Systems · September 04, 2026

WANTED: TEN YEARS' EXPERIENCE IN SOMETHING INVENTED IN 2022

WANTED: TEN YEARS' EXPERIENCE IN SOMETHING INVENTED IN 2022

Everyone became senior in 2023.

ChatGPT was released on November 30, 2022. As I write this it is three years and nine months old. It cannot legally hold a driver's license in any jurisdiction on earth, and there are job advertisements asking for ten years' experience with it.

One of those ads did the rounds in 2024, when the technology was not yet two. Somebody on the internet observed that the ideal candidate would need "good time-traveling skills." That remains the most precise piece of job-market analysis I have read this decade.

None of this is new. The species has form.

The man who was underqualified for his own invention
In July 2020, a developer called Sebastián Ramírez saw a job ad requiring four or more years of experience with a Python framework called FastAPI. He could not apply. He had a year and a half of experience with it, because a year and a half earlier he had written it. The tool's author did not meet the minimum requirement to use the tool.

That story was funny in 2020 because it was rare. It is less funny now because it is the standard operating condition of an entire industry. Nobody has ten years of experience in generative AI. A small number of people at Google, OpenAI and a handful of universities have something approaching nine years with the transformer architecture underneath it, and in my experience they are the least likely people alive to put the word expert in a bio. They are too busy.

The great renaming
According to an Indeed analysis reported by Business Insider, the number of frequently advertised job titles that explicitly mention AI went from 264 in 2022 to 822 in the first quarter of 2026. Nearly two thirds of them sit outside technology. AI marketing manager. AI learning specialist. Responsible AI counsel. AI transformation lead. These are, for the most part, existing jobs that have discovered a keyword.

AI now appears in more than four percent of all Indeed job postings. Postings mentioning it are up 134 percent on early 2020. Total postings are up six. The jobs did not multiply. The syllables did.

At the top of the org chart, postings for chief and senior AI roles have climbed by triple digits, on Indeed data reported by IEEE-USA InSight. Genpact has gone further and appointed an Agentic AI Officer. Around the same time, LinkedIn's own head of engineering told Fortune that nobody can clearly define what an AI agent is.

So we have arrived at the appointment of officers for a thing that has not yet been defined. Medieval courts did something similar with the Master of the Horse, though in fairness to the medieval courts, they had seen a horse.

An executive recruiter quoted by IEEE-USA InSight supplied the anthropological context. The chief AI officer follows the chief cloud officer and the chief e-commerce officer, roles that arrive with the technology, sit at the top table for a few years, and are then quietly absorbed into the furniture. The title is a temporary structure. Everyone standing in it is behaving as though it is load-bearing.

Senior
The word has been asked to do something it was never designed for.
In medicine, senior implies a decade. In law, longer. In generative AI, a senior anything has been doing it for at most three years and nine months, and that is the theoretical maximum, available only to people who started on launch day and have not slept since. In practice, senior means was paying attention in 2023.

This is not a criticism of the people holding the titles. Most of them are doing hard work with the experience that exists. It is a description of what happens when a profession compresses a forty-year seniority ladder into a single product cycle. Every rung is still there. They are just very close together.

The mayfly
The prompt engineer is the cleanest specimen. In 2023 it was the job of the future. The people who held it were called AI whisperers, and salaries were reported as high as $375,000 for roles that did not require a technical background. By 2025, industry commentators were declaring the job obsolete. By 2026 the standalone title was thinning out, while the skill it named had been folded into roles that do not mention it.
Born, celebrated, declared dead and quietly absorbed into everyone else's job description in three years. Careers used to take four decades to do that. Some insects have longer arcs.

A declaration of interest
I am building one of these systems. I have used words this year that I could not have defined the year before. I have described myself, in professional settings, with a title I am not sure I would recognize on someone else's card. The field guide includes its author.

Why the ads say ten years
The people writing the ads are not fools either. A hiring manager takes the old Senior Data Scientist description, which correctly asked for five years, and sprinkles LLM on it because the technology moved faster than the template. Years of experience were never the point. They were a proxy for someone who understands the pipes, not just the tap. The proxy has failed because the pipes are three years old.

What happens next is predictable. The applicant tracking system asks for years. The honest answer is zero, or two. The form rejects the honest answer. The field begins, gently, to select for people who round up.

The right unit
Years are the wrong measure for a technology this young. The right questions are shorter and harder to fake. What did you build? What broke? What would you do differently? Show me something that survived contact with a user.

Nobody has ten years of experience in this. Some people have three good ones. They are worth more than a decade of the other kind, and they are almost never the ones with expert in the bio.


