Short answer: the surface details change wildly — users, regulations, distribution, pricing, even what "good" looks like. The skill that transfers is modelling the user's actual job to be done at a level of specificity no one in the room currently has. Everything else is local. That one thing is portable.
I've shipped across six domains in seven years. AI consumer in two flavors — fashion and content editing. Last-mile B2C logistics in an emerging market. Pharma content SaaS. Vertical ERP for an underserved industry. Outcomes-driven ed-tech.
Every one of those felt completely unrelated when I started. By month three, the same problems kept showing up wearing different clothes.
Five things that turned out to transfer
1. The user's job is almost never what they say it is
In ed-tech, students said they wanted "good content." What they actually wanted was a credential employers cared about. The content was a means; we kept optimizing the means. Once we instrumented around credential-recognition, every product decision got easier.
In regulated SaaS, customers said they wanted "more features." They wanted audit trails that survived a regulator visit. The audit trail was the product. Everything else was packaging.
The pattern: ask "what would happen to them tomorrow if this disappeared?" Their answer is the job. If they say "I'd use a competitor," you haven't found the job yet.
2. Distribution dictates product shape more than product dictates distribution
In emerging-market logistics, our product had to work over patchy connectivity and for an operator earning $5/day. That constraint was the design.
In vertical ERP, our entire onboarding became a single-day data-migration concierge service because that was the only way customers would commit. The product was a Trojan horse for the migration.
You don't adapt your product to your channel. You build the product the channel rewards.
3. Roadmaps shrink the moment a real customer signs
Pre-customer, the roadmap is a wishlist. Post-customer, three things matter: keep this customer alive, fix what they hit yesterday, build the next thing they need to renew. Everything else is theatre.
I've watched teams burn six months on roadmaps that the first paying customer rewrote in a single Zoom call. Get that call earlier.
4. The team you have determines what you can ship
You can't ship a consumer product with an enterprise team. Or vice versa. The cadence, the polish bar, the way decisions get made — they're all team-shaped before they're product-shaped.
If you're scoping a new product, look at the team's last three things they shipped well. That's the shape your next thing has to match. Or you're changing the team.
5. AI doesn't replace product judgment. It compresses it.
The newest layer. In AI-native products, the demo is 80% of the way there in week one. The remaining 20% is where all the actual product work happens — evals, edge cases, the cost-vs-quality trade-offs, the careful prompt scaffolding that makes the magic reliable.
Same product judgment as before. Just faster from blank canvas to "huh, this isn't quite working yet."
The through-line
If I had to bet on one skill across all six domains, it's this:
Get to the specific job, then build the most boring version of the product that does it well.
The fancy version is always available later. The boring version is usually what the user wanted in the first place.
A small operational frame
Whenever I start in a new domain, the same four questions:
- What is the user's actual job? Test by asking what happens to them tomorrow if the product disappears.
- What is the distribution shape? What does this channel reward? Who pays vs. who uses?
- What's the smallest evidence that we're useful? Don't define "successful" — define "useful enough to be worth one more sprint."
- Where is the team strongest? Build to that. Don't fight the shape of the people in the room.
Seven years across six industries — and these four questions still do most of the work. Steal them.