Zach Wilson - Data Engineering in 2026, Traveling, and more - Freestyle Fridays (5/8/2026)
TL;DR
Stockholm feels like a real AI contender, not a side scene — After touring Lovable and hearing “agent” five times in his first café hour, Zach Wilson says San Francisco’s “we’re the only city doing AI” vibe looks increasingly out of touch.
Zach’s hottest take is about work, not tooling — He says traditional employment incentives often reward doing “the least amount of work possible to not get fired,” which is part of why he’d rather stay a business owner than go back to a job.
Data engineering in 2027 gets safer as your data gets messier — Wilson argues dashboards are “cooked” and that the safest roles will involve one or more of the 3Vs: high volume, real-time velocity with tools like Kafka and ClickHouse, or unstructured variety like PDFs, images, and transcripts.
For newcomers in 2026, the move is building, not just reading — If you already know Python, he recommends starting with orchestration and simple automations, using Claude to generate code and then learning by running it, breaking it, and watching data move.
AI is hard to teach because the playbook is still missing — Unlike data engineering, where best practices are clearer, AI still has fuzzy standards around prompting, agents, and evaluation, which creates space for vendors to sell certainty that may not really exist.
Most people are overspending on frontier models for tiny tasks — Wilson says Claude Opus is massively overused and compares it to asking Einstein to solve 1+1, arguing that better prompts and cheaper models often get the job done.
Summary
Rooftop in Stockholm, and a surprise AI vibe
Joe Reis opens from a rooftop bar in Stockholm with Zach Wilson in full Freestyle Fridays mode, and Zach immediately says SF “should be scared.” After touring Lovable’s office, he comes away impressed not just by the startup energy but by the fact that they have real revenue numbers to back it up.
Why Stockholm feels different from San Francisco
Zach says the first café he walked into felt exactly like Hayes Valley: laptops out, builders everywhere, and people casually talking about agents. But where San Francisco can feel snobby and “get rich or die trying,” he frames Stockholm as more balanced — a place with AI ambition plus a healthier social baseline and more life outside the grind.
Travel, transition, and a spicy take on jobs
He explains he’s in the Nordics partly traveling with his mom and partly figuring out his next chapter. Even though students want him to keep running AI boot camps, he doesn’t really want to — and that leads to his spiciest take: he no longer really believes in jobs, because many employment contracts incentivize people to do the minimum required to avoid getting fired.
What employees need now: don’t get “tokenized”
When Joe asks how to win as an employee today, Zach’s answer is blunt: know AI. He ties that directly to job displacement and jokes, referencing Joe’s phrase, that people do not want to get “tokenized.”
Data engineering in 2027: dashboards are cooked
Asked where engineering is headed, Zach says the classic Tableau-dashboard world isn’t where you want to be anymore, even if those jobs still exist. His framework is the 3Vs: volume, velocity, and variety — with high-volume data, real-time systems like Kafka and ClickHouse, and unstructured formats like PDFs, images, text, and transcripts all offering more durable value because AI still struggles with those edge cases.
If you’re new, start building with Claude
For beginners, his advice depends on your baseline, but if you already know Python, he says to build orchestration and automation right away. Instead of just reading, he says to ask Claude for code, run it, see what breaks, and learn by getting your hands dirty — a point Joe explicitly agrees with, even while they both still value fundamentals.
Why AI is messier to teach than data engineering
Zach says he keeps getting pulled toward AI even though he doesn’t entirely enjoy teaching it. The reason: unlike data engineering, AI lacks settled best practices, so questions like what makes a good prompt, agent, or evaluation often collapse into “whatever makes the business process work,” which is useful but much fuzzier.
Better prompting, cheaper models, smarter use
His concrete AI best-practice advice starts with prompting: don’t just type “fix this,” because underspecified prompts kneecap the model. He says giving the model 10% more context can make it 80% more effective, and he also argues people wildly overuse Claude Opus for trivial tasks — like hiring Einstein to do 1+1 — when a smaller, cheaper model would be perfectly fine.
Closing on the road
They wrap with Zach’s travel itinerary — Copenhagen, Oslo, Bergen — plus a quick nod to the upcoming Databricks AI Summit. The whole thing ends in good spirits, with rooftop-party energy, travel plans, and a sense that both the AI map and the data engineering career map are shifting fast.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.