"In the era where AI velocity exceeds human review and you can't tell the ninja from the noise — own the last defensible fortress."
The Lie We've Been Telling Ourselves
Every best practice in modern software engineering is a social contract pretending to be an engineering solution.
Code review. PR checklists. Linting pipelines. Clean architecture. SOLID principles. Testing pyramids. These work beautifully in a fantasy team — small, senior-heavy, stable, everyone reads the architecture docs, everyone cares.
That team doesn't exist at scale. It never really did.
And in 2025, it's completely, irreversibly dead.
AI Just Nuked the Last Pretense
The old control loop was: slow down output → review → gate → merge.
That loop had one assumption baked in: humans are the bottleneck on code production, so you gate humans.
That assumption is gone. The code is now infinite and free. Your junior dev with Cursor ships 10x what your senior shipped in 2019. Your contractor uses Claude Sonnet and delivers a feature in a day that used to take a sprint. Your intern doesn't even write code — they write prompts and review diffs they barely understand.
And here's the part nobody's saying out loud:
The tests are AI-generated too. They test what the AI thought it was building.
The PR looks immaculate. Consistent style. Well-documented. Green CI. And the business rule is subtly, quietly wrong — because the agent didn't know about the edge case your senior learned from an incident three years ago that never made it into a spec.
You can't add enough reviewers. You can't make the checklist long enough. You cannot out-review the velocity.
The Sacred Cow That Breeds From Slop
Here's the current reality of "business logic in application code":
Your rules live in service classes, middleware, serializers, validators — spread across layers, replicated across services, enforced by convention and prayer.
Every new dev who joins — bootcamp grad, offshore contractor, AI-assisted intern — touches that code. Every PR is a potential bypass. Every "quick fix" is a load-bearing card in a house of cards.
And now the code itself is generated by a machine that has no memory of why the rule exists, no context for the incident that created it, and no skin in the game if it gets it wrong.
You have perfect version control over slop. Git tells you exactly which commit introduced the wrong behavior. Congratulations. The audit trail of your own mistakes is immaculate.
Meanwhile, you're doing standup, sprint planning, architecture reviews, and PR approvals — all of it cope. Elaborate institutional cope to manage the entropy of logic that was never safe outside a controlled environment.
The Churn Math Nobody Wants to Do
Ask yourself honestly: which role churns faster at your company?
The JavaScript developer — framework-hopping, chasing the ecosystem, mentally rewriting their stack every 18 months. React → hooks → RSC. REST → GraphQL → tRPC. Webpack → Vite → whatever drops next Tuesday. Their knowledge doesn't compound. It rotates.
Or the bearded guy with the SQL wand.
The senior DBA who's been writing joins since before your current frontend framework was a GitHub repo. Whose CTEs from 2015 still run. Whose understanding of indexes, vacuuming, and query planning compounds year over year. Postgres 17 still respects everything he wrote in Postgres 10.
You also cannot fake SQL mastery in an interview. Run EXPLAIN ANALYZE on the query. No leetcode theater. No rehearsed system design answers. Either the query plan shows 3ms with an index-only scan, or it shows a sequential scan on 2 million rows and you're done. The database is the interviewer, and it has no unconscious bias and cannot be charmed.
Meanwhile your JS interview pipeline is a gymnasium for people who memorized 50 leetcode patterns for 3 weeks. They pass. They join. You find out 6 months later.
Own What AI Cannot Slop
Here's the reframe:
There are now two zones in your system.
Above the line: AI territory. Generate freely. Redeploy in minutes. Let the contractors, the juniors, the agents write it. Move fast. Break things. The BFF, the frontend, the API adapters — disposable by design. This is where velocity lives and chaos is acceptable.
Below the line: The fortress. Slow. Deliberate. Council-gated. Changes are rare and ritualistic. This is where correctness lives and velocity is the enemy.
The line is the database boundary.
And the crucial insight: you cannot prompt your way into changing a trigger. The AI can generate a million BFF variations. Every single one hits the same wall. The invariant holds not because anyone checked the PR, but because the system enforces it at a layer where velocity doesn't operate.
What the Postgres Temple Actually Looks Like
The PostgreSQL-centric architecture — sometimes called Planetary Architecture — inverts the conventional stack:
-
Raw tables are locked.
REVOKE ALLfrom everyone including the application user. Superadmin only. - Views are the only access surface. One table, many views — scoped by role, redacted by sensitivity, filtered by RLS policies.
- INSTEAD OF triggers are the only mutation path. Business logic lives in Postgres functions called by triggers. Not in your service layer.
- Row-Level Security is authorization. Not middleware. Not JWT decoded in a controller. The database itself decides what each role can see and touch.
- FDW forensic audit writes to a separate database server on a separate host. Append-only. If your main DB gets ransomwared, the audit trail is untouched.
The application user — the one your PostgREST or Django connects with — can see metadata on raw tables but never read a row directly. The BFF consumes views and calls functions. It has no idea what the underlying schema looks like.
A junior dev, an AI agent, an offshore contractor — they work on the BFF. They literally cannot bypass business logic because there is no surface to touch. The database rejects it at the query level.
The Seniors Become Force Multipliers
This architecture changes what "senior" means.
In a conventional stack, seniors spend their leverage on reviews, mentoring, and trying to slow down the entropy of a dozen people touching business logic simultaneously. They're playing defense constantly.
In the Postgres temple, the senior writes the fortress once, correctly. Encodes the rules into trigger functions, RLS policies, view contracts. Then steps back.
The chaos above the line doesn't require their supervision anymore — because the architecture does the supervision. One council of SQL wardens can safely oversee an arbitrarily large BFF team of agents, juniors, and contractors. The blast radius of any mistake above the line is hard-bounded by what PostgREST exposes, which is bounded by what views exist, which is bounded by what the senior built.
The more junior and AI-heavy your above-the-line becomes, the more valuable the below-the-line fortress gets. The chaos above justifies the rigidity below.
This Isn't a 90s Idea. It's the Correct Response to 2026
The Oracle shops of the 90s had this right. Business logic in stored procedures. Schema changes as rituals. DBAs as priests. Nobody pushed to production without the warden's blessing.
We abandoned it because startups needed to move faster than that culture allowed, and Rails made it feel unnecessary. We traded correctness guarantees for velocity, and it worked — when humans were still the velocity bottleneck.
That trade no longer makes sense. The velocity problem is solved. The correctness problem is now existential.
The bearded SQL warden who stays 10 years, whose fortress gets stronger over time, whose knowledge doesn't deprecate — that person is now the most strategically valuable engineer in your organization.
Not because SQL is magic. Because owning the one layer that AI cannot slop is now the entire game.
The One Sentence
Your code is a sacred cow bred by a slop machine, kept alive by PR prayers and best-practice inhalers.
Put your sacred cow in Postgres, where the DB is the priest, and no agent, no junior, no tired tech lead reviewing his 40th PR of the week can accidentally sacrifice it on a Friday deploy.
The fortress holds. The orbit burns. That's the architecture.
Inspired by PostgreSQL centric - Planetary Architecture by Ryo Suwito.
Tags: #postgres #architecture #systemdesign #webdev #programming
Top comments (0)