Break Into Tech in 2026: The 90-Day Plan for Unemployed South Africans
The SA junior dev market is full. The remote AI-fluent builder market is wide open. Here's the 90-day plan to ship 3 real projects and apply remote, in dollars.
By Akhil Boddu · AfricanTechBroRead this first
If you are unemployed in South Africa right now and trying to break into tech, you have probably been told the wrong thing. Get a CS degree. Send out CVs. Apply for SA junior dev roles. That pipeline is full.
Here is the actual 2026 picture, with receipts:
- Stats SA QLFS Q4 2025: 57% youth unemployment for ages 15-24. 43.8% for ages 15-34. (source)
- Stanford AI Index 2026: software dev jobs for ages 22-25 are down ~20% since 2022. (summary)
- Wellfound 2026: average remote software engineer comp is $124K/yr (~R2.2M). Range $45K-$210K. (data)
- OfferZen via TechCentral: SA junior devs are getting squeezed because companies hire AI-augmented mid-levels instead. (article)
The thing that "killed" SA juniors (Cursor, Claude Code, Copilot) is the same thing that gets you hired in 2026, if you use it daily and ship.
I run a bootcamp called Zaio so I am biased. But this guide is honest, names every option, and tells you the actual median time to a remote offer, not the hype number.
Honest timeline. 90 days to ship 3 portfolio projects + first 5 applications. 6-12 months for a realistic remote junior offer if you grind daily. 12-18 months is the median for self-taught devs (Career Karma data). If anyone tells you 90 days to a remote SWE job, they are selling you something.
Who this is for
- 22-30, unemployed, in South Africa or anywhere in Africa
- Either has no CS degree, or has one but can't get hired
- Has a laptop and ~5 hours/day to work the plan
- English-comfortable enough to write a LinkedIn post (we will give you templates)
If that is you, this guide is the path.
The wedge in one sentence
Stop applying for SA junior dev roles. Apply remote-first as an AI-fluent builder. Same effort, ~10x the rand.
flowchart LR
A[Unemployed in SA] --> B{Old path}
A --> C{New path}
B --> B1[CS degree or generic bootcamp]
B1 --> B2[Apply for SA junior roles R18-25K]
B2 --> B3[Pipeline full -20% globally<br/>Most never get hired]
C --> C1[Pick one stack + AI-fluency drill]
C1 --> C2[Ship 3 real projects in 90 days<br/>Cursor / Claude Code daily]
C2 --> C3[Post each project on LinkedIn + X]
C3 --> C4[Apply remote in USD<br/>Wellfound, Turing, Toptal, OfferZen]
C4 --> C5[6-12 mo: first remote offer<br/>$45-80K junior, $80-124K mid]
The 90-day plan, week by week
Week 1-2: Foundations + AI-tools setup
Week 3-6: Project 1 — clone an existing tool with Cursor/Claude Code
Week 7-9: Project 2 — solve a real problem you actually have
Week 10-12: Project 3 — ship in public + start applying remote
You are not "learning to code in 90 days." You are shipping 3 real things in 90 days while learning the tools recruiters now screen for.
Week 1-2: Foundations + AI-tools setup
What you are doing
Pick one stack, install the AI tools, and get to "I can build a button that does something" by end of week 2.
Pick one stack: Next.js + TypeScript + Postgres
Reason: this is what most remote-friendly companies hire for in 2026. JavaScript + React is the largest hiring surface on Wellfound, OfferZen, RemoteOK. Don't pick Java, .NET, or Go for your first stack — fewer remote junior roles, higher gatekeeping.
| Tool | What for | Cost | Link |
|---|---|---|---|
| VS Code | Editor, free, runs everywhere | Free | code.visualstudio.com |
| Cursor | AI-native IDE, the new default | Free tier ok, $20/mo Pro | cursor.com |
| Claude Code | Terminal AI agent, ships features | $20/mo Pro | claude.com/claude-code |
| GitHub | Where your code lives | Free | github.com |
| Vercel | Deploy your app in one click | Free hobby | vercel.com |
| Supabase | Database + auth, batteries-included | Free tier | supabase.com |
Free curriculum, pick ONE and stick to it
Don't tab-hop between courses. Pick one and finish through "build your first full-stack app." All four below are honest paths.
| Path | Best for | Time to first project | Cost |
|---|---|---|---|
| The Odin Project | Self-discipline, free, deep | 6-10 weeks | Free |
| freeCodeCamp | Cert-driven, structured | 4-8 weeks | Free |
| Zaio (I run this — disclosure) | Want a structured cohort + accountability + SA mentors | 4-6 weeks | Paid |
| QCTO occupational certs | Want a recognized SA qualification | 6-12 months | Subsidised options exist |
Tip. If you can self-teach and stay disciplined, freeCodeCamp + Odin Project are excellent and free. If you have tried that and bounced, a paid cohort (Zaio or another) buys you the structure. There is no shame in either path. Pick the one you will actually finish.
