THE DEV BENCH
🪜

Get Hired

Entry paths into technology, ranked by how likely you are to actually get hired — deliberately not by what they pay. Written for someone with no degree, no work experience, and no idea where to start.

A path that pays $130,000 but never hires a beginner is worth less here than one paying $48,000 that does. Every figure below is sourced, and everything we could not verify is marked as such.

This is the map, not the course

The research behind this ladder is done and sourced. The courses mostly are not built yet — 2 partly built, 5 still to come of 7. Each rung says which it is, so nobody clicks expecting a course that is not there. The upper rungs — Kubernetes, Terraform, Linux, Azure, Security+ — already exist over on Paths.

The number to beat

$30,110a year · $14.47/hr

Fast-food work

BLS figure. The honest comparison — not the $7.25 federal minimum, which almost nobody actually earns.

Every path below clears it. The real questions are by how much, how fast, and how reliably you get in.

The market you are walking into

Everything below is true and the paths still work. This is the part most career advice leaves out, and you should have it before you choose.

An entry-level posting is not an entry-level competition

In May 2026, 49% of applications from workers with ten or more years of experience went to entry-level roles, and experienced workers made up 30% of all entry-level applicants. You are not only competing against other beginners.

The market is tilting toward seniority

Entry-level postings were down 7.5% year over year while senior postings rose 14.7%. In software development specifically, entry-level made up 4.5% of postings and senior 69.3%.

The AI hiring boom is a senior boom

Software development postings rose about 15% between February 2025 and mid-2026, but 71% of that rise was senior roles. AI has not opened a door at the bottom; it has raised the floor.

The squeeze on new entrants is measurable

Entry-level hiring at twelve large tech employers is down about 65% against 2019, and about 76% at early-stage startups. Payroll data puts employment for 22-to-25-year-olds in AI-exposed occupations roughly 19% below where it would otherwise be, driven by reduced hiring rather than layoffs. The researchers call these early descriptive findings, not proof of cause.

"Fastest-growing job" can mean almost no jobs

Growth is a percentage, and a percentage of something small is still small. The two fastest-growing occupations in the whole US economy add fewer than 20,000 jobs between them — the government's own statisticians point this out. Always ask how many openings there are a year, not how fast the number is rising. That applies to every list you will read, including the ones quoted on this page.

This is a low-hire, low-fire market

The hiring rate sat at 3.4% in June 2026 and total postings were down 2.9% year over year. Fewer people are being let go, and fewer are being taken on. Expect the search to be slow and do not read that as a verdict on you.

The ladder

Ranked by probability of landing a first job. Start at the top unless you have a reason not to.

Every rung is scored on two different questions, because they have different answers. Getting hired is whether an employer will take you on with no experience. Earning it yourself is whether you can sell the work directly, without anyone choosing you first. Help desk scores top on the first and zero on the second. Video editing is the reverse. Neither number is the real one — which matters more depends on whether you need a job or you need income.

1
🎧

IT Help Desk / Tier 1 Support

The highest volume of genuinely entry-level openings, the lowest bar of anything here, properly remote, and it is the trunk every infrastructure path branches from.

Partly built
Getting hiredHighEarning it yourselfEffectively none
Why those two differ
The sharpest split on the ladder. Nobody buys freelance tier-1 support from an individual, so there is no way to earn from this without being hired. That is fine — it is the trunk, and what it branches into does sell.
Pay
$48,154 average. Most roles $40,000–54,000, top decile $63,000. Remote-specific $25.03/hr
Against the benchmark
About $18,000 a year more than fast food
Certificate
CompTIA A+ — $530. Public workforce funding may cover it, but the amount is set by your state and local workforce board, not nationally — ask your local American Job Center whether the course is on your state's approved provider list
Remote
Remote-native
How many jobs
Very high — the most abundant true entry-level technical role
How hard to get in
Lowest bar on this page. Routinely hires people with no prior experience
Leads to
Cloud Support → Cloud Engineer → DevOps/SRE, or SysAdmin → Platform

Entry job titles

Help Desk Tier 1Remote Support TechnicianService Desk AnalystTechnical Support Specialist
Closest path we have built
2
🎬

Video Editing

Remote-native, works as a job or a business, and uniquely — if you already edit your own gaming or streaming clips, you are not starting from zero.

