# Marcus Tran, Engineering Lead at Fieldline (38-person B2B SaaS) — read of learns-yo, 2026-05-20 > 9 years writing TypeScript, the last 4 running a 6-engineer team that ships from Linear tickets and Notion specs every single sprint. I have a 3-year-old who starts preschool in September and I listen to podcasts during the 40-minute bus commute downtown. I used GitHub Copilot for a year before I got annoyed enough to stop. --- ## How I got here Someone in the TypeScript Discord posted a link saying "found this, looks interesting, anyone tried it?" I opened it during lunch. I was specifically looking for something that could stop the pattern where a junior engineer takes a perfectly clear spec, ignores our existing db client, and ships generic Prisma boilerplate that I then rewrite in review. That is a real pain. I clicked expecting a tool I could try this afternoon. --- ## What I clicked first The code block in the hero. That was the right call to put that front and center. Showing actual idiomatic output with aliased imports (`@/lib/db`, `@/lib/logger`, `@/errors`) and the real `db.users.findUnique` pattern instead of raw SQL or `fetch` calls -- that is the most honest thing on the page. I read it twice. If the product does exactly what that block implies, someone on my team would use it. The line "TypeScript that passes review on the first try" is doing a lot of work. I'm skeptical but I didn't close the tab. --- ## Where I paused The spec table. Specifically: "Context: Up to 200k tokens of repo context per translation." That is the number I needed to see and did not expect to find. Our services directory is not small. I actually stopped and did the math. Then right below it: "Privacy: BYO model key, self host the indexer, no training on your code." That's the answer to the question my CTO would ask before I could even finish the sentence. Those two facts together made me read the whole page instead of bouncing. --- ## What I distrusted Three things, in order of how badly they hit: First: "Wishdeal Factory listing, pre revenue, taking design partners." That is buried in a spec table like it's just another field. It is not just another field. That is the disclosure that this product does not exist yet in any customer-validated form. I had to re-read the page to understand what I was actually looking at. Am I a potential user? A potential founder who should build this? Both? The page does not decide. Second: the section called "How honest is this idea, really?" with Fermi scores and "Year-1 take-home: $-20,990" and "1 in 9 meaningful-success odds." That is not product copy. That is a pitch to an entrepreneur who wants to BUILD this product. That section is for a completely different person than the TypeScript engineer described in the hero. Halfway through this page I stopped being a potential user and became a prospective operator buying a $99 idea kit. Those are two different sales. Running them on one page breaks both. Third: the "Who this is for" box near the bottom says "Tutors with $500K+ revenue, training-program operators, certification course creators." That appears to be copy-pasted from a different idea in this same factory. It has nothing to do with TypeScript shops. That killed a lot of trust instantly. --- ## What would convince me I want to see a real pull request diff, not a hero code block. Show me the before-and-after on a non-trivial feature: some service that touches three internal helpers, has a conditional branch, and uses an internal error type. Show me the initial output, then show me what the indexer learned on the second similar file. That is the claim worth proving. I would also want a number like "after indexing a 40k-line codebase, output matched senior-engineer patterns on X% of translations without any manual edits." Even a rough one from a design partner would be better than nothing. The "30-day fix-it-or-stop-billing" guarantee is actually good. I almost dismissed it as a standard line but the framing is specific enough to be credible. --- ## What I'd ask in an email reply 1. The indexer -- does it run once and stay static, or does it re-index on each commit? Because our codebase drifts fast and a stale index is worse than no index. 2. When the output is wrong -- and it will be wrong sometimes -- what does the correction loop look like? Do I annotate the bad output and it learns from that, or do I just re-run and hope? 3. The "design partner" framing -- what does that actually mean in terms of my time commitment? I'm not looking for a co-founder relationship, I'm looking for a tool. If design partner means weekly calls and feedback surveys, that changes the math on whether I try this. --- ## Verdict: on-the-fence The core mechanic is exactly the right thing to build and the code block in the hero is the clearest demonstration of the value I've seen in this category. But the page is half aimed at me (the user) and half aimed at a founder who wants to buy the idea and build it themselves, and that confusion made me lose confidence halfway through. I would not reply today, but I bookmarked it and I would check back in 60 days to see if there are any design partner case studies posted. --- *Memo by skeptic persona, generated 2026-05-20. Studio breaks own self-grading loop.*