Alisa Rae keeps citing the same number: 95%. That's the share of user experience bugs she believes companies never hear about. The form that mysteriously rejects certain email formats. The checkout flow that traps users in an endless redirect loop. The dashboard that implodes when opened on an iPhone.
They're happening right now, on your site, and unless someone manually scrubs through hours of session replay footage—which, let's be honest, nobody does—they'll keep happening.
Rae thinks she's found a different approach. Her company, Lucent, emerged from Y Combinator's Winter 2026 batch with a straightforward proposition: let AI watch the replays humans don't have time for. Connect your PostHog account, grant read-only access, and Lucent's models will start hunting for the friction you're missing.
The public launch came in March, right as the broader session replay market started filling with similar promises.
Reading Between the Clicks
Lucent doesn't own the infrastructure that captures user sessions. That's deliberate. Rather than rebuild what PostHog, FullStory, or Hotjar already do well, the product operates as a layer on top—initially focused on PostHog, with expansion plans for other platforms still in the works.
Once plugged in, Lucent's AI begins its vigil. Rage clicks, unexpected crashes, error messages that pop and vanish, pathways that dead-end without explanation. When the system detects something worth investigating, it doesn't simply flag the anomaly. It compiles reproduction steps, identifies which users hit the issue, and delivers the package wherever engineering teams actually spend their days: Slack channels, email threads, or ticketed straight into Linear.
During the launch window, Rae offered prospective customers a free diagnostic pass—the first 300 PostHog sessions, analyzed at no cost. A try-before-you-buy gesture that doubles as proof of concept.
She shared one demo that spring: a form validation bug rejecting perfectly valid email addresses. The kind of silent conversion killer that persists until someone with enough patience manually reviews session after session, looking for the pattern that explains why users keep abandoning at the same step.
These are the bugs that hurt most because they never announce themselves.
Two Acquisitions Before 25
Rae brings previous exits to the table, which matters in a market where session replay startups are multiplying fast. Before Lucent, she built Stella AI—acquired in 2024. Earlier still, she was a founding engineer at MagicBrief, the ad creative platform that Canva bought in June 2025.
(Australian media sometimes refers to her as Alisa Wu; same person, different contexts.)
She closed Lucent's pre-seed last October—$1.3 million from a collection that includes Horizon's Sandy Kory, Joshua Browder's Browder Capital, Long Journey Ventures, Weekend Fund's Vedika Jain, and Firestreak Ventures. Australian outlets pegged the round at $2 million AUD and called it oversubscribed. At the time, according to coverage from that period, Rae was 22 and operating as a solo founder.
Traction and the Pivot
By March, Lucent had processed north of 45,000 session replays in a single week, per Rae's own reporting. That volume suggests early adoption, though the company hasn't disclosed formal user counts or retention metrics.
The current product represents a course correction. Earlier iterations of Lucent explored browser-agent training data infrastructure—something more foundational, perhaps, but harder to sell. The shift to devtools and QA automation appears to have resonated better. One early tester, commenting on a LinkedIn demo Rae posted, wrote: "I thought Den (YC X25) had 0 bugs till I used Lucent."
Third-party industry roundups have suggested 30-plus YC companies adopted the tool, though Lucent hasn't confirmed those figures officially. Startups talking to other startups, in other words—the usual early-stage playbook.
Two Minutes to Integration, Then Continuous Surveillance
Setup is deliberately minimal. Generate a read-only PostHog API key, connect it, and step back. No instrumentation changes, no new JavaScript snippets to audit and deploy. The AI operates continuously after that, analyzing sessions as they stream in rather than requiring batch reviews at week's end.
The workflow integration is what makes the pitch credible. Bugs don't accumulate in some separate Lucent dashboard that engineering teams have to remember to check. They surface in Slack, where alerts already flow. Or they arrive pre-formatted in Linear, with context attached and reproduction steps included.
The promise: eliminate the hours currently spent manually scrubbing through session replays. Or more accurately, eliminate the backlog of replays nobody watches because there aren't enough hours in the week.
Crowded Terrain
Lucent didn't invent this category—it walked into one already taking shape.
PostHog itself launched "PostHog AI" features. FullStory, Hotjar, and LogRocket are layering their own intelligence onto replay data. Other YC companies have similar offerings: Decipher AI uses vision language models to parse replays, while Prism pitches its system as "AI that watches session replays and tells developers what to fix."
The market is fragmenting before it's fully formed.
Lucent's angle is depth over breadth—focusing on automated detection and workflow integration rather than trying to own the capture layer. That creates dependency risk. Right now, the company lives or dies with PostHog's API access and feature set. Expansion to other platforms will determine whether the model holds at scale.
But the underlying dynamic is hard to dispute. Session replay became standard practice faster than companies built the capacity to analyze what they were capturing. The data piles up. AI seems like the obvious answer, which is why multiple startups are converging on the same insight simultaneously.
Whether they can coexist, or whether the incumbents simply absorb this functionality into their core products, is the open question.
What Happens Next
Lucent presented at YC's Demo Day on March 24. Third-party sources have referenced pricing in the range of $1,200 monthly for 50,000 sessions, though the company's website doesn't publish rates publicly. The marketing copy calls it "AI-Native Session Replay for Modern Web Apps" and emphasizes automatic issue detection—"flags problems for you."
The bet, ultimately, is that the gap between what users experience and what developers know about remains wide enough to justify another tool in the stack. That teams using PostHog want automated bug detection badly enough to pay for it separately. And that 95% of unreported bugs represent a genuine market opportunity, not just a compelling statistic.
Rae's track record suggests she knows how to build products people will buy. The session replay market's momentum suggests demand is real. Whether Lucent captures enough of it before the incumbents close the window—that's the part that hasn't been written yet.
For now, the bugs keep happening. The question is who finally starts catching them.