Not built yet
Getting hiredModerateEarning it yourselfHigh
Why those two differ
The reverse of help desk. Job postings are thinner than the hiring ranks above, but the work sells directly from the first reel and no employer, degree or certificate stands in the way.
Pay
Beginner $25–40/hr or $150–400 per video. Employed average $65,728, top 10% over $101,000
Against the benchmark
Clears it at beginner rates, from the first paid job
Certificate
None exists — Free — your reel is the credential
Remote
Remote-native
How many jobs
High, and heavily freelance-shaped rather than posting-shaped
How hard to get in
Low to enter, and the craft is learnable on free tools
Can also be sold directly
Creator retainers $1,500–3,500/month for four long-form videos plus social cutdowns
Leads to
Live stream technician → Broadcast → Live stream producer

Entry job titles

Video EditorContent EditorShort-form EditorCreator's Editor
3
☁️

Cloud Support

Pays more than any other true entry role and Amazon hires it at volume — but it expects some Linux and networking first, so it is a step up rather than a start.

Not built yet
Getting hiredModerateEarning it yourselfLow
Why those two differ
Cloud contracting pays very well but clients do not hand infrastructure to someone with no track record. The earning comes later, from the same skills — not at entry.
Pay
Cloud Support Associate averages $108,280 and starts at $70,819. Junior Cloud Engineer remote $71,799
Against the benchmark
About $40,000–78,000 a year more
Certificate
AWS Cloud Practitioner or Azure AZ-900 — $100 / about $99
Remote
Remote or hybrid
How many jobs
Healthy, though counts vary sharply by source — treat the big numbers as optimistic
How hard to get in
Harder than the three above. Expects foundational Linux and networking
Leads to
Cloud Engineer → Platform → DevOps/SRE

Entry job titles

Cloud Support AssociateJunior Cloud EngineerCloud Operations Associate
Closest path we have built
4
🔍

SEO and AI-Answer Visibility

The certificates are free and take an afternoon, the work is remote-native, and it sells directly to small businesses as well as it hires. 🔴 Note which half hires: SEO does, and AI-answer visibility does not yet.

Not built yet
Getting hiredLowEarning it yourselfHigh
Why those two differ
🔴 Moved down a place in s204 on new data. Marketing job postings have sat more than 20% below their pre-pandemic level for three years running, while the job market as a whole recovered — so being hired into this is harder than the free certificates suggest. Selling it is a different story: it is still the easiest thing on this page to sell to a local business, and recurring, which is what turns gigs into a living.
Pay
Under a year of experience $52,333. Remote entry average $67,388. Coordinator roles $54,061
Against the benchmark
About $22,000–37,000 a year more
Certificate
HubSpot SEO, Semrush Academy, Google Analytics, Google Ads — Free — 3–5 hours each
Remote
Remote-native
How many jobs
Moderate — a few hundred remote specialist openings at any time
How hard to get in
Moderate. No gatekeeper, but one demonstrable result beats any certificate. 🔴 Do not aim at a GEO job title — an independent count in July 2026 found roughly 50 dedicated AI-search roles in the entire country, and they are Director and VP positions filled by poaching ten-year SEO leads. GEO is a specialisation you add to SEO and a service you can sell, not a door you walk through
Can also be sold directly
$500–1,500/month per client. Single-location trades sit at $800–1,500
Leads to
SEO Lead, or your own agency on recurring revenue

Entry job titles

SEO SpecialistDigital Marketing CoordinatorSEO AnalystContent & SEO Associate
5
💬

Community Management

Remote-native with a low bar, no certificate exists to gate it, and running a Discord or a guild is genuinely relevant experience.

Not built yet
Getting hiredModerateEarning it yourselfLow
Why those two differ
Sells only once you have run something visible. Until then there is nothing for a client to buy, and no certificate exists to stand in for it.
Pay
Average $58,937. Junior roles $45,060–60,080
Against the benchmark
About $15,000–29,000 a year more
Certificate
None exists — Free — show the community
Remote
Remote-native
How many jobs
Moderate and growing with creator and game communities
How hard to get in
Low. The credential is a community you have actually run

Entry job titles

Community ManagerDiscord Community ManagerCommunity ModeratorSocial Coordinator
6
🛰️

Drone and Satellite Data

The flying is on-site but the processing is not, the desk half pays better than the flying, and the same skills carry you all the way into the space sector.

