Career Change to AI: My Honest Plan to Pivot Into AI, Data & Engineering

The moment I decided to do this wasn’t dramatic. I was staring at a campaign dashboard, untangling why a conversion-attribution model was quietly lying to us, and it hit me that I hadn’t done anything that looked like “marketing” in weeks. I was cleaning data, reasoning about a model’s assumptions, and automating the boring parts. The job title still said marketing. The work had become data and engineering.

This is the first article on this site, and it’s the honest one: why I’m pivoting into AI, data, and engineering, and the actual plan I’m running. Not a motivational leap-of-faith story — a rational response to a market that has changed underneath all of us. If you’re considering the same move, treat this as a companion plan you can copy and argue with.

Why now

The standard advice is either “it’s too late, AI ate the entry-level jobs” or “just grind LeetCode and you’ll be fine.” Both are wrong, and they’re wrong in the same way: they treat “tech jobs” as one thing.

They’re not one thing anymore. The market split in two. That framing is important enough that it gets its own article, but the short version drives everything in my plan:

• The generalist-junior lane contracted. Software-developer employment for the 22–25 cohort is down roughly 20% since 2024 (Stanford HAI’s 2026 AI Index), and general software postings sit well below where they were pre-pandemic.
• The AI-adjacent lane grew. LinkedIn data from early 2026 showed AI/ML engineering postings up about 34% year over year while overall tech postings fell around 8%. “AI engineer” is one of the fastest-growing roles in the world.

So “is it too late?” is the wrong question. The honest question is: which lane am I aiming at? I am deliberately not aiming at the contracting one.

What changed, and why it matters to a career changer

Two more facts shape the plan. The World Economic Forum’s Future of Jobs 2025 names AI and big data the fastest-growing skill set through 2030, and LinkedIn estimates roughly 70% of the skills used in the average job will change by then. Meanwhile Stack Overflow’s 2025 developer survey found about 84% of developers already use or plan to use AI tools — but their trust in those tools is mixed.

That last point is the opening. A churning skills market plus low trust in AI output is precisely the environment where someone who shows judgment — who can say where the tool breaks, not just that it’s exciting — is valuable. I don’t need to out-credential a CS graduate. I need to be demonstrably useful in the lane that’s growing.

The skills I’m keeping (the part nobody tells you)

The biggest myth in career changing is that you start over at zero. You don’t. A decade in digital marketing and ad tech left me with skills that map almost one-to-one onto data and AI work:

• Experimentation. A/B testing is the experimental method. Designing a clean test, worrying about sample size and confounds — that’s day-one data science thinking.
• Attribution modeling → causal reasoning. Years of arguing about what actually drove a conversion is years of practicing causal inference under messy, real-world conditions.
• Audience segmentation → clustering. Grouping users by behavior is the intuition behind a whole category of ML.
• Campaign analytics → BI and metrics design. I already know that a dashboard nobody acts on is worthless, and how to choose a metric that drives a decision.
• ROI thinking → model business value. I instinctively ask “what is this worth?” That question is rare and prized on technical teams.

I go deep on this in What Ad Tech Taught Me About AI, because it’s my single most unfakeable advantage. A big SEO site can out-rank me on “what is a neural network.” It cannot copy ten years of running experiments against real budgets.

What I’m deliberately de-prioritizing

A pivot dies from trying to learn everything. So I’m explicit about what I am not doing right now:

• Not chasing a CS degree. I have no degree in this and I’m treating that as a framing problem, not a wall — more in How to Get Into AI Without a CS Degree.
• Not collecting certificates as a substitute for shipped work.
• Not learning three languages and five frameworks “to be safe.” Depth over breadth.
• Not leaning on a Tokyo / bilingual angle. I’m based in Tokyo but I’m not Japanese-bilingual, so that’s flavor, not strategy.

The runway plan (the unsexy part that makes it real)

This is where most “follow your passion” posts go quiet, so I’ll be specific about the principle. A pivot without a financial plan is a bet you can be forced to fold at the worst moment.

My rule: keep a 6–12 month reserve and learn alongside paid work rather than quitting into a void. The reserve isn’t for “if it fails” — it’s so I can make slow, correct decisions instead of panicked ones, and so I can say no to the wrong first role.

The trade-off is honest: learning while working is slower and more tiring than a full-time bootcamp sprint. I’m choosing it anyway, because it keeps the proof-of-work real (I’m solving actual problems) and the runway intact.

My learning roadmap for the next 6–12 months

Sequenced so that every phase produces something shippable, not just consumed:

1. Foundations that compound: SQL first (it made messy problems feel tractable faster than anything else), then Python, then Git and a little cloud. The minimum viable stack, not the maximal one.
2. One Kaggle competition, done properly — not for the score, for the reasoning. The value is the documented thinking, not the leaderboard.
3. Build something real with an LLM API — a tool that solves a problem someone actually has — and make it trustworthy, not just a demo: evals, monitoring, the boring reliability work.
4. Learn the AI tools well enough to be honest about them — fluent use and a clear view of the failure modes.

The projects that prove I can do the work

Résumé lines don’t survive contact with a good hiring manager; deployed, used artifacts do. My bar is two or three real tools on real (messy) data, each with a write-up of the decisions, trade-offs, and the things that broke. I’m building toward that now — and as those projects ship, the write-ups will land here. That’s the whole point of doing it in public.

How I’ll track progress (and what year one actually looks like)

The failure mode of “learning in public” is consuming in public — endless courses, no output. So my success metric isn’t hours studied; it’s shipped. Concretely, year one looks like: the minimum stack in working order, a couple of deployed tools that someone other than me has used, a steady stream of honest write-ups, and at least one foot inside an AI-adjacent role or project.

The plan is simply to keep writing it down as I go — what I set out to do, what I shipped, what failed, and the numbers. Doing it in public is the forcing function.

If you’re making the same move: don’t aim at the lane that’s shrinking. Keep the skills you already have — they’re worth more here than you think — pick a runway you can defend, and measure yourself by what you ship. That’s the whole plan. The rest of this site is me actually running it.

---

This is the start of an ongoing, in-public pivot. New here? The articles linked above are the best next reads, and each one ends with a “next best read” of its own.