# Marcus Okafor, Senior Data Engineer at Threadline (e-commerce, ~220 employees) — read of BigQuery AI Table Analyzer, June 22 2026 > 8 years in data, started as a SQL analyst, now own the warehouse at a Series B apparel company. Stack: BigQuery, dbt, Airflow, Looker, and a Confluence graveyard nobody opens. Coach my daughter's U9 soccer team Saturday mornings, which means I do most of my side-project research on Friday nights after she's in bed. --- ## How I got here My team lead pinged me Tuesday asking for "table-level docs" before our next sprint review. We have 74 tables. I've documented maybe 12. I typed "auto document bigquery tables" into Google, clicked through three results that were all dbt meta or generic AI wrappers, then hit this on page two. The "No Auth" in the hero CTA was the reason I didn't close it immediately. I've been burned by "free trial" signups that just collect email before showing me anything. --- ## What I clicked first The terminal animation in the hero. Specifically this: > `$ bq-analyze --project my-project [Scanning 47 tables...]` That got me. CLI tools that do one thing well are usually worth five minutes. The output showing `users`, `events`, `revenue`, `metrics` is obviously staged but the framing is honest about what it does. I read the next line below the hero: "reads your dataset and explains what each table means in plain English." Fine. Clear enough. I kept scrolling. --- ## Where I paused The Relationship Detection feature: > "Automatically identify foreign keys, nested structures, and cross-table patterns." I stopped here for a bit. BigQuery doesn't enforce foreign keys. There are no FK constraints in the schema to read. So either this is doing something clever with naming conventions and column overlap, or it's hallucinating relationships based on column names that look similar. That's a real difference. "Find dependencies you didn't know existed" either means it's doing something genuinely smart or it's going to tell me `user_id` in my `events` table "probably" relates to `id` in my `users` table and call that a discovery. I'd want to know which one before I trust it on anything important. --- ## What I distrusted Two things. First, the numbers: "cutting documentation time by 90 percent" and "Cut dataset exploration time by 80 percent." Where do these come from? No attribution. No "based on 12 data teams" or "internal testing across X tables." Just percentages floating in the air. After eight years I ignore these on sight. Second, and this is the bigger one: I scrolled to the bottom expecting pricing or a signup form and found this instead: > "Honest disclosure: we don't have live customers on this idea yet. We shipped the strategy package; you ship the customer conversations." So the hero says "Try Free (No Auth)" but the bottom of the page tells me there are no live customers and this is an idea being sold as a package for someone else to build and operate. The "Try Free" button is either a demo or a dead end. I don't know which because I didn't click it, but that contradiction is the kind of thing that makes me distrust everything above it retroactively. The "Three Steps to Full Clarity" are numbered 1 through 4, which is a small thing but when I'm already suspicious I notice small things. The whole Wishdeal Factory framing also confused me. I came here looking for a tool. I found a product idea marketplace. Those are different things. --- ## What would convince me A working demo against a real dataset, not a canned one. Let me point it at a public BigQuery dataset, like the Chicago taxi trips table or the GitHub archive, and show me what the output actually looks like. Not a screenshot. Not a markdown snippet. The actual export. If it can describe `trip_seconds` as something meaningful beyond "integer column, nullable," I'm interested. If it just echoes the column name back at me with "likely represents trip duration in seconds," I'll know it's a thin wrapper. On the relationship detection specifically: show me one example where it found a connection that wasn't obvious from column naming alone. One real example is worth more than the whole features section. --- ## What I'd ask in an email reply 1. The "Try Free (No Auth)" button in the hero: is there actually a working product I can point at a BigQuery project right now, or is this a mockup of what the product would look like once someone adopts the idea? 2. For relationship detection in BigQuery (where FK constraints don't exist at the schema level), what signals are you actually using? Column name matching? Value overlap sampling? Something else? 3. The pricing says "full source code access" across all plans. Does that mean I can self-host this entirely, or is it a local CLI that still phones home? --- ## Verdict: dismissive Not because the idea is bad. Auto-documenting BigQuery schemas is a real pain I have right now. But the page presents itself as a live product and then reveals at the bottom it isn't one. That bait-and-switch, even if unintentional, breaks the trust I need to hand over my GCP project credentials. If this were a real working tool with a real demo, I'd be testing it tonight. --- *Memo by skeptic persona, generated 2026-06-22. Studio breaks own self-grading loop.*