Not built yet
Getting hiredLowEarning it yourselfModerate
Why those two differ
Employers want tool fluency and prior surveys. But mapping is bought as a service by builders, farmers and surveyors, and a Part 107 plus a processed map is enough to quote for that work.
Pay
Drone mapping $22.48/hr. GIS analyst $55,000–95,000. Photogrammetry $48.20/hr. Satellite imagery analyst $77,626
Against the benchmark
About $10,000–65,000 a year more depending on the rung
Certificate
FAA Part 107 — $175 — no degree, no prerequisites
Remote
Remote
How many jobs
Moderate, and broadening as drone survey work becomes routine
How hard to get in
Moderate. Tool fluency in QGIS, Pix4D and ArcGIS is the real requirement
Leads to
GIS Analyst → Satellite Imagery Analyst → Remote Sensing Scientist

Entry job titles

Drone Mapping TechnicianGIS AnalystPhotogrammetry SpecialistSatellite Imagery Analyst
Closest path we have built
7
📐

CAD and PCB Design

The desk side of 3D printing and electronics — remote, freelance-friendly, and the design tools are free.

Partly built
Getting hiredModerateEarning it yourselfModerate
Why those two differ
Drafting hires steadily and pays modestly. Board layout sells for two to four times the drafting rate, but only once a portfolio exists — which is why the portfolio is the path, not an afterthought.
Pay
CAD entry $20.51–26.53/hr. PCB freelance $50–90/hr in North America
Against the benchmark
About $12,000–25,000 a year more at entry
Certificate
CSWA (SolidWorks) or Autodesk Certified User — $99 / $94–148 — and KiCad itself is free
Remote
Remote
How many jobs
BLS projects 10% growth for drafters to 2034, roughly 16,200 openings a year
How hard to get in
Moderate. A portfolio of real drawings is required, not optional
Can also be sold directly
Freelance drafting and board layout, sold directly

Entry job titles

CAD DrafterDesign DrafterPCB DesignerLayout Designer
Closest path we have built

Doors that are already open

Programmes that verifiably take people with no degree, no experience and no money. These matter more than any job title, because someone else is holding them open. Each one states its catch, including the good ones.

Per Scholas

Tuition-free technical training. Tracks include AI Native Software Development, AI Native Business Intelligence, Cybersecurity with AI Tools, IT Support (Google IT Support plus CompTIA A+) and an 8-week Data Center Technician course.

Who can apply
18 or over, high school diploma or equivalent, US work authorisation. No degree.
Remote?
🎯 Has a National Remote option — the only verified remote route here.
The catch
Geography rules vary by course, and some remote tracks still require living within reach of a campus. No placement rate or salary outcome is published, so do not assume one.

AWS re/Start

Free cohort-based cloud training aimed at people who are unemployed or underemployed. Prepares for AWS certification and includes interview coaching.

Who can apply
18 or over. No degree, no technology background required.
Remote?
Not stated on the programme page.
The catch
Programme length, delivery format and placement rates are all unpublished. Verify before committing time.

Registered apprenticeships — the 14 national AI sponsors

Apprenticeship.gov lists national sponsors you can join WITHOUT an employer sponsoring you, including Apprenti, ACI Learning, OpenClassrooms, CareerWise and Step IT Up America. Paid from day one, rising wages, and a nationally portable credential. Over 58,000 apprentices were served across technology, cybersecurity and AI occupations in 2025.

Who can apply
No education prerequisites stated. Apprenti says no prior experience is required for most of its programmes.
Remote?
Varies by sponsor and is mostly unstated.
The catch
⚠️ The Department of Labor's AI apprenticeship portal itself is built for EMPLOYERS designing programmes — it lists no live openings. The named sponsors are the actionable part; the portal is not.

Claude Corps (Anthropic)

A funded fellowship placing 1,000 people at over 400 nonprofits. $85,000 plus benefits, twelve months, and it is real W-2 employment rather than contract work.

