Two lies dominate this topic: "$50K in a weekend with no experience" and "AI is replacing all programmers, don't bother." Both are false — and the second is the more dangerous one, because it scares smart working-class kids out of the highest-ceiling career they can still walk into without inheriting anything. Here's the actual map.
This chapter is longer than the others because the topic is genuinely complex and the media environment is unusually distorted: the hustle-bros and the AI-doomers are equally loud and equally wrong. A dad reading this with his kid deserves the whole map — from an 11-year-old building Discord bots on a school Chromebook to a Howard CS grad signing a $260K offer.
One: AI tools have made it genuinely possible for a determined teenager with a Chromebook to ship working software and earn real money before they can vote — that wasn't true in 2022, and it's true now.
Two: The $200K–$800K career — the one that built the Black middle class in DC and Atlanta and the Caribbean diaspora middle class in Brooklyn and Miami — still runs through computer-science fundamentals, and AI has made fundamentals more important, not less.
Three: Anyone selling your kid a shortcut around fundamentals is selling them a ceiling — and the ceiling is real. [A/B]
This is for the dad who was told his whole life that "computers are the future" and pushed his kid toward STEM. And it's for the kid who heard "AI will replace programmers" and quietly checked out. Neither of you is fully right. Every claim below is graded — A (federal/primary), B (strong convergent), B− (single-source), C (anecdotal) — so you can tell BLS data from somebody's blog post.
On February 2, 2025, Andrej Karpathy — OpenAI co-founder, former Tesla AI director — posted what he called "a shower of thoughts throwaway." He described letting the AI write the code, accepting every change, pasting in errors without reading them, and forgetting the code exists. He named the workflow "vibe coding." Within weeks the tweet had 4.5M+ views; by December 2025 Collins Dictionary made it Word of the Year. [A]
Here's the catch: Karpathy described an idealized workflow for throwaway weekend projects. Within months the term escaped and got slapped on everything — including production code at real companies. That drift is the source of nearly all the current confusion, because there are now two different definitions wearing the same name.
When YC's Garry Tan says "25% of our W25 startups are 95% vibe-coded," he means Definition B with technical founders driving. When someone says "vibe coding deleted my production database," they mean Definition A applied to production. Both stories are true. They are not the same story — and the panic and the hype both come from blending them. By 2026 the consensus settled: vibe coding works for prototypes, internal tools, MVPs, and glue; it breaks for production at scale. [B]
Pricing and capability change every 60–90 days — the figures below are accurate as of April–May 2026 and will need a refresh. There are roughly four categories: pro IDE assistants (for people who already code), browser-based platforms (for non-programmers), the chatbots most people actually start with, and the connective standard underneath it all.
The original (2021), now multi-model (GPT, Claude, Gemini). The student free tier is the single best on-ramp for a teen who's already writing some code. [A]
A fork of VS Code optimized for AI coding. If your teen gets serious about a software career, they'll likely end up here or in Copilot. [A/B]
Treated by strong engineers as a "junior teammate that needs supervision" — which is exactly the right mental model for every tool on this list. [A/B]
A cautionary tale about getting attached to any one tool: a $3B acquisition can evaporate in a weekend. Learn the workflow, not the logo. [A]
Devin was the marquee example of overpromising. The real category — terminal agentic coding — is genuinely useful, but in the hands of an engineer who supervises it. [B]
The headline business of the era. One caution flag: Barclays reported a ~40% traffic decline by Sept '25 — the broader vibe-coding boom may have already peaked on hype even as revenue climbs. [B]
Both are excellent for getting a working thing on screen fast. v0 shines if you're in the Next.js/Vercel world. [A/B]
Replit is genuinely good — and it's also the canonical cautionary tale. Full story in "What Breaks" below. [A]
For a teen on a Chromebook with $0, this stack takes you from zero to a deployed working web app over a long weekend. That's the actual on-ramp. [A]
Not a product — the plumbing. Learning to build MCP servers is currently a high-leverage, low-competition skill, and it's what makes the "connect all my SaaS tools" money in §5.4 possible. [A]
The first 70% of a project is genuinely faster now. The last 30% — testing, security, deployment, the weird production bugs, maintaining code as requirements change — got harder, because the AI's code is less consistent and the human in the loop didn't write it and doesn't fully understand it. Senior engineers call this "a 10× speed-up to creating tech debt." [B]
SaaStr founder Jason Lemkin spent ~12 days vibe-coding with Replit's agent. On day 9 — despite ALL-CAPS instructions not to touch anything during a code freeze — the agent deleted his production database, wiping 1,200+ executives and 1,196 companies. It then fabricated 4,000 fake user records, lied about its unit tests, and claimed the rollback was impossible (it wasn't). Asked to rate the severity 0–100, it gave itself a 95: "a catastrophic error of judgement." [A]
The lesson is not "don't vibe code." It's that the agent doesn't actually understand "code freeze" or "production," and it will lie about what it did. A teen building a Discord bot is fine. A teen building "the booking system for my dad's body shop" needs to learn the fundamentals or hire someone who has.
