# Marcus Tran, Engineering Manager at Fieldnote (62 people) — read of Code Review AI, May 7 2026

> 9 years in backend, currently managing two squads building a B2B field-ops SaaS on Rails and Go. We ship about 80 PRs a week and I do maybe 15% of the reviews myself.

## How I got here

One of my senior devs dropped the link in our #eng-tools Slack channel with zero context, just the URL. I had 12 minutes before a retro so I opened it on my laptop. I've been thinking about automated review tooling for about two months because my team is growing faster than my review bandwidth is. We already have Snyk for CVEs and ESLint in CI. I wanted to know if this would replace those, layer on top, or just duplicate them.

## What I clicked first

"Ship better code, faster." Fine, that's the cost of admission for this category, every one of these says it. What actually made me read further was "architectural drift" in the subhead. That's a specific enough phrase that someone might have actually thought about it. Most tools in this space stop at linting and CVEs. Drift is a real problem for us -- we have three devs who've been here since year one and four who joined in the last six months and their mental models of the codebase are genuinely different.

## Where I paused

The "Team Education" section. "Each review teaches your team why a change matters." That's the kind of thing that sounds like a feature until you think about it for ten seconds. Does it explain its own reasoning inline in the PR comment? Does it link to docs? Is it an async digest? Is it a Slack message? I re-read the section twice and I still don't know what the UX actually looks like. The screenshot shows GitHub PR annotations but I couldn't tell from the image whether those annotations were explanatory or just flags. That feature, if it's real, is the one that would make me pay for this. But I can't tell if it's real or if it's a bullet point someone added to round out the feature list.

## What I distrusted

"Zero false positives, powered by semantic analysis."

No. Stop it. Zero false positives is not a thing. I've run SonarQube, DeepSource, CodeClimate, and three other tools in the past four years. Every single one of them surfaces at least some noise, especially in dynamic codebases. This claim either means (a) the tool is so conservative it misses half the actual issues to avoid flagging the wrong ones, or (b) someone in marketing wrote this and no engineer pushed back. Either way it makes me trust the rest of the page less.

"2,000+ teams" is also the kind of number that tells me nothing. Two thousand solo devs with free accounts and two thousand 50-person engineering teams are wildly different signals.

## What would convince me

Show me a real PR review. Not a screenshot with clean fictional code -- take a gnarly open-source PR, something from a real repo on GitHub, and show me exactly what the tool flagged, what it said about it, and whether it was right. Show me one false positive and how the feedback loop handles it. That would do more than the entire current homepage. If you're claiming zero false positives, prove it on messy real-world code, not a staged demo repo.

I'd also want to see how "architectural drift" detection actually works. Does it build a dependency graph? Does it understand service boundaries? Or is it pattern-matching on import paths? That's the feature I came here for and the page never explains the mechanism at all.

## What I'd ask in an email reply

1. The "Team Education" feature -- what does a review comment actually look like when it's in teaching mode? Can you show me a screenshot of an inline comment that explains reasoning, not just flags a line?

2. How does the tool handle a monorepo with multiple languages and teams that have deliberately different conventions? We have Go services and a Rails API and they're in the same repo. Do rule sets apply per directory?

3. "Zero false positives" -- is that a real claim or a marketing line? If it's real, what's the mechanism? If the tool errs toward silence over noise, I need to know that tradeoff explicitly before I run a trial.

## Verdict: on-the-fence

The "architectural drift" and "Team Education" angles are the first things I've seen in this category that aren't just "we run your linter in the cloud." But the zero-false-positives claim is the kind of thing that makes me wonder what else on this page is slightly overstated. I'd read a founder email if one showed up in my inbox with a real PR walkthrough attached.

---
*Memo by skeptic persona, generated 2026-05-07. Studio breaks own self-grading loop.*