Who can apply
🎯 18 or over, UNDER two years of full-time work experience, US work authorisation. No degree requirement.
Remote?
🔴 Full-time in person, with relocation expected.
The catch
The first cohort closed in July 2026. Remaining cohorts start January 2027 and August 2027 — so this is one to watch and prepare for, not one to apply to today.

Data centre technician

The operational floor of the AI buildout. Microsoft states the minimum as a high school diploma, GED or equivalent experience. Google runs an apprentice track where holding a degree in the field is actually DISQUALIFYING, which is rare and genuinely favours this audience.

Who can apply
High school diploma or GED. No degree.
Remote?
🔴 On-site, always. Twelve-hour shifts including nights, weekends and holidays, with travel up to 25%. There is no remote version of racking a server.
The catch
⚠️ The permanent headcount is far smaller than the coverage implies. A quasi-experimental study across roughly 1,500 US facilities found a typical county gains only 100–200 permanent jobs, with local wages unchanged. Virginia produced one direct permanent job per $54 million invested. The large labour demand in the buildout is construction, which is temporary per site.
What about certificates
🔴 Do not buy a data-centre certificate. Across 220 hyperscaler technician postings, ZERO required any third-party certification. A high school diploma appeared in 108 of 200 Amazon listings; Google's said 'bachelor's degree or equivalent practical experience' in 18 of 20 and named no certificate in any minimum requirement. If you spend anything, BICSI Installer 1 is the one no-prerequisite credential with real employer text behind it — it is hands-on tested and Amazon names it in cabling roles. The data-centre credentials sold hardest are aimed at managers and consultants, one of them at nearly $5,000, with zero hits across Amazon's entire global board.
The tell
41 of 200 Amazon postings list a Cisco certification in their preferred qualifications that was retired in February 2020. They are recommending a credential that has not existed for six years. That block is copied-forward boilerplate, not a hiring criterion — which is roughly what 'preferred qualifications' is worth generally.

Is AI itself a career you can start in?

Learning AI will not get you hired on its own. It makes you the stronger candidate for the job you were going to apply for anyway, and it is the fastest route to your own first invoice.

Why it is not a rung

  • Standalone prompt-engineer postings are down about 30% since 2024 and the title has effectively disappeared at frontier labs. Roles asking for the skill grew about 3×.
  • The AI job titles that do hire — AI engineer, ML engineer, AI product manager — are senior roles that expect an existing career to step from.
  • Entry-level technical hiring overall is down sharply, and Anthropic's exposure research puts the measurable harm precisely on new entrants aged 22–25. AI is raising the floor to get in, not removing the ladder.

The certificate trap

GitHub Certified: Agentic AI Developer (GH-600) No formal prerequisites, so it reads like an entry certificate anyone can take. Its stated audience is architects, platform engineers, DevOps engineers and security engineers operating agents in production, and it expects prior software-lifecycle, GitHub-workflow and code-review experience.

Certificate difficulty and job accessibility are different axes and they do not correlate. Treat this one as a destination marker, never as the entry bet.

The one entry-level AI job there is

AI output review and quality — auditing what the model produced

The evidence
A live remote posting at $45,000–50,000, listed as entry level, whose only stated requirement was the right to work. No degree, no years of experience. The work is auditing AI-generated support answers for accuracy, hallucinations and compliance.
The work
Read what the model said. Decide whether it is true, safe and on-policy. Say why.
The catch
🔴 Rare. That was ONE posting across everything searched, so treat it as a real category and not as a job market. Learn the skill because every rung needs it; do not plan on the title being there when you look.

Where AI does open a door: AI automation for small businesses

Getting hiredLowEarning it yourselfHigh
Time to competence
About two weeks on a workflow platform such as n8n
Tool cost
Zero — n8n is open source and self-hostable, with 400+ integrations
Entry rate
$25–40/hr with no completed projects behind you
Per project
$200–1,500 for a one-off workflow build. A simple form → CRM → email chain goes for $500–2,000
Retainer
$1,000–3,000/month per client
Gatekeeper
None. Nobody has to choose you first
The catch
There is no job advert for this. You find the client, scope the work and invoice it yourself, and the day-one hourly rate is not far above the fast-food benchmark until you have proof of work.

