# Marcus Okafor, VP Engineering at Fieldpoint (42 people, Series A) — read of Documentation AI, May 12 2026

> 14 years in backend engineering, last 4 managing teams. Currently wrangling a monorepo with Python services, a TypeScript frontend, and a Rust CLI nobody wants to document.

## How I got here

Google. I typed "auto sync docs with github commits" because we just had a post-mortem where a new hire followed a setup guide that referenced an env variable we killed in January. I was in a Lyft back from the airport, had 20 minutes, and clicked the third result. Not a recommendation from anyone I trust. Just me and my frustration and a search bar.

## What I clicked first

The headline pulled me in: "Your code and documentation drift apart with each release." That is the exact sentence I would say to my team. I kept reading. The follow-up is good too: "Six months later, your README is a liability." I have said those exact words. So the top third of this page is earning my attention.

Then I hit "Engineering teams waste 40+ hours per quarter rewriting docs that should be automatic." No source. No "in a survey of X teams." Just a number floating there. I let it slide because it felt plausible, but I noticed it.

## Where I paused

The example API doc block. It is clean. Better than I expected. The error codes, the parameter table, the response shape. If that is genuinely what the output looks like on a real repo, I want to see it on my repo, not on a toy endpoint with "Jane Smith" and a fake UUID. That example is doing real work on this page and it should be pushed harder.

## What I distrusted

Two things, one small and one large.

Small: "AI learns your team's voice and style over time." Every documentation tool says this. Every AI writing tool says this. It means nothing until I see a before/after from a real codebase after 3 months of use. That line reads like filler.

Large: halfway down the page, the product breaks the fourth wall completely. There is a section titled "How honest is this idea, really?" with an Adoptability score of 69/100, a Fermi estimate of -$22,500 year-one take-home, and this line: "Honest disclosure: we don't have live customers on this idea yet. We shipped the strategy package; you ship the customer conversations."

So this is not a product. This is a business idea being sold as a blueprint. The $29/month pricing I was looking at is hypothetical. The "Start Free Trial" button is selling me on a concept, not software. I had to reread that section three times to confirm I was not misreading it. That is a significant trust problem. The top of the page reads like a live SaaS. The bottom admits it is an idea deck with a $5 unlock and a $99 code starter.

I am not mad about it. The transparency is kind of remarkable. But the page structure creates a genuine bait-and-switch feeling whether that is intended or not.

## What would convince me

A real repo. Not "Jane Smith" with a fake UUID. Show me you ran this on Rails or a FastAPI monorepo with 80 endpoints and this is what came out. Show me the diff between the AI-generated output and what a senior engineer would have written. Show me one line that was wrong and how the system handles that.

The "Stack-Aware" claim in particular needs proof. I have a mixed Python/Rust/TypeScript repo. Tell me what that analysis looks like. Not a bullet point. A screenshot or a short Loom.

## What I'd ask in an email reply

1. The page says "no live customers yet" but the pricing and feature list look complete. Is there a working demo I can point at a public repo right now, today?
2. How does it handle internal-only functions, proprietary logic, or code we explicitly do not want documented? "Lock sensitive sections" is mentioned but I want to know the mechanics.
3. The "Freshness Tracking" feature sounds useful. What does the health dashboard actually surface? Page-level staleness, or can it flag a specific function whose signature changed but whose docs did not?

## Verdict: on-the-fence

The pain is real and the page communicates it clearly for the first half. But the second half reveals this is an idea package, not software, and the hero copy does not prepare me for that at all. I would not reply today. If they shipped a working beta I could test against a real repo, I would reply within the hour.

---
*Memo by skeptic persona, generated 2026-05-12. Studio breaks own self-grading loop.*