What "AI-fluent" actually means
This is the gatekeeping skill of 2026. Recruiters screen for it. Most people fake it. Be honest with yourself:
- You can describe a feature in plain English and have Claude Code or Cursor draft it
- You can read the AI's output and spot when it is wrong
- You can debug AI-generated code without re-prompting in a panic
- You know when to write code by hand vs delegate to AI
- You commit AI-assisted code with a clear message and PR description
If you can't do all 5, you are not AI-fluent yet. That's fine — by end of project 1 you will be.
Daily drill (every day, weeks 1-2)
- Spend 60-90 min on your chosen curriculum.
- Spend 30 min using Cursor or Claude Code to build something tiny (a button, a form, a list). Commit to GitHub.
- Read one real codebase on GitHub for 15 min (e.g. shadcn/ui — readable, modern Next.js).
Done when: you can spin up a new Next.js app, push it to GitHub, deploy to Vercel, and it loads.
[SCREENSHOT: Cursor IDE with a Next.js project open + a Claude Code terminal session]
Week 3-6: Project 1 — clone an existing tool
What you are doing
Pick a tool you actually use and rebuild a stripped-down version of it. Not from scratch. With Cursor or Claude Code as your pair-programmer.
Why a clone first
You learn faster from rebuilding something that already works than from staring at a blank canvas. The reference is the spec.
Pick one of these
| Clone target | Stack drill | Why it teaches you a lot |
|---|---|---|
| Linktree clone | Auth, single DB table, public profile pages | Auth + dynamic routes + SEO |
| Pomodoro timer + history | State, localStorage, charts | React state and persistence |
| URL shortener | Forms, DB writes, redirects, analytics | Full-stack request lifecycle |
| Lightweight Notion clone | Rich text editor, tree of pages | Real product complexity |
| Habit tracker | Auth, daily check-ins, streaks | DB design + cron-like UX |
How to do it
- Day 1: write a one-page spec in plain English. What it does, what it doesn't.
- Day 2: ask Claude Code to scaffold the skeleton. Commit.
- Days 3-20: build features one at a time. Every commit message names the feature.
- Days 20-25: deploy to Vercel. Get a real URL.
- Days 25-28: write a README a stranger could understand.
Watch out. Don't add 17 features. The grader (recruiter) wants to see "this person finishes things," not "this person is ambitious." Three working features beat ten broken ones.
The AI-fluency move
Every time Cursor or Claude Code writes something you don't fully understand: stop, read it, ask it to explain it, then refactor part of it by hand. That loop is the entire skill.
[SCREENSHOT: Vercel dashboard showing a deployed project with a real URL]
Done when: the project is live on a Vercel URL, has a README, has at least 30 commits, and you can demo it in 60 seconds without notes.
Week 7-9: Project 2 — solve a real problem you actually have
What you are doing
Build something you would use yourself. This is the project that goes at the top of your portfolio, because the story behind it is real.
Why a real problem
Recruiters are tired of TODO apps and movie databases. A scrappy tool that solves your own annoyance is more memorable than a polished clone of something that already exists.
How to find a real problem
Open your phone. Look at the last 3 things you Googled this week. The annoyance behind one of those is your project.
Examples that have worked for SA-based devs:
- An app that scrapes minibus taxi rank schedules in your city
- A WhatsApp bot that splits restaurant bills with friends
- A site that tracks load-shedding stage history with notifications
- A tool that converts ZAR to USD with the EFT fee baked in
- A dashboard that pulls your bank's CSV and categorises it
How to do it
- Days 1-2: write the spec + the user story. One sentence: "I want to ___ so that ___."
- Days 3-5: build the dumbest version that solves it for you alone. No auth, no polish.
- Days 6-15: add auth, deploy, share with 3 friends, fix what they complain about.
- Days 16-20: write a launch post for LinkedIn (template below in week 10).