AI training and data annotation — read this before you sign up

Headline rates of $80–125/hr are real and they are for physicians, attorneys, quants and senior engineers. The tier actually open to a beginner pays $2–15/hr. The genuinely open platforms enforce floors of about $8/hr, and the one credential-free lane on the biggest job board pays $10–15/hr.

Their own published numbers, divided

  • One platform advertises $15 million paid to over 240,000 contributors. That is about $62 each, for the life of the platform.
  • Another: $20 million to over 100,000 contractors. About $200 each.
  • The largest: over $500 million to more than 900,000 people. About $555 each.
  • A fourth: $100 million to over 100,000 fellows. About $1,000 each.
  • 🔴 And one platform reports nearly five million registered experts against roughly 30,000 working in a given week. Well under 1% of the people signed up are working at any moment.

One of those platforms builds its home page out of screenshotted posts from workers celebrating $40,000, $75,000, even $200,000. If the company has paid $20 million in total, it cannot contain even a hundred such people without paying everyone else nothing.

How open is it really
🔴 Less open than advertised. The big platforms require a bachelor's degree or 'undergraduate-level expertise', one accepts only the top 1% of applicants, and another has no public sign-up at all. 'No AI experience required' is true and is a different claim from 'no degree required'. What is genuinely open: research-participation platforms with an $8/hr floor, the lowest-paying task sites, and household-video recording at $10–15/hr.

What can go wrong, documented

  • Read the terms. One major platform reserves the right to deactivate you 'at any time, immediately without prior notice for any reason', states that a violation 'may result in non-payment to you of some or all amounts accrued', and requires binding arbitration with class actions waived. They can drop you without a reason, keep money you already earned, and you have signed away the group remedy.
  • It happens at scale. One platform shut down in Kenya, Nigeria and Pakistan outright; another cut 200 staff and 500 contractors and closed its generalist team; a third laid off around 500 annotators in a day.
  • Across 776 cloudworkers in 100 countries, 31% had experienced non-payment and 40% did not know of any appeal process.
  • A misclassification case against one platform settled for $12.5 million covering two of its brands. Others are live, one naming a chief executive personally.
  • To get this work you hand over government ID, banking details, biometrics and screen-recording rights. One platform lost roughly four terabytes of exactly that — facial biometrics, tax and banking data, monitoring screenshots and Social Security numbers.
  • The generalist tier is being automated directly. Training methods that grade against programmatic checks rather than human preference eat precisely the maths, code and structured work that made up most of it.

⚠️ Task-scam reports to the FTC went from effectively zero in 2020 to about 20,000 in the first half of 2024 alone — nearly 39% of all job-scam reports. The mechanic is always the same: unsolicited message, simple repeated tasks, small real payouts to hook you, then you must deposit your own money to 'unlock' earnings. 🎯 One rule defeats all of it: never pay anyone in order to get paid. No legitimate platform in this sector ever asks you for money. Type the domain yourself and never follow a link from a message.

Do not trust review scores here. Two platforms show 4.6 and 4.7 stars from thousands of reviews — written by applicants praising the automated interview, who have never been paid anything. Cross-check against sites where reviewers actually worked, where the same platforms score far lower on pay and job security.

🔴 Treat any money from this as a windfall, not a budget line. The most commonly reported outcome is an empty dashboard after unpaid onboarding. Screenshot your tasks and earnings, never let a balance sit past one payout cycle, and never make it the plan.

The New York Fed examined job postings in May 2026 and found little sign of a distinct AI-driven decline in labour demand — the gap between exposed and unexposed occupations opened BEFORE 2022, with no clean break after ChatGPT, and no junior-versus-senior divergence. Set against PwC's seniorization finding, the honest position is that the direction is contested. The 2022 hiring bubble deflating explains a great deal of what gets attributed to AI.

