There's a peculiar waste that happens inside successful software companies. Product teams dutifully record thousands of user sessions each week—every click, every hesitation, every abandoned shopping cart faithfully captured by session replay tools. And then almost no one watches them.
The recordings pile up like unread emails. Engineering managers occasionally scrub through a few clips when a frustrated customer escalates a complaint, but by that point the problem has already cost the company users, revenue, or both. Ask around and you'll hear the same sheepish admission: we capture everything, we watch almost nothing.
Lucent, a graduate of Y Combinator's W26 batch, thinks this is exactly the kind of drudgery AI was meant to eliminate. The startup has built a platform that connects to existing session replay tools and automatically watches every recording, flagging bugs and user experience breakdowns before they snowball into support nightmares.
Whether that turns into a viable business—or just gets absorbed into the very platforms Lucent integrates with—is the more interesting question.
The Pitch: Let the Machines Do the Watching
Here's how it works. Instead of expecting engineers to manually review hours of session recordings (they won't), Lucent's AI processes them in real time. When something breaks—a payment flow throwing a server error, an email address incorrectly flagged as invalid, a dead link dumping users onto a 404 page—the system clusters similar failures, identifies which users were affected, and generates reproduction steps.
Founder Alisa Rae says the platform analyzed over 45,000 session replays in a single week during early testing. That number hints at the scale of the problem: even a moderately successful SaaS product can generate thousands of sessions daily, burying critical issues in a mountain of data that no one has time to excavate.
The output goes beyond simple alerts. Lucent surfaces which specific users encountered the problem, provides context on what they were trying to accomplish, and can automatically generate tickets in Linear with debugging details already filled in. Notifications route through email and Slack, directing issues to whichever team is best positioned to fix them.
It's a straightforward value proposition, though not exactly a novel one.
Riding the PostHog Wave
For now, Lucent has focused its integration efforts on PostHog, the open-source analytics and session replay platform that's become something of a default choice among developer-centric startups. The company's launch offer reflects that focus: PostHog users can connect their account with read-only API permissions and get 300 sessions analyzed for free.
That narrow initial wedge makes strategic sense. PostHog has carved out meaningful territory in the startup ecosystem, and Lucent is positioning itself as a specialized layer on top—similar to how a handful of other emerging tools are building around PostHog's replay data. The marketing emphasizes "ready to integrate in minutes" setup, the kind of low-friction onboarding PostHog's audience expects.
Beyond PostHog, Lucent connects to Slack, Gmail, and Linear for notifications and workflow integration, though there's no public mention yet of support for other major replay platforms like FullStory or Hotjar.
A Category That's Getting Crowded Fast
Lucent has company. Several startups that passed through recent Y Combinator batches are chasing variations on the same thesis: that vision models and language models can finally "watch" user sessions at scale in ways humans never could.
Decipher AI, Prism, and PathPilot have all launched in the past year with similar promises—using AI to parse massive volumes of replays, generate semantic summaries, and surface problems. Meanwhile, established players like New Relic and Dynatrace have started weaving AI-assisted insights into their enterprise monitoring suites. PostHog itself continues expanding native AI features across its platform.
The technology clearly works—AI can analyze session replays. The harder question is whether a standalone tool adds enough value to justify another line item in the software budget, or whether this functionality simply gets absorbed by the replay platforms themselves. History suggests the latter happens more often than founders in this space would like.
The Things Users Never Report
Rae frames the core problem with a statistic that's difficult to verify but resonates with anyone who's run a product team: "94% of users don't report bugs." Most broken experiences just... disappear. A user encounters a 500 error during checkout, assumes the site is having issues, closes the tab. No ticket gets filed. No alert fires. Just quiet churn.
Traditional monitoring catches catastrophic failures well enough—servers crashing, error rates spiking, databases going offline. But it's remarkably bad at detecting what Lucent calls "silent breakages": edge cases, subtle friction points, bugs that only surface for specific user segments or particular flows.
That gap is what Lucent aims to fill. By observing actual user behavior instead of relying solely on instrumented error tracking, the system can theoretically catch problems that slip through conventional monitoring. Perhaps more importantly, it can catch them before they reach critical mass.
Rae previously worked as a software engineer at Atlassian and served as a founding engineer at MagicBrief prior to its acquisition by Canva. She raised a $1.3M pre-seed round in November 2025 from investors including Sandy Kory at Horizon, Joshua Browder from Browder Capital, and Vedika Jain at Weekend Fund.
What It'll Cost (Eventually)
The company hasn't published pricing yet—the website directs interested teams to book a demo call. For now, the free PostHog scan doubles as both a lead-generation tactic and proof-of-concept: let the AI watch a few hundred of your sessions, see what breaks, then decide whether the insights justify paying for it.
Whether Lucent establishes itself as an essential part of the modern DevOps toolkit or eventually gets folded into the platforms it integrates with remains uncertain. The outcome likely depends less on the technology itself—which appears sound—and more on how quickly PostHog and its competitors build comparable features in-house.
But the underlying insight feels solid: if your users won't tell you what's broken, and your team won't watch thousands of session replays, something else needs to. The question is just who ends up providing that something.
