Adapting to the IT downturn without becoming another “learn AI” cliché

Every downturn grows a slogan. In 2001 it was “get certified.” In 2009 it was “learn mobile.” In 2020 it was “pivot to cloud.” In 2026 it is “learn AI,” said with the confidence of people who have not had to sit across from a hiring manager and explain what that actually means.

I spent five years in technical recruiting before I started writing about education and career transitions. I have also advised bootcamps, which is a polite way of saying I have watched placement decks get more creative as the market got worse. The “learn AI” pitch is not new advice wearing a new logo. It is the same volume play that failed last time: take a scared cohort, point them at the loudest category, and hope the category is still hiring when they finish the module.

Adaptation is real. The downturn is real. Treating a twelve-week prompt course as the adaptation is how you spend a year becoming easier to ignore.

The slogan is doing a specific job

Companies need a story for why they hired fewer people. “We are more productive with AI” is a cleaner story than “we over-hired and finance wants the number down.” Education companies need a story for why their enrollment should not collapse. “The jobs are in AI” is a cleaner story than “the jobs are in the same systems as last year, with a smaller headcount.” Individuals need a story for why this stretch of unemployment or under-employment is temporary. “I’m upskilling” is a cleaner story than “I do not know what to do with Tuesdays.”

All three stories can be partly true and still be a bad plan for you. AI tools are in the workflow. Some roles really did get cheaper to staff. A lot of ticket work that used to justify a junior seat is now a session with an agent and a senior who reviews the diff. None of that implies that the winning move is to become the thousandth person this month with “LangChain” on a resume and no production constraint attached to it.

When I still sat in intake meetings, hiring managers did not ask for “someone who learned AI.” They asked for someone who could own a system the team was afraid to touch. Sometimes that system had a model in it. Usually it had a queue, a database, a messy vendor, and a pager. The model was not the job. The ownership was the job.

What “learn AI” usually produces

I have read the portfolios. They have a shape. A chatbot over a PDF. A weekend wrapper around someone else’s API. A notebook that classifies sentiment on a dataset that came with the course. A blog post that restates the vendor’s getting-started guide in a warmer tone.

None of that is shameful as a first afternoon. It is fatal as a year of strategy. The hiring manager already has three people on the team who can do that on a Thursday. They do not need to pay you to demonstrate it again.

The other product of the slogan is a kind of identity panic. People who were perfectly good backend engineers start introducing themselves as “transitioning into AI.” They bury the five years of payments or identity or data platform work under a new headline that they cannot defend past the first follow-up question. I have watched that happen in screens. The candidate spends ten minutes on the course and two minutes on the system they actually understand. The hiring manager hears: this person is running from their own experience.

If you want to work on models, work on models. Evaluation, cost, latency, data leakage, the boring failure where the demo is fluent and the production path is wrong. That is a lane. Wrapping an API and renaming your LinkedIn headline is not a lane. It is a costume.

Adaptation that is not a course catalog

The downturn changed the demand curve. It did not erase software. It concentrated the remaining seats on people who reduce risk. That sounds abstract until you translate it into a week of work.

Risk, from a manager’s chair, looks like this: a service nobody wants to deploy on a Friday; a migration that has been “almost done” for two quarters; a vendor contract that will explode if one person leaves; a compliance hole that sales keeps promising around; an on-call rotation that pages the same three people. If you can walk into that and make it less frightening, you are adapting. If you can only talk about tools you studied, you are waiting.

I would rather see a mid-level engineer spend the next three months becoming the person who can explain — and change — one painful system than spend those months collecting certificates. The first path is ugly. It means reading logs, writing the doc nobody wrote, pairing with the person who is tired of being the only one who knows the cron. The second path has a progress bar. Progress bars are why the slogan sells.

Own something that already exists

If you still have a job, the adaptation is internal before it is external. Volunteer for the leftover. Not the glamorous rewrite. The leftover: the report finance actually uses, the retry path that drops messages, the test suite that only passes on one laptop. Those are the places where teams discover they cannot afford to lose you. They are also the places that give you stories that survive a frozen interview loop.

If you do not have a job, you cannot own a company’s leftover. You can still own a finished artifact that looks like leftover work: a migration write-up, a load-test that found something, a small tool that replaces a manual ritual you used to hate. Finished is the word. A graveyard of “exploring RAG” folders is the course catalog in another costume.

Move adjacent, not theatrical

Most surviving reqs are not “AI engineer.” They are the same jobs as 2022 with a thinner bench: platform, data, security, mobile release, SRE, integrations, the unglamorous middle of a product. Adjacent moves still work. A backend person who learns enough of the data platform to stop throwing work over the wall. A frontend person who can own the performance budget. A QA person who can make the test suite a product instead of a complaint.