Public money for training

Can it fund this
Yes. In August 2025 the Department of Labor issued guidance explicitly encouraging workforce boards to spend existing funds on AI skills training, and the regulations already make any program leading to an industry-recognised certificate eligible. ⚠️ Note the word existing — that guidance came with no new money attached.
How much
There is no national amount, and anyone quoting one is wrong. Caps are set by each state and local workforce board. Across the boards checked, they ran from $5,000 to $7,500 per person. Yours is whatever your local board says it is.
The rule most guides leave out
🔴 Training funds are for people who cannot get the money elsewhere. If a Pell Grant would cover your program, this funding is meant to be secondary to it, not instead of it. That single rule catches a lot of people out on a college-delivered course.
What is actually approved
CompTIA programs are the best-covered by far — over 1,100 approved programs across 28 states. AWS appears in at least nine states. Google's certificates appear in at least eight, but always under the college or nonprofit delivering them, never under Google. Explicitly AI-titled programs exist and are growing but remain thin — roughly a dozen nationally against those 1,100 CompTIA entries.
Also new
Separately, Workforce Pell began on 1 July 2026 and covers short programs of roughly 8 to 15 weeks at accredited institutions. It is a genuinely new source rather than a reshuffle of an old one.
How to start
Find your local American Job Center through careeronestop.org or on 1-877-US2-JOBS. Ask two questions: which funding stream you qualify for, and whether the specific course you want is on the state's approved provider list.

Who qualifies

  • Adult funds: the entire eligibility test is being 18 or older. But priority is required by law for people on public assistance, on low incomes, or who are basic-skills deficient — and the Department pushes states to fill at least 75% of these places from those groups. Outside them you are competing for what is left.
  • 🎯 Dislocated Worker funds carry NO priority test. If you were laid off, lost a job to a plant closure, or were self-employed and lost the work to economic conditions, this is the cleaner route — and it is also the funding stream least at risk right now.
  • Veterans and eligible spouses get priority of service across all of it.

The rules that decide it

  • The money only pays for programs on your state's approved provider list. 'Is this course on my state's list?' is the decisive question — not whether the course is any good.
  • Registered apprenticeships are added to that list automatically and are exempt from the usual approval process.
  • 🎯 You may choose a provider outside your local area, or in another state, as long as it is on your state's list.
  • ⚠️ A common myth: you do NOT have to complete career services first. There is no federally required waiting period. Enough assessment to determine eligibility is required; a sequence of appointments is local policy, not law.

🔴 Act on this sooner rather than later. Current funding runs to 11 December 2026, and in June 2026 a House committee advanced a bill cutting these formula grants by 62% — eliminating the Youth program entirely and almost all of the Adult program, while leaving Dislocated Worker largely intact. It is not law and may not become law. But if you intend to use this money, the window is now and the Dislocated Worker stream is the durable one.

What a certificate is actually worth

Every path on this page names a certificate, so you are owed an honest account of what one buys — especially if you do not have the money spare.

Employers are buying AI skills, not AI certificates. The strongest evidence is not a posting count at all: the labour-market firm that supplies this data CAN count certifications and does so for other fields — and when it lists the credentials that actually appear in generative-AI job postings, it names security clearances, MBAs, project-management and cloud certificates. It names no AI certification at all. Not one.

CompTIA Security+ is named in 70,019 open cybersecurity roles — about 13.6% of the 514,359 openings tracked nationally. The best-performing AI certificate measured appears in dozens. That is a gap of three to four orders of magnitude, not a close race.

Security+ dominates because it is a legal requirement for US defence work, not because employers judged it the better credential. No regulator mandates any AI certification anywhere, which is why none of them has an equivalent floor under it.

What a certificate signals to the person reading it

Surveyed hiring managers say they value microcredentials as a sign of ambition, discipline and a learning mindset rather than as a measure of specific skill, and report a lack of trust in their validity. A certificate reads as a personality signal, not a skill signal — which is precisely the wrong thing if what you needed to prove was competence.

'Google Career Certificate' returned 110 postings — and almost all of them were at one restaurant chain, where it is offered to staff as an employee benefit rather than asked for as a qualification. That is what 'recognised by 150+ employers' amounts to in practice: those employers commit to CONSIDERING graduates, a verb that costs them nothing.

