AI Application Builder
You build real products on top of AI models like Claude and GPT, and make them reliably useful in something people use. It’s the fastest-growing job title in tech right now, and the hard part is judging whether the AI’s output is any good.
Related: Software Engineer, Product Manager, Founder
The day in the life
PlantPal · the plan
Answers customers' plant questions
- •Answers from Devi's own notes
- •Friendly, short, shop's voice
- •What a 'good' answer looks like: still undecided
what it should do, before you write anything
You write the rules PlantPal follows. What is its number one rule for talking to customers?
Try a day as an AI application builder
A short, playful taste of the real work.
The single fastest-growing job with computers right now, and there are more of these jobs going than there are people who can do them.
No degree rule. What gets you in is a handful of things you made with AI that other people use every day.
The much bigger numbers you may have seen, $600K and up, are for the researchers who build the models themselves, which is a different job from building things on top of them.
What you’d do all day
The picture is training cutting-edge AI models, but that’s mostly a different job. This one is building on top of models that already exist: writing prompts, wiring the model into a product, and testing it.
Because AI output isn’t predictable, a lot of the work is checking whether the thing is reliable. The real skill is judgment: knowing what models are good and bad at, and verifying the output instead of trusting it.
- Building & coding45%
- Integration & deployment20%
- Testing & iteration15%
- Product & design decisions10%
- Meetings & coordination10%
Nearly half your day is building the thing itself. Most of the rest is connecting it to the other tools and systems it has to work with, then checking all of it works. Nobody asks you what to build yet.
What you build it for changes the job a lot.
- Chatbots & assistantsbuilding something a person talks to and gets real help from.
- Automated workflowswiring AI into a task so it runs itself instead of a person doing it by hand.
- Internal toolsbuilding something a company’s own team uses, not something the public sees.
- Consumer appsbuilding an AI feature the public downloads or signs up for.
A typical early-career day
- 10:00Design the prompt
Work out what you want the AI to do and how to ask for it: the system prompt and the rules around it.
- 11:30Wire it in
Build it into the product with code, connect the model, handle the inputs and outputs.
- 1:30Test on real cases
Throw real, messy inputs at it and find where it breaks. AI fails in weird ways you have to hunt for.
- 3:30Evaluate the output
The core question: is it reliable enough to put in front of people? Judge it, don’t assume it.
- 5:00Direct, don’t trust
Point the AI at the work, then verify, your judgment about what’s good is the whole value.
The role is new and the day varies a lot. Making the model reliably useful, and checking that it is, stays the same everywhere.
The outlook
Where it’s going
This is the hottest corner of tech hiring. "AI Engineer" was the single fastest-growing job title in the US in 2026, with postings up over 140% in a year, and demand is outrunning the number of people who can do it. Companies buy frontier AI models like commodity tools and need people to build real products on top of them, so this is becoming both its own premium career and a baseline skill across all software.
Right now
This is the rare bright spot (it’s hiring hard) against the broader squeeze on entry-level tech jobs. Two honest catches: getting in as a pure generalist is still tough (the routine junior coding is the most-automated work), and the field rewards a portfolio of AI things you have built and put online over degrees, so the way in is to build and publish.
Would you like it?
Worth a look if you like building real things, and work where the answer is never quite the same twice. A lot of the job is deciding whether the AI got it right often enough to let other people use it.
Tap any that sound like you.
In practice, people realize it’s their thing when…
…and it probably isn’t their thing when
How people get in
- 4
- Years it takes on average
- $48K
- Average cost
- $107,000
- What the degree earns four years out
Every way in
A computer science degree
About 6 in 10
Four years · About $48K for the degree
Build things and show them
About 4 in 10
As long as it takes you · Your own time
Both of those are real ways in. A certificate is not a third one: it is something you add on top, and it comes with an end date.
How your first job gets decided
There is nothing to pass and nobody to apply to. The certificates keep getting switched off underneath people, so the things you have built are the part that lasts.
What employers check
- Whether you can work with AI tools at all. About 13 in every 100 job adverts now ask for it, and among the jobs meant for beginners that has nearly tripled in a year.
- The cheapest and most visible AI certificate is not a builder one. The company selling it says on its own page that it is meant for people who do not necessarily build, and the jobs it lists beside it are sales, marketing and project management.
Nothing about this gets decided for you. There is no matching system, no licence and no season, so what you have built is what you turn up with.
- A computer science degree earns about $107,000 four years after you finish. The best paid quarter clear $152,000. The lowest paid quarter are under $74,800.
