with Aki?

This is the resource behind the Lightning Lesson of the same name. Join the live session to go through the plan step by step and ask about your own situation. Browse upcoming sessions →