What is genuinely recognised

  • ANSI/ISO 17024 accreditation and US Department of Defense 8140 approval are the only structural recognition found anywhere. They are hiring rules, not claims. CompTIA A+, Network+, Security+, Cloud+, Linux+, CySA+, PenTest+ and SecurityX carry both.
  • Microsoft's certifications are proctored, renew free each year, and appeared in 3,221 postings.
  • ISC2's Certified in Cybersecurity requires no work experience, is ISO 17024 accredited and DoD 8140 approved. ⚠️ Its free exam programme closed to new entrants in May 2026, so it now costs about $199 plus annual membership.

What is being sold as one and is not

  • CompTIA's own AI Essentials and AI Fundamentals are unproctored, and CompTIA itself calls them certificates rather than certifications. They carry no accreditation.
  • The same pattern repeats across four separate vendors: a respected certification brand whose AI variant has almost no hiring signal. One vendor's traditional data engineering certificate appeared in 99 postings; its generative-AI version appeared in 2.
  • A widely-taken introductory AI certificate takes about four hours and has been earned by roughly two million people. A credential two million people obtained in an afternoon cannot distinguish between candidates.

The size of the AI skills market, honestly

  • Even in IT and computer science, the most saturated field there is, about 14% of postings mention any AI skill. Which means 86% mention none. Marketing is 8%, engineering 4%, finance 1.4%, healthcare 0.5%.
  • Roughly 90% of all AI-related job postings in 2025 came from just 1% of companies, and only about 5.7% of US firms are AI adopters at all. A general-purpose AI certificate has a very narrow set of employers it could possibly impress.
  • About one in seven 'AI mentions' in job postings is the employer advertising its own AI recruiting software rather than asking for a skill. In some sectors it is nearly all of them. Any raw count of AI postings is an upper bound, including the ones we quoted earlier.

Traps that cost real money

  • The Google IT Support certificate does NOT grant CompTIA A+. Google's own wording is that graduates who ALSO pass the A+ exams earn a dual credential — you still pay for and sit both proctored exams.
  • Where those course certificates genuinely earn their roughly $300 is ACE college-credit recommendations — up to 15 credits. That is real, independently administered recognition, but it is ACADEMIC, not employer recognition, and each institution decides separately.

The intersection of free, no prerequisites, and genuinely recognised is now empty. The last credential occupying it closed its free programme in May 2026. Recognition requires independent proctored assessment, and that costs money to administer.

So spend your money in this order

  1. 1Learn on the free tiers. AWS Skill Builder alone carries 900+ courses at no cost, and Microsoft Learn, NVIDIA's institute, Hugging Face and fast.ai are all genuinely free. This is where the capability comes from, and none of it costs anything.
  2. 2Build public artifacts — repositories, published work, a profile someone can read. In AI and ML hiring this outranks every completion certificate measured, costs nothing, and is what those 19,256 postings are actually screening for.
  3. 3Spend money only on proctored exams that actually appear in the postings you are answering. Security+ or A+ for IT and defence work, a Microsoft certification, or the AWS AI Practitioner at $100 if you want an AI certificate — it has the highest posting count and the lowest price of the group, which is not a coincidence.

Advice that has already gone stale

  • The Azure AI-900 exam retired in June 2026 and was replaced by AI-901. Any guide still saying 'take AI-900' has not been updated.
  • CompTIA's own page estimates the current Security+ version retires during 2026 without naming a successor. Check which version is current before booking.

The desk beside the workbench

Every hands-on trade has a design, data or software half — and that half is remote-capable, better paid at entry, and reachable without a workshop.

Hands-on, on-siteThe desk half, remoteWhere we teach it
3D printing and fabricationCAD and product designAdditive ManufacturingFreeCADCode-CAD with build123d
Electronics and solderingPCB design and schematic captureElectronics & PCBKiCad Layout
Flying dronesPhotogrammetry, GIS and mappingDrone TechnologyFAA Part 107Satellite OperationsOrbital Compute
Building robotsSimulation and robotics softwareROS 2 Robotics
The datacenter floorCloud and infrastructure operationsrung 3, not built yet

The physical skill is not the way in. The desk beside it is.

What not to pursue

A guide that only tells you what to do is lying by omission. These are the paths the numbers say to avoid right now. Every entry has evidence behind it — “everyone says it’s dying” is not evidence.