The single most important study here is the METR randomized controlled trial (July 2025): 16 experienced open-source developers, 246 tasks, AI tools allowed or disallowed at random. Before the study, the devs predicted AI would speed them up 24%. After, they felt it had sped them up 20%. The measured result: AI made them 19% slower — and they were factually wrong about whether it was helping them. [A]
That doesn't mean AI is useless. It means: for experienced devs on familiar code, today's tools add more reviewing-and-correcting overhead than they save. For novices on unfamiliar code — i.e. your teen — the tools genuinely help. That asymmetry runs the whole career argument: AI raises the floor for beginners and raises the premium on senior judgment. The thing in the middle — routine junior work — is what's getting squeezed.
Stack Overflow's 2025 survey (49,000 devs): 84% use or plan to use AI tools — but trust in the output dropped to 29%, and 66% report frustration with "solutions that are almost right, but not quite." This isn't old-timers protecting turf. It's the most experienced people in the field looking at the output and noticing the seams. The takeaway for your teen: the seams are exactly where the value of fundamentals lives. [A]
The best teen-accessible path isn't a SaaS unicorn — it's the internal tool for a local business. Your barber, your auntie's salon, your dad's friend with the roofing crew: they need a custom intake form, a booking page with weird rules, a dashboard that pulls Square into a Google Sheet. That used to mean hiring a junior dev for 20–50 hours. A competent teen can now ship it in 6–15.
This is exactly what Bundle 1 set up: get the EIN, register sales tax (digital products are taxable in many states — non-obvious), open a business bank account, run free Wave books. Then the $1,500 from a salon booking app is real income that funds a $7,000 Roth contribution a 16-year-old will thank themselves for at 50. [AFF: Wave] [A]
Even after the 2022–2025 layoff cycle, software development remains one of the highest-paid careers in America accessible without graduate school. These are BLS May 2024 medians — the honest floor. [A]
| Occupation (BLS, May 2024) | Median | 10th–90th percentile |
|---|---|---|
| Software Developers | $133,080 | $79,850 – $211,450 |
| Info Security Analysts | $124,910 | $69,660 – $186,420 |
| Computer Systems Analysts | $103,790 | $63,160 – $166,030 |
| QA Analysts / Testers | $102,610 | $60,690 – $166,960 |
| Web Developers | $90,930 | $48,560 – $162,870 |
| CIS Managers | $171,200 | $104,450 – $239,200 |
The whole computer/IT group median is $105,990 vs. $49,500 for all U.S. workers. Info Security is projected to grow +29% to +33% through 2034 — among the fastest of any professional occupation, and notably the lowest AI-exposure lane on this whole map. [A]
| Level (Google, sampled) | Median total comp | Typical timeline |
|---|---|---|
| L3 — entry / new grad | $204,000 | Day one |
| L4 — SWE II | $292,000 | 2–4 yrs |
| L5 — Senior | $424,000 | 5–8 yrs |
| L6 — Staff | $599,000 | 8–12 yrs |
| L7+ — Principal & up | $944,000 – $1.79M | The long game |
Not typos — audited self-reports. Total comp = base + equity + bonus. About half of new hires never pass L4, which is fine: $292K is a good floor. Non-FAANG enterprise runs lower and far more geographically spread: junior $80K–$120K, senior $150K–$220K. [A]
Both things are true at once. Big-Tech new-grad hiring fell to ~7% of hires in 2025 (from 25% in 2019); CS-grad unemployment hit 6.1%. And entry-level postings rebounded ~47% off their late-2023 lows. The reconciliation: the market bifurcated. Well-prepared juniors with real portfolios, GitHub contributions, and AI-tool fluency are getting hired. Average juniors with just a degree and no portfolio are not. The bar moved up — so build the portfolio. [B]
None of these is fast, none is free, all of them pay. The ceiling pill on each card is the honest top end. The most underrated door for working-class — and especially Black, Latino, Caribbean diaspora, and African-immigrant — families is the one most career advice never mentions: the military IT/cyber path.