Sources
1. Sebastián Ramírez, post on X, July 11, 2020: a job ad required four or more years of FastAPI experience. He had a year and a half, having created the framework. FastAPI's initial release was December 2018. https://x.com/tiangolo/status/1281946592459853830

2. PYMNTS, AI Is Coming for Your Job Title, July 18, 2026, reporting an Indeed analysis carried by Business Insider: frequently advertised job titles referencing AI rose from 264 in 2022 to 822 in Q1 2026, nearly two thirds outside traditional technology fields, with examples including AI marketing manager, AI learning specialist, responsible AI counsel and AI transformation lead. Also reports Indeed Hiring Lab data: AI in a record 4.2 percent of postings in December 2025, postings mentioning AI up 134 percent on February 2020 against total postings up 6 percent. https://www.pymnts.com/news/artificial-intelligence/2026/ai-is-taking-your-job-title/

3. IEEE-USA InSight, Chief AI and Senior AI Job Postings Soar by Triple Digits, But Face Uncertain Future, April 29, 2026. Indeed data on triple-digit growth in chief and senior AI postings. Executive recruiter on the chief cloud officer and chief e-commerce officer cycle of emergence and absorption. https://insight.ieeeusa.org/articles/chief-ai-and-senior-ai-job-postings-soar-by-triple-digits-but-face-uncertain-future/

4. Fortune, Sage Lazzaro, Why Genpact has an Agentic AI Officer and why she dislikes the term 'human in the loop', November 14, 2025. https://fortune.com/tag/chief-ai-officers

5. Fortune, John Kell, No one can clearly define an AI agent. LinkedIn's engineering boss says that's okay. Mohak Shroff, SVP and head of engineering at LinkedIn, on the absence of a crisp definition of an AI agent. https://tech.yahoo.com/articles/no-one-clearly-define-ai-144513712.html

6. Salesforce Ben, Prompt Engineering Jobs Are Obsolete in 2025, May 1, 2025. Prompt engineers labeled AI whisperers in 2023. Roles advertised with salaries as high as $375,000 without requiring in-depth technical knowledge. https://www.salesforceben.com/prompt-engineering-jobs-are-obsolete-in-2025-heres-why/

7. 4Geeks, AI Prompt Engineer in 2026, August 2026. The standalone Prompt Engineer title in decline while roles requiring prompt-engineering skills expand. Secondary source, used for direction only. https://4geeks.com/en/blog/ai-powered-learning/ai-prompt-engineer

8. Memebase, Job posting demands 10 years experience using Generative AI, April 15, 2024. Source of the widely shared 2024 ad and the "time-traveling skills" line. https://cheezburger.com/25458949/job-posting-demands-10-years-experience-using-generative-ai-despite-chatgpt-coming-out-less-than-2

9. Medium, Y. Dedeepya, The Great AI Paradox: How to Have "5 Years of Experience" in a 3-Year-Old Tech, February 1, 2026. On hiring managers reusing older data-science job descriptions with generative AI terms added, and years of experience as a proxy for depth.
https://medium.com/@dedeepyayarlagadda/the-great-ai-paradox-how-to-have-5-years-of-experience-in-a-3-year-old-tech-bfa0d70f1705

The newsletter

One email, when there is something worth sending.

A tool from one of the books, a sample chapter, an early copy before publication, word of a new edition. Every template and excerpt on this site arrives through the same door.

Unsubscribe in one click. Your address is never shared.

Related Books

Stop Feeding the System

A field guide for recognising and recovering from narcissistic abuse, written from inside five decades of it. Eight named tools, from the Pattern Ledger to the Self-Trust Rebuild. Not a diagnosis. A map.

Related Articles

The money went out. Nobody counted what came back.

Spending on AI is recorded precisely. What it returns, in most organisations, has never been written down. This piece sets out what the published research actually says, including where the widely quoted numbers have been compressed past what the studies support. It covers the Australian position after the mandatory guardrails were shelved, the automated decision-making disclosure obligation that takes effect on 10 December 2026, and the finding that thirty-seven per cent of Australian boards have audited what their organisation is running. It closes on a conflict of interest the category has not addressed and an eighteen-month mistake of the author's own.

Your business is not missing from AI. It was never a candidate.

Everyone is selling AI visibility. Almost nobody has a denominator. We put buyer questions to AI assistants about ten trading Melbourne businesses, repeated them, and counted. Out of 570 chances to be named, they were named 18 times. Four were never named once. Then the finding that matters: asked whether it knew the specific business, nine of the ten got both answers from the same machine, some runs yes and some runs no. An independent study from SparkToro found the same instability from a different direction, with under a 1 in 100 chance of the same brand list appearing twice. This article covers what that does to every AI visibility report on the market, why the failure happens before your content is read, and the four questions to ask any vendor selling you a number. It includes our own business as the worked case, along with two findings we published about ourselves that turned out to be wrong.

The Slow Ledger

A man who once managed a hundred and sixty-five billion dollars announces the end of the arrangement between ad reads. The essay takes what holds in his warning, that bubbles form around true ideas and that advisers answer to a different number than yours, then turns his own instrument on him. Underneath sits one portable mechanism: fast numbers celebrated, slow numbers drawn down, the bill arriving later, addressed to someone younger.