Game development

8,000–12,000 jobs cut globally in the first half of 2026 alone. A third of US developers laid off within two years. Entry openings collapsed, and applicants now compete against laid-off veterans willing to take junior roles.

QA was the historic side door and AI has largely closed it. Learn Godot to learn programming — it is genuinely good for that. Do not aim at a studio job.

Administrative and clerical work

Cashiers are projected to lose 313,600 jobs by 2034, the largest decline in the US economy. Office assistants lose 177,800. Clerical typists are down 36%, phone operators 27%. Admin roles are six of the twenty largest declines.

If the day-to-day work is receiving information in one format and retyping it in another, it is being automated. Do not train toward it.

Robotics engineering (as a first job)

Entry-level robotics engineers average $105,605, which looks inviting — but most engineering roles require at least a bachelor's degree, and no certificate substitutes. Remote openings are thin.

A destination, not a door. Reach it after the software ladder with a portfolio behind you. Do not let the pay figure sell a door that is shut.

Space engineering (as a first job)

Ground-segment software pays $85,000–115,000 at entry but requires a degree. Defence space work requires US citizenship and a clearance, and cleared work is on-site by definition.

The front door is shut. The back door is Earth-observation data — the same skills as drone mapping, and it is rank 6 on this ladder.

But do not read that as “tech is finished”

Research measuring which automatable tasks are actually being automated at work complicates the story in both directions:

  • Computer programmers are the single most-exposed occupation, at 75%. "Learn to code" is not a safe harbour from AI. Anyone selling it as one is wrong.
  • No systematic rise in unemployment for highly exposed workers since late 2022. Exposure is not the same as displacement. Your field is probably not dying.
  • But hiring is measurably down for workers aged 22–25 in exposed occupations. The effect lands on new entrants, not on people already inside. That is precisely who this ladder is for.

The honest summary: AI is not deleting these careers. It is raising the floor to get into them — which is exactly why the next section is not optional.

Working with AI, as engineering

This is not a separate path. Every rung on the ladder teaches these seven rules against its own work.

The wage premium for AI skills is real, and the two best measurements of it disagree. PwC, across more than a billion job adverts in 27 countries, tracks it at 25% (2024) → 57% (2025) → 62% (2026). Lightcast, across 1.3 billion postings, puts it at 28% — about $18,000 a year — for a posting listing AI skills against an otherwise identical one without them. Both are credible and they measure differently, so treat the range as the honest answer rather than picking the bigger number.

Surface familiarity with AI tools is now table stakes, not a differentiator. The premium goes to people who have fixed things when they broke — and can explain how.

  • Veracode: 45% of AI-generated code contains an OWASP Top 10 vulnerability.
  • CodeRabbit, across 470 pull requests: AI co-authored code carried 1.7× more major issues and 2.74× more security vulnerabilities.

The line between professional and amateur is review, not tooling. Same models, same prompts — the difference is whether a human understood and verified the output.

  1. 1

    Use more than one, deliberately

    Claude Code, OpenAI Codex, Google Gemini. Same task, different tools, compare the results. Being vendor-agnostic is a habit before it is an architecture — the same idea as multi-cloud, applied to AI.

  2. 2

    Never ship what you have not read

    If you cannot explain a line, you cannot ship it. This single rule is the difference between the 62% premium and the 2.74× vulnerability rate.

  3. 3

    Specify before you generate

    State the requirement, the constraints and what done looks like. Then generate. Then check the output against what you asked for.

  4. 4

    Verify by running it, not by reading it

    Syntax checks pass happily on silently broken work. Execute it, and look at what actually came out.

  5. 5

    Learn the failure modes

    Confident wrong answers, invented functions that do not exist, security defaults quietly omitted, and scope that grows without anyone asking for it.

  6. 6

    Be able to work without it

    Someone who only functions with an agent running has a tool, not a skill. The tool will change. The skill is what you keep.

  7. 7

    Be straight about using it

    AI-assisted work is normal professional practice in 2026. Hiding it is the problem; using it is not.

Where to go next

The upper rungs of this ladder already have full courses built — Kubernetes, Terraform, Linux, Azure, AWS, Security+ and the NVIDIA AI-infrastructure track.

Browse all learning paths