If your search is how to learn AI from scratch and turn it into paid work, this is your start-here page: 6 phases in order, each with honest difficulty from 1 to 5, real time estimates, a clear exit test so you never lie to yourself, and free vs paid paths. Read it once fully, then execute it top to bottom.
How to use this roadmap
One rule: never advance without passing the exit test. Tutorial hell — 40 hours of video with nothing published — is cured by measuring output, not hours. The reference pace is 5 hours a week; scale the timelines to your life, but keep the order.
The free path costs $0 beyond internet: free tiers of the chats, the free guides linked here, and trial-mode automation tools. Paid is an accelerator, never a requirement, compared on pricing. If you want the gentlest possible preamble first, read AI for beginners: where to start and the free online AI course for beginners before phase 1.
Phase 0 — Anchor case and goal (difficulty 1, one evening)
Before studying anything, write one concrete sentence: what job do you want AI doing for you within 90 days? "Summarize 15-page reports into 1 page" or "produce my shop's 12 monthly posts". That sentence is your anchor case: every phase gets applied to it first. People who study AI "in general" quit; people solving their own problem continue.
Exit test: sentence written plus one real document from your work set aside for exercises. Career-switchers should also read the AI learning path for career changers to see how these phases map onto a full job change.
Phase 1 — Professional chat fundamentals (difficulty 1, 2 weeks)
Learn to talk to AI like a professional: give context, demand exact formats, iterate, and verify every answer. The classic mistake is treating chat as an oracle; professionals treat it as a blazing-fast junior that needs review.
Drills: build a library of 20 prompts from your real work, request the same text in 3 tones, ask the AI to attack its own answer, and compare its draft against what you would write by hand until you reach 80%-usable output. The beginner guide above is your daily script for these two weeks; the free online course structures the same ground.
Exit test: 20 saved reusable prompts plus 3 real work deliverables made with AI. No real usage, no advance.
Phase 2 — Files, spreadsheets, reusable templates (difficulty 2, 2 weeks)
Empty chat is a toy; AI with your files is a tool. Learn to attach PDFs and spreadsheets, demand actionable summaries with risks and next steps, extract tables from long documents, and build templates that turn a messy recurring document into a finished deliverable in 15 minutes.
Phase project: take the heaviest recurring document in your work and build a prompt template that converts it to final form. Measure before and after in minutes — that number goes into your portfolio later.
Exit test: 1 documented template saving at least 1 hour a week, time measured. Difficulty 2 because real files arrive broken, messy, and in the worst format imaginable.
Phase 3 — Light automation (difficulty 3, 2–4 weeks)
The phase that separates users from billable professionals: one flow that works while you sleep. A single useful flow is enough — triage inbound messages and draft replies, or turn form submissions into draft proposals plus alerts. The full playbook is how to automate repetitive tasks with AI: follow it end to end.
Difficulty 3 comes from accounts, API keys, tests, and the mysterious breakage on demo eve — normal, and part of the learning clients pay you not to suffer. Free tiers and trials cover everything here; the only sensible early spend is cheap hosting so a client flow runs 24/7.
Exit test: 1 automation genuinely running for 7 straight days, with an execution log and hours saved.
Phase 4 — Portfolio with 2 documented projects (difficulty 3, 2–4 weeks)
Nobody hires "AI knowledge"; they hire proof. Publish one simple page — a shared doc works — with 2 projects: your phase-2 template and your phase-3 automation. Each follows before/after/how: the problem, what you built, time or money saved with numbers, screenshots as evidence. The hands-on AI projects guide gives you extra project patterns if you want a third piece.
Exit test: a published page with 2 number-backed projects. No numbers, no portfolio — just promises.
Phase 5 — Specialize by field (difficulty 2–4, 2 weeks)
Generalists learn; specialists charge. Pick ONE lane and go deep: B2B sales with AI plus AI for sales teams for revenue roles; AI training for marketing teams for marketers; how to choose AI sales training if you sell enablement itself. The right specialization combines your work history with demand you can reach — an ex-salesperson sells prospecting automation better than a technician with no network.
Exit test: a 5-minute plain-language pitch of exactly what problem you solve, how, and for how much — tested on 3 real people.
Phase 6 — First sale and pricing (difficulty 4, 2–4 weeks)
The hardest phase is commercial, not technical. The honest plan: 1 free or steeply discounted job for a testimonial plus result numbers, then 2 entry-priced paid jobs, then raise. Realistic 2026 entry points: simple automation projects $300–$1,500; monthly content or support retainers $400–$2,000; paid diagnostics $150–$500 credited toward the project. Starting points, not a rate card — proof moves them up.
Outreach order: your existing network first, then local businesses with visible pain (slow replies, dead socials, manual spreadsheets), carrying your phase-4 portfolio and phase-5 pitch. Every 10 conversations, revise the pitch; every 30 without closing, revise the offer.
Exit test: 1 delivered job with testimonial plus 1 paid job closed. From here the roadmap becomes operations: deliver, document, prospect, repeat.
