Journey Uncommon Logo
JourneyUncommon
Stay valuable

AI can write the code.
It can't make the call.

The SDK wiring is free now. What keeps you valuable is knowing what to hand to AI and what you must own yourself. Learn the tech around the product, the judgment that doesn't get automated away.

The problemWhat you ownHow to practice

Start with the problem

Payments

Real money, irreversible — where a bug costs customers and trust.

It's the one place a bug doesn't throw an error. It quietly double-charges a real customer, or loses revenue and tells no one.

You own

The idempotency strategyThe reconciliation logic
See what to own + practice →6 to learn · 4 to own

Observability

Knowing what your system is actually doing — before users tell you.

Checkout is failing for 3% of users on one browser right now. Nobody filed a ticket. AI wrote the code but can't tell you it's on fire.

You own

Deciding what's worth measuring (the targets/SLOs)Judging signal vs noise during an incident
See what to own + practice →6 to learn · 4 to own

Auth & sessions

Get this wrong and it's a breach, not a bug.

Get this wrong and it's not a bug, it's a breach. 'Works on my machine' is exactly where a real attacker starts looking.

You own

The token storage + expiry/refresh strategyThe authorization rules (who can do what)
See what to own + practice →6 to learn · 4 to own

Caching

The cheapest speed win — and the source of the weirdest bugs.

Cache makes slow things fast, until a user sees a stale balance, an old price, or a permission that should have changed. Then it's the weirdest bug you'll chase all week.

You own

What's safe to cache and for how long (the staleness call)The invalidation strategy
See what to own + practice →6 to learn · 4 to own
▸Which AI to use, by area (7)

The give/own split is durable; the tool names move fast. Mid-2026 snapshot grounded in real user sentiment. Tool names, prices, and limits move in weeks — re-check before publishing. The give/own split is durable; the named tools are not.

AI coding agent

Claude Code — Most-praised coding brain for hard multi-file refactors and architecture.

Give: The grind: multi-file edits, boilerplate, make-it-pass loops.
Own: The architecture, and the diff you actually merge.

Frontend / UI

Claude Code (Sonnet) — Cleanest React structure and the best design taste in a real repo.

Give: First-pass components, Figma-to-code, throwaway prototypes.
Own: Design-system fit and the last 30% to production.

Backend / data / SQL

Claude Code (Opus) — Cleanest, most idiomatic server/API code; large context helps on big codebases.

Give: Endpoints, migrations, scripts, SQL against a real schema.
Own: Schema design, prod data, and the query that hits RUN.

Review / test / debug

CodeRabbit — The low-noise default first-line PR reviewer; predictable flat pricing.

Give: First-pass PR review and edge-case test generation.
Own: Whether to merge, and judging the generated tests.

DevOps / infra

Claude Code — Strong at writing Terraform/Helm/K8s manifests and CI/CD.

Give: Drafting IaC, diagnosing clusters, explaining errors.
Own: `apply`, state, and any live-cluster remediation.

Which model, for what

Claude — Hard, long-horizon, multi-file coding and deep reasoning — the 'own-it' work.

Give: The task you can verify cheaply.
Own: The task where a wrong first pass is expensive.

PM / design / data

Claude — Writing-heavy work — PRDs, specs, natural prose, clean SQL.

Give: First drafts, transcripts, theme extraction, SQL boilerplate.
Own: Judgment, context, and the final call.

Don't just read it. Practice it.

Work a real open-ended scenario and get AI feedback on your reasoning — assumptions, structure, the call you'd make. Free.

Practice your judgment →