AI Giant Secretly Develops Rival To Google's Search Engine
By 813 Staff
A major product shift is underway — AI Giant Secretly Develops Rival To Google's Search Engine, according to Machina (@EXM7777) (in the last 24 hours).
Source: https://x.com/EXM7777/status/2031798550850654340
In a nondescript server room in San Francisco last week, a Perplexity AI engineer watched a prototype agent execute a 47-step task involving travel booking, expense documentation, and calendar management entirely autonomously. This quiet test, described by a person familiar with the run, is a glimpse into Project OpenClaw, the company’s ambitious and previously unreported initiative to build a generalized AI agent that can reliably operate software and complete complex, multi-application workflows. The effort, confirmed by internal documents and chatter among engineers close to the project, aims to move beyond today’s single-task automation bots to create an AI that can understand high-level goals, reason through the steps, and manipulate the digital interfaces humans use every day.
The project’s existence entered the public sphere via a cryptic post from the well-followed industry observer Machina (@EXM7777), who noted simply that “Perplexity [is] building an OpenClaw that works.” The name itself is telling, suggesting an open-ended tool for gripping and manipulating the digital world. Internal roadmaps frame OpenClaw not as a feature but as a new product pillar, positioning it as a logical but aggressive expansion from Perplexity’s core competency in answer engines into the fiercely competitive arena of AI agents. Engineers describe a system that combines advanced computer vision for understanding on-screen elements with a sophisticated reasoning model to navigate unpredictable software environments, from enterprise CRM platforms to consumer-grade creative tools.
The rollout has been anything but smooth, however. Early technical memos reveal significant hurdles in achieving consistent reliability, a problem that has plagued every major player in the agent space. The core challenge is failure recovery; when an action doesn’t produce the expected result, the agent must diagnose and correct course without human intervention. Sources indicate Perplexity is tackling this by generating massive, synthetic datasets of software interactions, including edge cases and error states, to train the models. The commercial implications are vast, targeting both enterprise clients looking to automate back-office processes and a broader user base that could delegate tedious digital chores. This moves Perplexity into direct, if unannounced, competition with projects from OpenAI, Google, and a slew of well-funded startups.
What happens next hinges on Perplexity’s ability to close the reliability gap. Industry insiders suggest a limited alpha release to select developers could occur within the next two quarters, though no public timeline has been set. The major uncertainty is whether Perplexity can achieve the necessary step-change in robustness to make OpenClaw more than a compelling demo. If they succeed, it would mark a fundamental shift for the company, transforming it from a source of answers into an entity that takes action. For now, the project remains a closely guarded bet, one that could either redefine productivity software or become another promising agent that couldn’t quite handle the chaos of the real world.
