In the beginning of the year, everyone was rushing to build "Openclaw for normies / enterprise." In the normie category, Muse / Instinct / Grok Bot are clearly seeing some signs of life. In the enterprise category, Claude Tag / countless startups have not landed with buyers yet. There are both model and interface gaps for the enterprise agent: - On the model side, recent models have crossed the threshold for being useful as personal assistants, but the threshold for providing enterprise value (at least worth the token spend at scale) is higher. Employees are paid to do work, so offloading menial tasks isn't really worth as much. - Being proactive in an enterprise is harder than being proactive as a personal assistant because the bar is higher: there's usually a lot more context and existing efforts for enterprise needs than personal needs. - On the interface side, the current paradigm of connecting MCPs / custom connectors on behalf of either individuals or companies is the only scalable / secure solution that fits into the current shape of companies, but this is too inconvenient and requires more human coordination that most companies can achieve. - The most valuable use cases for the enterprise are mostly already covered by Claude Code / Codex / verticalized solutions. The bar is high for a new product. All this said, I've been blown away by the internal agent that we built for ourselves. We've solved a few of the model and interface problems and are making inroads on more. More benchmarks and rigorous details to come soon.