# Marcus Webb, Lead TypeScript Engineer at Fendrel (28-person B2B SaaS) — read of Code Morph, May 31 2026

> 9 years writing TypeScript, last 3 running a 5-person frontend team, currently drowning in onboarding two junior devs who keep reinventing our error-handling wheel.

## How I got here

Somebody in the TypeScript Discord server I lurk posted a screenshot of the generated `createOrder` code and said "this is actually not embarrassing." That's a low bar but I clicked. I was already in a browser tab comparing GitHub Copilot Workspace and Cursor's background agents so the timing was right. I did not watch the 30-second explainer. I never watch the 30-second explainer.

## What I clicked first

The hero copy landed okay: "Write pseudocode. Get production TypeScript instantly." Fine. But the line that actually held me was "Learns from your codebase to write in your exact coding style, patterns, and conventions." I've heard that before from three tools that then produced snake_case variables in a camelCase codebase. What kept me reading was the inline example, specifically the note at the bottom: "Notice the error handling, logging, type annotations, and null checks? Code Morph learned those from your codebase." I wanted to see whether the code actually reflected that or whether it was generic boilerplate labeled as learned.

The `createOrder` example is the page doing its best work. That's a real-feeling function. The zod parse, the 401 vs 402 split, the `logger.error` call with structured context, the 201 on success. Whoever wrote that demo knows what a real API route looks like. I don't know if the tool actually outputs that or if someone hand-crafted it for the page, but the example passes a smell test that most AI tool pages don't.

## Where I paused

Bottom of the page. There's a section that reads: "Honest disclosure: we don't have live customers on this idea yet. We shipped the strategy package; you ship the customer conversations." And there's a Fermi estimate showing "$-20,990 Year-1 take-home." On the product homepage. That stopped me cold. Is this a product or is this a business-idea marketplace selling build packages? Reading more carefully, it looks like "Wishdeal Studio" built a concept and is selling the rights to operate it for $99-$199. So I'm not buying a SaaS. I'm buying a starter kit for a SaaS that doesn't exist yet. That's a fundamentally different thing and it's buried.

## What I distrusted

"Your code never leaves your account" is doing a lot of work with no support. Who hosts the RAG index? Where does the codebase context get vectorized? What's the data processing agreement? For a 28-person company with a SOC 2 in progress, "your code never leaves your account" on a marketing page is not a data security story.

Also: $29 per month, unlimited translations, all team members. That's pricing for a student project, not a production RAG pipeline with per-token inference costs. Either the unit economics don't work or there's a usage cap buried in the fine print that isn't disclosed here. I'd bet on the latter.

The "61/100 Adoptability" badge with "uniqueness: 5/10" is a strange thing to put on a product page. You are telling me your own tool rates itself mediocre on differentiation. I appreciate the transparency in theory but it mostly just reminded me that GitHub Copilot, Cursor, and Codeium all do style-matching in some form now.

## What would convince me

One real team, not anonymized, saying specifically what patterns it got right that their previous tool got wrong. Not "it learned our error handling" but "we use a custom `AppError` class that wraps Zod issues and it started using it correctly after 12 accepted translations without us ever telling it to." That's the kind of evidence that would separate this from generic autocomplete.

Also a clear technical answer to: does the codebase context go to OpenAI or Anthropic or do you run your own embeddings? If it's going to a third-party LLM, what's in the system prompt and who can see it?

## What I'd ask in an email reply

1. The page says "Each accepted translation becomes context for the next one" -- is that context stored per-repo, per-user, or globally? And what's the latency difference between a cold first translation and a warm hundredth one?

2. The `createOrder` example uses `req.userId` as if it's already been attached by middleware. Did the model infer that from a real codebase or was that hand-written for the demo? I need to know if you're showing me actual output or aspirational output.

3. You're selling this as a $99 build kit, not a live product. If I buy the "adopt" tier and then build this myself, am I competing against you or are you walking away from the idea?

## Verdict: on-the-fence

The code examples are the strongest I've seen for a tool in this category and the pain is real. But the page confused me about what I'm actually buying for the first three scrolls, and "we don't have live customers yet" is not a footnote -- it's the headline risk, and it's not treated like one.

---
*Memo by skeptic persona, generated 2026-05-31. Studio breaks own self-grading loop.*