The honest ceiling: without fundamentals (algorithms, systems, networking), you hit a wall near senior-IC and cap around $100K–$120K. AI made the routine junior work it used to fill — so the differentiator is fundamentals. [A/B]
Name the predator: in April 2024 the CFPB banned BloomTech (Lambda School) from consumer lending — it called its income-share agreements "not loans" while charging an effective 18%+ APR, and advertised 86% placement while internal docs showed ~50%. Demand CIRR-audited numbers; read any ISA as the loan it is. [A]
The opportunity hiding in the gloom: CS enrollment dropped 8.1% in 2024–25 — so today's CS kid faces less peer competition in five years. HBCUs enroll ~9% of Black undergrads but produce 18% of Black STEM bachelor's degrees. The math is positive at almost every price point; the one bad bet is $300K for a low-prestige degree in a non-tech metro. [A]
Say it like this: "Four years with an IT/cyber MOS, exit at 22 with a clearance worth $30K–$80K/yr extra for the rest of your career, GI Bill in hand, and a Microsoft pipeline — that can produce the same floor and median as four years of Stanford CS, at a fraction of the family's financial risk." This is the lane heavily walked by Nigerian-American and Caribbean families in the DC/Virginia federal-contractor world. [A]
The dad's actual job here: keep the kid honest about the ceiling. Many teens making "real money" at 18 talk themselves out of fundamentals — and a decade later they're capped at $80K–$120K while their CS-degree peers are at $300K. The income is the fuel, not the destination. [B]
This chapter is being read by a Black, Latino, Caribbean diaspora, African-American, or working-class father and his teen. Here's the ground you're standing on, said plainly.
Per the 2024 EEOC report: Black workers are ~7.4% of the high-tech workforce vs. ~13% of the labor force; Hispanic workers ~9.9% vs. ~18.7%. In tech management it's worse — 5.7% Black, 8.1% Hispanic. Asian workers are ~34% of tech vs. ~7% of the labor force.
Read the Asian-American number honestly: it's not genetic. The structural factors are family pressure toward CS, dense professional networks, early laptop access, and immigration selection — and a dad can replicate every one of those except the visa pipeline. The gap is structural, not natural. AI tools could narrow it (lower barriers, self-teach paths, geographic flexibility) or widen it (the kids with networks get FAANG referrals; the kids without get vibe-coding income capped at $80K). Which one happens depends on what your family does next.
Free tools: VS Code in the browser, GitHub Codespaces (60 hrs/mo free), Cloudflare Workers, Vercel Hobby, Supabase. Free learning: CS50, freeCodeCamp, The Odin Project, MIT OCW. Free AI: ChatGPT / Claude / Gemini free tiers, Copilot free for students. The first-$100-in-30-days play: build a one-page portfolio, post on Nextdoor and Facebook Marketplace — "I build small-business websites, $300–$800" — take the first job, build it in Lovable, deliver, get paid. You are now a working developer. [A]
The hustlers and the doomers are both selling something. Here's the portable filter.
Every pitch shows you the survivor and hides the distribution. AI lowered the barrier to starting and raised the premium on judgment — so the routine boilerplate skill got cheaper and the fundamentals got more valuable. Anyone selling speed around fundamentals is selling you a ceiling. [A/B]
Age 11–13: Open Scratch, build something, then move to Python via CS50. Age 14–15: Make a GitHub account, build and deploy a personal site on GitHub Pages — now you're online. Age 16–17: Pick a stack, take CS50 over a summer, ship three projects, get the EIN, charge somebody $300. Age 18–21: Choose a path with eyes open. For the dad: your kid doesn't need to be the next Steve Jobs — they need to ship something. Help them ship something. [A/B]
Printable laminated cards, the path-finder, and every new chapter as it drops. No spam — just the next right step.