Should AI Design Your Database Schema? Why the Data Model Needs Review
By Matt Cullerton, founder of Mavric and Apso
A TikTok from @abhibuilds1 argues that you should never let AI one-shot your data model. The advice holds up, though the spoken version is softer than the title card: he recommends using AI as a co-pilot that interviews you and helps with trade-offs, with a human reviewing the schema before it ships. The strongest support for his point comes from the PostgreSQL manual, which documents why schema mistakes get expensive as a table grows.
What the video actually recommends
The title card reads "Never let AI one-shot your data model." The description softens it to "Don't let AI one shot core things that get harder to change as your product grows." The audio is softer still: "You can definitely use AI, but you don't want to use AI to do the whole thing." He suggests you "could have it interview you" and make trade-off decisions, "but you definitely don't want it to essentially one shot the data model and you didn't review it."
So the card bans AI from schema work, while the speaker wants AI in the loop from the requirements interview onward, with a human signing off. A viewer who reads only the card gets the wrong instruction.
The slide behind him for the whole clip asks "Who pays for a mistake?" and sorts work into three tiers: a demo on fake data, a product with real users, and shared production owned by a team or company. The audio never mentions those tiers. It talks only about the data model.
Why is a database schema hard to change later?
The video says the data model is "going to be crucial to your app, your analytics, to, you know, everything basically." The PostgreSQL manual's ALTER TABLE page gives the specific reason.
It states that "An ACCESS EXCLUSIVE lock is acquired unless explicitly noted." It also states that "Changing the type of an existing column will normally cause the entire table and its indexes to be rewritten." Adding a column with a volatile default, a stored generated column, an identity column, or a domain type with constraints triggers the same full rewrite. The page warns that rebuilds "may take a significant amount of time for a large table, and will temporarily require as much as double the disk space."
That is the mechanism behind "harder to change as your product grows." A column type chosen wrongly on day one costs a full table rewrite and an exclusive lock to fix later, and the cost scales with the data you have accumulated.
GitHub hit this at scale. Its gh-ost announcement from 1 August 2016 says the team reached the point of "considering many migrations as 'risky operations'" and describes testing on tables "as small as empty and as large as many hundreds of GB." GitHub built and open-sourced a migration tool to avoid lock contention on large tables.
How to have AI interview you before it writes the schema
The interview step he describes already exists as a command. GitHub's Spec Kit repository, created on 2025-08-21, includes a clarify command described as "Identify underspecified areas in the current feature spec by asking up to 5 highly targeted clarification questions and encoding answers back into the spec."
The README's flow runs /speckit-constitution, /speckit-specify, /speckit-plan, /speckit-tasks, /speckit-implement and /speckit-converge, with the instruction to "Run these skills in your agent's chat, not your terminal, one at a time, reviewing each result before continuing." On 2026-10-22 the repository showed 142,436 stars and release v1.2.3.
That gives a viewer something more concrete than the video's advice to "communicate with it really well."
What "co-pilot" means, according to GitHub
The video closes by saying of co-pilot, "just keyword is co, you're, you're a part of that as well." GitHub's Copilot launch post from 29 June 2021 calls it "a new AI pair programmer that helps you write better code" and does not frame the name around the prefix.
The conclusion is still right, and GitHub states it directly. Its responsible-use card for inline suggestions says "Users are responsible for reviewing and validating suggestions before accepting them to ensure they are accurate and appropriate." It also warns that "Copilot may generate code that appears to be valid but may not actually be semantically or syntactically correct." Output that looks right and is not is the failure the video describes.
Does AI make developers slower?
The video cites no study, and the most-shared number in this debate is out of date. METR's July 2025 study had 16 experienced developers complete 246 issues in repositories they maintained, using Cursor Pro with Claude 3.5 and 3.7 Sonnet. With AI tools, they took 19% longer, with a 95% interval from 2% to 39% longer. The same developers forecast a 24% speedup beforehand and reported a 20% speedup afterwards.
That page now carries a banner reading "These results are out of date." METR's 24 February 2026 update describes a second study with 57 developers, more than 800 tasks and 143 repositories. Completion time came out 18% lower for returning developers (95% interval from 38% lower to 9% higher) and 4% lower for new developers (15% lower to 9% higher). Both intervals cross zero. METR reported that "30% to 50% of developers told us that they were choosing not to submit some tasks because they did not want to do them without AI," and treats its estimate as a lower bound.
So the case for reviewing AI-generated schemas rests on how databases behave and on vendor documentation, and the productivity research does not settle it either way.
What to check in an AI-generated data model
The video never shows a schema, migration or tool. The documented checks are:
- Will any column type need to change later, forcing a rewrite under ALTER TABLE?
- Does any column use a volatile default?
- Will a future migration hold an ACCESS EXCLUSIVE lock long enough to matter at your table size?
- Were the questions a clarify step would ask ever answered?
Bottom line
The advice is sound, and the title card overstates it. Use AI to interview you and draft the schema, then review it yourself, because PostgreSQL documents exactly how costly a wrong column type becomes once a table is large.
Sources
- ALTER TABLE, PostgreSQL documentation - lock behavior and table rewrite conditions
- gh-ost announcement, GitHub Blog, 1 August 2016 - GitHub on risky schema migrations
- Introducing GitHub Copilot, GitHub Blog, 29 June 2021 - the "AI pair programmer" framing
- GitHub Copilot inline suggestions application card - the user review obligation
- github/spec-kit - spec-driven development commands, including clarify
- METR early-2025 developer productivity study - the 19% slowdown result, now marked out of date
- METR experiment design update, 24 February 2026 - second study results
- METR research index - list of METR publications
Published October 22, 2026. Based on a TikTok by @abhibuilds1. Full video notes (markdown)