Theatrical moves are the ones that require you to pretend your last five years did not happen. I am not against career changes. I am against career changes that start with discarding the only evidence you have. The market is not kind to people whose story begins at the certificate.

Which skills to keep and which to let rust is its own argument, and it is not the same as “add AI to the list.” If you are mid-career and trying to decide what still compounds, that list is more useful than another foundation-model explainer. The short version for this piece: depth in a system you can still touch beats a shallow survey of whatever venture capital is funding this quarter.

Use the tools without making them the headline

I use coding agents. You should too, if they help you ship. That is not the same as becoming an AI person. Using a compiler did not make you a compiler engineer. Using Copilot does not make you a researcher. The adaptation is output and judgment: you produce more of the work the team already valued, and you catch the cases where the agent is confidently wrong.

If a hiring manager asks how you use AI, answer with a constraint. “I let it draft the migration script and I still wrote the rollback.” “It suggested a query that would have locked the table; I caught it because I have seen that lock.” That is a 2026 answer. “I completed the prompt engineering course” is a 2024 answer wearing this year’s date.

The education market will not save you

I need to be blunt about the industry I used to advise. A lot of programs that sold career-switcher volume cannot place that volume anymore. Some closed. Some rebranded the same curriculum with an AI module and a new landing page. The instructors are often sincere. The incentive is still to enroll people who are scared.

If you are considering a paid program, ask the questions I used to ask when I was the one trying to make the placement number look honest. How many graduates in the last two cohorts got roles that match the pitch, not “any job in tech”? How many of those roles existed before the program started talking to the employer? Can I speak to a graduate who was hired for the work, not for a partnership discount? If they cannot answer without a slide about “the future of work,” you are buying a slogan.

Free courses are cheaper and they have the same failure mode: they feel like motion. Motion is not the same as becoming harder to replace. I would rather you read the incident report from your last job, or from a public outage in your stack, and write down what you would have done. That afternoon will serve you longer than another certificate PDF.

What I would do with the next ninety days

This is not a job-search playbook. That problem — frozen reqs, referrals, take-homes — is a different article, and if you are already hunting, what still works in the hiring freeze is the operational next read. This is about what you become while the freeze is on, whether you have a desk or not.

Pick one system, not five tools. Name it in a sentence a manager would recognize: the billing worker, the identity sync, the deploy pipeline, the inventory feed. Spend the quarter making that sentence longer and more precise. If you cannot name a system, you are still shopping for an identity.

Write the ugly doc. Architecture that exists only in Slack is a gift to whoever is willing to pin it down. The person who writes the current-state doc becomes the person other people ask. That is adaptation you can do without permission and without a course ID.

Get one number you did not have before. Latency, error rate, cost per job, time to restore, number of manual steps. A downturn is full of people who can talk about impact and empty of people who can show a before and after. You do not need a dashboard career. You need one metric you did not invent in the interview.

Talk to three people who still have the job you want. Not influencers. People who shipped last week. Ask what they stopped doing, not what they are learning. I have found that the honest ones say things like “I stopped pretending I would rewrite this” and “I got good at saying no to work the agent can draft.” That is more useful than a syllabus.

Put AI in the workflow, not on the masthead. If a tool saves you an hour, keep it. If a tool becomes the story you tell about yourself, delete the story and keep the hour. Your future manager is buying the hour and the judgment, not the brand of the tool.

A note for people who actually want the AI jobs

Some of you should ignore half of this. There are real seats in evaluation, applied research, data platforms that feed models, inference cost, safety review, and the grim work of making a demo survive a customer’s documents. Those seats were never going to be filled by a general “learn AI” cohort. They are filled by people who already had a related craft — search, data, ML ops, a product with messy inputs — and then went deeper.

If that is you, go deeper. Publish the failure. Show the eval set. Talk about the cost curve. Do not hide in a survey course with people who are there because they are scared of Java. You are not in the same market as them, and pretending you are will make you look like them.

If that is not you, stop apologizing for it. The industry is loud about models because models are a stock story. Most of the remaining software work is still the unglamorous kind: keep the system up, make the change safe, explain it to someone who writes the checks. That work did not become unserious because a chatbot can draft a function.

The downturn is a filter. Do not filter yourself out

I am not going to tell you the market will bounce in a quarter. I do not know. I know that every time we have had a scare like this, a crowd has sprinted toward the same certificate, and the people who came out the other side with jobs were usually the ones who could still talk about a system they had hurt their hands on.

Adaptation, for most of us, is becoming more specific. Specific about what you own. Specific about what you will not pretend to be. Specific about the hour of work you can point at when someone asks what you did with this ugly year.

You can use the new tools. You should. Just do not let a slogan talk you out of the experience you already paid for. The cliché is cheap. Your last five years were not.