Tip. Resist scaling the idea. The portfolio piece is "I had this problem, I built this thing, here is the GitHub." Bigger is not better.
Bonus AI-fluency drill
Use Claude Code to write the tests for the project. Read every test before committing. If a test passes for the wrong reason, fix it. This is the senior-engineer skill that compresses 5 years of experience into 5 weeks.
Week 10-12: Project 3 — ship in public + apply remote
What you are doing
Two tracks in parallel: build project 3 with public daily updates, and start applying remote.
Project 3: ship in public
Pick something AI-flavoured. Recruiters in 2026 want to see you can deliver an AI-augmented product, not just consume AI tools.
Ideas:
- A summariser for a niche newsletter (RSS in, summary out)
- An AI study-buddy for a specific SA matric subject
- A "Cursor for X" — pick a workflow and embed AI into it
- An AI agent that watches a GitHub repo and posts daily commit summaries to a Slack channel
Build in public:
- Daily LinkedIn post during weeks 10-12. Short. Specific. Show code.
- Tweet/X thread per week with the demo.
LinkedIn post template (steal this)
Day 14 of building [project name].
Problem: [one sentence].
What I shipped today: [one sentence].
Tool I leaned on: [Cursor / Claude Code / specific package].
What broke: [one honest thing].
GitHub: [link]
Tomorrow: [one sentence].
That format works because it shows you ship, you read your own code, and you are honest about what broke. Recruiters DM people who post like this.
Apply remote — where to actually apply
| Platform | What it is | When to apply | Link |
|---|---|---|---|
| Wellfound (formerly AngelList) | Startup remote roles, weekly | Week 10+ once project 1 is live | wellfound.com |
| Turing | Vetted remote engineering, US clients | Week 12+ (needs interview prep) | turing.com |
| Toptal | Top-vetted freelance, hard screen | Month 6+ | toptal.com |
| RemoteOK | Public job board, lots of indie | Week 10+ | remoteok.com |
| OfferZen | SA-based, mix of local + remote SA-friendly | Week 10+ | offerzen.com |
| Hacker News "Who is hiring" | Monthly thread, indie + senior | Week 12+ | hn.algolia.com |
| Direct DMs | Founders on X who post "we're hiring" | Anytime | Search "hiring junior" on X |
Application volume + the honest math
Send 5 applications/day for 60 days. That is 300 shots. Expect:
- 80% no response
- 15% rejection
- 4% phone screen
- 1% offer
That math is brutal but normal. The portfolio is what flips your conversion rate from 0.5% to 4%. Without a portfolio, the math is much worse.
Done when: you have 3 live projects, 12+ public posts on LinkedIn/X, and 100+ applications sent.
What "AI-fluent" looks like to a recruiter
The new junior screen has changed. Here is what they look at, in order:
- GitHub commit history. Daily-ish over 90 days. Real commit messages. Real PRs.
- Live deployed project. A URL they can click. Loads fast. Works.
- README quality. Can a stranger run this in 5 minutes?
- AI-collab evidence. PR descriptions that mention how you used Cursor or Claude Code, what you accepted, what you rejected.
- One LinkedIn / X post per week showing you ship.
- Open-source contribution. Even one PR to a small lib goes a long way.
If you have 1-5, you skip the cold-CV pile entirely.
What this guide does not promise
- Not "$124K in 90 days." The $124K is the Wellfound 2026 average across all remote SWEs, mostly mid-senior. Junior remote offers cluster $45-80K/yr in Year 1. Aim there, climb up.
- Not "no degree, no problem, no work." The work is real — 5 hours/day for 90 days is the floor. If you can't put that in, you can't break in via this path.
- Not "Zaio is the answer." Zaio is one path. The Odin Project, freeCodeCamp, and QCTO certs are also valid. Pick the one you will finish.
What's next
- Pick your stack (Next.js) and your curriculum (one of the four), today.
- Install Cursor + Claude Code + GitHub + Vercel + Supabase before the weekend.
- Start the daily drill on Monday.
- Comment BREAKIN on the original video if you want me to send check-in nudges every 2 weeks.
This guide powers the "Comment BREAKIN" CTA on ATB-220.
Sources
- Stats SA QLFS Q4 2025 — Media Release
- Stanford AI Index 2026 — junior dev hiring drop (summary)
- Wellfound 2026 — remote dev hiring data
- TechCentral / OfferZen — SA juniors squeezed by AI
- Career Karma — self-taught time-to-first-job benchmarks