For years, code review looked essentially the same: scan the diff, leave comments on questionable lines, hope the developer on the other end understood your intent. Rinse, repeat. But somewhere between the tenth consecutive AI-generated pull request and the mounting backlog of unreviewed code, that familiar workflow started to crack.
Enter Assert Labs, a San Francisco outfit founded by a pair of infrastructure engineers who spent late May rolling out what amounts to a visual aid for overwhelmed reviewers. The pitch is deceptively straightforward—render the user interface changes before and after a code modification, right there alongside the traditional diff. No context-switching to staging environments. No guessing whether that CSS tweak will actually center the button or shove it halfway off the screen.
The timing feels deliberate, perhaps more so than the founders initially expected. As AI coding assistants churn out code at a clip that would make any human developer blanch, the bottleneck has shifted from writing to reviewing. Assert's founders, Devin Plumb and Cal Lavicka—veterans of Unify and Bubble, respectively—are betting that the solution isn't more automation on top of automation. It's giving human reviewers sharper tools.
Seeing Is Believing, Apparently
The core idea landed on May 29: inline previews that surface before-and-after renders of UI changes, positioned directly next to the code that caused them. Reviewers can drop comments onto the visual output itself, not just the lines of code buried in some CSS file three directories deep.
Assert calls this "evidence-based review," which is consultant-speak for a fairly intuitive concept—let engineers see what actually changes on screen without firing up a local build. For teams wading through AI-generated front-end tweaks, the visual confirmation carries weight. A diff might pass muster at first glance; the rendered result might reveal something else entirely.
By late June, the platform had expanded to handle rich file previews—SVG, HTML, Markdown, JSON—with slider, blend, and side-by-side comparison modes that feel borrowed from photo editing software. The company also introduced what it calls "merge candidates" on June 29, a feature that resolves rebase conflicts in a sandbox and shows reviewers only the delta relative to the proposed merge. Useful, if your team is drowning in merge hell.
Reviewable Units and the Ghost in the Machine
Beyond the visual layer, Assert chops pull requests into "reviewable units"—compact, dependency-aware chunks meant to impose some reading order on sprawling, multi-file changes. The goal is to sidestep the wall-of-text problem that makes large PRs feel like archaeology projects.
Then there's the feature that feels most attuned to the current moment: agent traces. For code produced by AI agents, Assert surfaces the prompts and sessions that generated each change. It's a breadcrumb trail for reviewers trying to understand intent when the original "author" wasn't a person with a brain and coffee habits, but a bot following instructions.
The platform also includes re-review mode, which highlights only the lines altered since the last review pass; move detection that links refactored code across files; and comment rebasing that keeps inline discussions intact even after force-pushes. Standard table stakes for any tool claiming to modernize review workflows, though the execution seems competent enough.
The Backers Talk Louder Than the Funding Round
Plumb, the CEO, was a founding engineer at Unify after studying computer science at Princeton. Lavicka, the CTO, came from a staff role at Bubble with a Carnegie Mellon pedigree. Solid credentials, but hardly unprecedented in a city full of Stanford and MIT dropouts building developer tools.
What's more telling is the angel roster: Guillermo Rauch of Vercel, Thomas Wolf of Hugging Face, David Cramer of Sentry, Charlie Marsh of Astral. These aren't passive check-writers—they're infrastructure and open-source operators with decades of collective credibility. Their involvement signals something, even if the company hasn't disclosed funding amounts publicly. (Startups rarely shout about small seed rounds unless they're startlingly large or come from marquee VCs.)
LinkedIn lists the team at somewhere between two and ten employees as of mid-2026, which tracks for a product still in its early innings. The company runs a public status page and maintains a GitHub integration that posts review links to pull requests and syncs all Assert activity back to the mother ship.
A Suddenly Crowded Room
Assert didn't exactly stumble into a greenfield opportunity. The market for code review tooling has gotten thick—perhaps uncomfortably so—in recent months.
Anthropic shipped Claude Code Review in early March, explicitly framing it as a triage tool for "the flood of AI-generated code." Linear followed with its own Diffs feature in late May, landing just a day before Assert's debut. Vercel Labs open-sourced OpenReview. Qodo, another entrant in the verification space, raised $70 million in March. Then there's the alphabet soup of other players: CodeRabbit, What The Diff, Diffen, Ito, Diffy—each offering some variation on AI-assisted review, automated fixes, or runtime behavior checks.
Assert's angle is narrower, and arguably more conservative. It doesn't auto-review your code. It doesn't auto-fix bugs. It gives human reviewers better context—especially visual context—so they can move faster without shipping product-level disasters.
The company's mission framing is pointed, if not particularly original: code review hasn't fundamentally evolved in over a decade, but the volume and velocity of code certainly have. With AI assistants accelerating development, review tooling needs to catch up. Whether Assert's particular blend of visual previews and dependency-aware chunking represents that catch-up moment is another question entirely.
The Bet on Human Judgment
Assert doesn't publish pricing, which suggests either early-stage flexibility or a strategy still being workshopped. The product integrates with GitHub via an installable app or personal access token. Documentation, recently updated, covers webhook-based updates and pull request comment workflows. A Trust Center link exists, though it remains gated—common enough for B2B tools still building out compliance infrastructure.
For now, the bet is straightforward: show reviewers the UI, not just the diff, and trust them to decide what ships. It's a vote of confidence in human judgment at a moment when much of the industry seems intent on automating it away.
Whether that confidence pays off depends less on Assert's feature set and more on whether engineering teams believe visual previews solve a problem worth paying for. The answer, as always, is in the pull requests.
