# Garrett Lim, Senior Software Engineer at Peopleworks -- read of SemanticSQL, June 9, 2026

> 9 years full-stack, side-project hunting on Saturday mornings while my kid is at soccer practice. Looking for something I can ship to $5k MRR before I turn 35.

## How I got here

Watched a YouTube video last week about "AI tool-layer startups" where the presenter mentioned text-to-SQL as an underexplored niche. I Googled "build nl2sql saas idea" and this came up second or third on page one. I clicked mostly because the domain sounded like an actual product, not a landing page for a course or a newsletter.

## What I clicked first

The hero says "Query all your data in plain English" and there's a "Start Now (Free)" button. I clicked it assuming it was live software. It is not. The button and "Get API key" in the nav are aspirational copy for an idea that doesn't exist yet. I spent about 30 seconds confused about whether I was on a real product page or not. That confusion cost them my goodwill before I even read the features.

## Where I paused

The scoring block. Specifically: "financial upside: 1/10" and "$-19,774 Year-1 take-home (Fermi)." Negative take-home. They scored their own idea's financial upside at one out of ten. I sat with that for a minute. On one hand, that is genuinely more honest than anything I have read in a SaaS idea newsletter. On the other hand, if they know it is a 1/10 financial upside, why are they selling me the playbook? That is either real transparency or the transparency itself is the product.

## What I distrusted

"Reduce your database query costs by 40% through intelligent deduplication." That number has no source. No customer said it. No benchmark backed it. Someone wrote it because 40% sounds credible without being too round. I have seen that number in three other SaaS feature lists this month.

Also "production-grade SQL automatically." The whole problem with text-to-SQL is that it generates plausible-looking queries that JOIN the wrong tables or silently drop a WHERE clause. Calling it production-grade without showing error rates or failure modes is the exact claim that collapses in a technical demo.

The bigger trust issue: the product description reads like a launched SaaS. The disclosure buried below says "we don't have live customers on this idea yet." Those two things are doing different jobs on the same page. The hero is for dreamers; the disclosure is for CYA. The gap between them is about 500 pixels.

## What would convince me

One real buyer conversation. Not a testimonial, not a persona slide -- an actual transcript where a specific person said "yes, I would pay X per month because my current workflow costs me Y hours per week." That is the thing the Fermi math cannot replace.

Also: the competitive angle. Outerbase, TextQL, Defog, Vanna.ai -- this category has at least five funded players already. The dossier might have a differentiation argument, but the page does not hint at one. "Unified interface" is a feature, not a moat.

## What I'd ask in an email reply

1. The Fermi math shows negative Year 1 take-home. What is the path where this is not negative? What specific assumptions flip it to positive, and have any of those been tested even lightly?

2. You scored buyer clarity 10/10. Who is the buyer -- a data analyst at a Series B startup, or a solo developer running a personal data stack? Those are completely different acquisition motions and I cannot tell from the page.

3. The feature list describes a finished product. Is there working code in the $99 "adopt" tier or is it a scaffold and spec? What percentage of the build is done versus described?

## Verdict: on-the-fence

The radical honesty about financial upside is the most interesting thing on this page and I genuinely have not seen that framing before. But selling me a playbook for $99 while scoring the idea 1/10 on financial upside is a weird pitch, and the mismatch between "real product" and "product idea" is a solvable UX problem they chose not to solve.

---
*Memo by skeptic persona, generated 2026-06-09. Studio breaks own self-grading loop.*