Free path vs paid fast lane
Phases 0–2 are 100% free with this roadmap and the linked guides — spend nothing before finishing phase 2, since you do not yet know whether you prefer automation, content, sales, or code. In phase 3, the useful spend is cheap flow hosting, not a course. From phase 4, paid makes sense as acceleration: structured courses with certificates and ready-made libraries. All compared at /pricing: full access for completists, focused tracks per phase.
What NOT to study (saves you 3 months)
No neural-network math unless you want to do research — zero value in your first 200 applied hours. No 5-tools-per-category collecting; master one chat and one automation tool. No "advanced prompt engineering" course before you have 20 real prompts failing in production. No chasing weekly model launches; switch models quarterly, not per headline. And no second programming language before your first AI dollar. AI rewards shippers, not collectors.
Mistakes that kill the roadmap (and the fixes)
Collecting courses with zero output is killer number one — the exit tests exist to prevent exactly that. Number two is studying tools instead of problems: nobody pays for "knowing n8n", they pay for "never losing a weekend lead". Number three is prospecting on promises instead of portfolio numbers. Number four is comparing your week 1 to a creator's year 3. Number five is quitting in phase 3, the most technically frustrating: shrink the flow, ship smaller, but ship.
Your next 4 weeks
Week 1: phase 0 plus your first 10 phase-1 prompts on the anchor case. Week 2: all 20 prompts plus 3 real AI-assisted deliverables. Week 3: the phase-2 template with measured savings. Week 4: choose automation (phase 3) or straight to portfolio if your network is commercially active. Mark today-plus-90-days on the calendar: the honest deadline for paid gig number one at 5 hours a week. Start now, start small, but start on real work — tell one person your 90-day deadline, because social commitment beats private intention — and when you need structure, the courses and pricing are here.
The 10-term minimum dictionary
You need no more vocabulary than this to finish. Model: the text-predicting engine. Prompt: your instruction. Context: what the model "sees". Hallucination: confident invention — hence verification. Token: the unit of usage and cost. Context window: how much it holds at once. Agent: a loop of steps with tools — automation with decisions, not magic. Workflow: the fixed sequence that runs alone. RAG: connecting the model to your documents so it answers from your data. Fine-tuning: retraining a model — expensive and almost never needed for your case. Everything else is combinations. If a course starts anywhere else, start another course.
The certificate question: when it matters and when it does not
Certificates never closed a deal or got anyone hired on their own — portfolio proof does that. But certificates do three real jobs: they structure your learning when you cannot self-direct, they signal seriousness to employers screening career changers, and they unlock team training budgets that never approve "a guy from YouTube". Get certified in phase 4 or 5, never in phase 1: a certificate with no portfolio behind it reads as exactly what it is. If you are freelancing, spend the certificate money on flow hosting first; clients buy outcomes. If you are job hunting or upskilling inside a company, the structured courses with certificates pay back in interviews and promotion packets. Match the credential to the gate you actually face.
Sticking to it: the streak system
Motivation is weather; systems are climate. The system that works: one fixed weekly block (or daily 45 minutes), a visible tracker with the phase exit tests as checkboxes, and a rule that a missed session means repeating the last completed step — never skipping ahead to "catch up". Find one accountability point: a friend, a study group, or a public build-in-public thread. And keep an evidence file from day one: every prompt that worked, every hour saved, every screenshot. On the day you want to quit (around week 4 for most people, right in the automation frustration zone), the evidence file reminds you how far zero-value tutorial watching never took you. Finishers are not more talented; they just never broke the chain twice.
If you only have 2 hours a week
The roadmap still works, stretched to ~20 weeks: one phase per month instead of per fortnight. What fails is scattered 20-minute snacking with no output — with little time, every session must end with something saved. Keep one fixed weekly block, guard it like a medical appointment, and apply everything to the anchor case so threads never go cold. Slow and steady beats intense and intermittent: 80% of quitting comes from broken streaks, not lack of talent. If a month stalls, repeat the last completed phase instead of leaping ahead — the roadmap stretches, it never skips.
Frequently asked questions
How long does it take to go from zero to paid AI work?
Eight to sixteen weeks at 5 hours a week per the phase breakdown. Computer workers with a real anchor case trend to the short half; total beginners with a full-time job should plan the full range without guilt.
Do I need to learn programming to work with AI?
No to start. Prompts, no-code automation, and ready agents cover most bought services; programming is an accelerator from phase 4.
What is the free path and what does paid add?
Free uses this roadmap, blog guides, and free tool tiers. Paid adds structure, certificates, and ready libraries — compare on pricing.
Where do I literally start today?
Phase 1, exercise 1: one real work task solved in chat today, with the beginner guide as your daily script. Nothing purchased before week one ends.
I tried AI before and was disappointed. Will this be different?
Usually disappointment means context-free prompting, which phase 2 fixes. Retest with context and demanded format.
What comes after the 6 phases?
Track A (business automation services), B (freelancing/consulting), or C (AI inside your current job), each with linked continuation guides.