- Nobody measures pay for this job on its own yet, because officially it is not a job. What a computer science degree earns is the nearest thing anybody can tell you.
When to apply
Nothing here has an application deadline. What does carry dates is the certificates, and every one of those dates is an ending.
- In high schoolNothing to decide and nothing to sign up for. Build things, and keep them somewhere a person can go and use them.
- Every three yearsAmazon certificates stop counting three years after you earn them. One you get at 17 is dead by 20 unless you sit the exam again, which they will sell you at half price.
- Before you start any certificateLook up whether it is being switched off. Microsoft ended 15 of its AI certificates between June 2025 and July 2026, 12 of them in 2026 alone. One was for building agents that go off and do work on their own, which barely existed as a thing when that certificate launched. The list of what goes next is published in advance.
How long it takes
There is no licence and no exam, and the government list of every job in the country has no name for this one. Most people doing it came through a four year computer science degree. The tools they actually use every day are younger than that degree is long.
Certificates here die faster than you can finish one. Something you built that other people actually use does not, and you can have the first one of those by the end of a weekend.
Try building something people can useShip an AI App People Can Actually Trust
Build an app that does a real job, then make it dependable enough for someone else to rely on: write down the things you will try to break it with, find where it breaks, decide how often it has to be right for the person using it, and fix the worst failures first. Deciding what the tool should do, what it must refuse to do, and how good is good enough is what this work really is.
Or try one of these
- Build an iPhone Shortcut Your Friends Will Use2-3 hours
- Automate a Task You Hate Doing3-4 hours
- Build an AI That Actually Knows Your Stuff3-5 hours
- Build an AI That Sorts Your Inbox for You3-4 hours
- Build a Voice Assistant That Does Real Things3-4 hours
- Build an AI Tool for a Real Local Business5-7 hours
- Build an AI Health-Info Helper That Knows What to Refuse3-5 hours
- Build a Custom AI Assistant and Get 10 People Using It2-3 hours
Where these numbers come from
Written by the Sidequest team, from the sources below.
- Pay: No dedicated BLS code; LinkedIn 2026 Jobs on the Rise (AI Engineer #1); Robert Half 2026 Salary Guide; Levels.fyi / Ravio 2026 comp bands.
- The outlook: LinkedIn 2026 Jobs on the Rise; Dice 2026 (fastest-growing role); PwC 2025 AI Jobs Barometer (skills premium). Dated June 2026.
- Getting in: How many people take each route is our own estimate, rounded. O*NET OnLine occupation keyword search for artificial intelligence, across the full taxonomy, read August 2026: no occupation carries the phrase in its title, so no Job Zone, SVP range, education response, median wage or employment projection exists for this career. The nearest real code is 15-1252.00 Software Developers, which the Software Engineer page carries in full; it is that occupation figure and not this one. National Association of Colleges and Employers, Job Outlook 2026 Spring Update, surveyed February 12 to March 17 2026 across 185 employers, read August 2026, for the 13.3% of job postings requiring AI skills and the 10.5% at entry level, which nearly tripled from fall 2025; the survey is across industries rather than specific to this work. Amazon Web Services, AWS Certified AI Practitioner certification page, read August 2026, for the foundational level, the 90 minutes, the 65 questions, the 100 USD price, and the intended candidate wording naming people who are familiar with but do not necessarily build AI solutions, with listed roles including business analyst, marketing professional, sales professional and project manager. AWS Recertification page, read August 2026, for the three year validity, the qualifying recertification exams and the 50% discount voucher. Microsoft Credential retirement register, read August 2026: 36 dated entries parsed, of which 15 carry AI or machine learning names, retired between June 20 2025 and July 17 2026 with 12 of those in 2026, including Azure AI Engineer Associate on June 30 2026 and the Applied Skill for creating agentic workflows on July 17 2026. That register lists what was withdrawn and not what replaced it, and the AWS recertification page shows at least one newer AI certification exists, so no claim is made that the credential layer is shrinking. U.S. Department of Education College Scorecard, field of study data, CIP 11.07 Computer Science at bachelor level, 767 programs and 28,067 graduates, read August 2026, for the $107,000 median four years out, the quarter under $74,800 and the quarter over $152,000; that is computer science broadly rather than anything AI specific. College Board, Trends in College Pricing and Student Aid 2025, public four year in state tuition and fees of $11,950 a year, for the four year total. Dated September 2026.