# Bento

> Bento is the closed-loop production platform for AI agents — OpenTelemetry-native observability, plain-English alerts, automatically authored skills and nudges, and regression-proof evaluation in one stack. Built by BentoLabs (YC P26).

This file is a curated index for AI agents. Every link below points to a clean Markdown version of the corresponding page. Read [/index.md](https://bentolabs.ai/index.md) first for product context, then the individual blog posts under `## Blog` for research and field notes.

## Pages

- [Home](https://bentolabs.ai/index.md): What Bento is, who it's for, and how the closed-loop platform works across monitoring, improvement, and evaluation.
- [About Us](https://bentolabs.ai/about.md): Who built Bento, the five-phase origin story from Emergent to YC, founders, vision, mission, and values.
- [Blog](https://bentolabs.ai/blog.md): Index of every published research note and field essay from the BentoLabs team.
- [Contact](https://bentolabs.ai/contact.md): How to reach the founders directly and the general inbox.
- [FAQs](https://bentolabs.ai/faqs.md): Platform, monitoring, improvement, evaluation, security, deployment, pricing, and onboarding answers.

## Blog

- [Why you can't vibe-code your way to a better production agent](https://bentolabs.ai/blog/vibe-coding-production-agent.md): Why Claude Code with max thinking is the right primitive for point-in-time work but the wrong tool for the population-scale, longitudinal problem of improving a production agent.
- [Production AI agents need research infrastructure, not iteration](https://bentolabs.ai/blog/production-ai-agents-need-research-infrastructure.md): Why production agents are time-varying distributions, not snapshots — and what improving one actually requires.
- [Why your AI agent stops following instructions mid-run](https://bentolabs.ai/blog/why-ai-agents-stop-following-instructions.md): How the attention U-curve causes instruction drift in long agent runs, and the zero-code fix for it.
- [Nature vs. Nurture in AI agents](https://bentolabs.ai/blog/nature-vs-nurture-in-ai-agents.md): A framework for diagnosing whether a production failure lives in the model's pretraining or in the harness around it.
- [2.6× higher scores on ARC-AGI-3 with a self-learning layer](https://bentolabs.ai/blog/self-learning-ai-agents-arc-agi.md): Same agent, same tools, same budget — 2.6× the score, 34% cheaper per successful outcome, three first-ever solves.

## Legal

- [Privacy Policy](https://bentolabs.ai/privacy.md): How BentoLabs AI collects, uses, and protects your information.
- [Terms of Service](https://bentolabs.ai/terms.md): The legal agreement governing your use of the Bento platform.

## Optional

- [Sitemap](https://bentolabs.ai/sitemap.xml): All indexable URLs on the site.
- [Book a demo](https://cal.com/team/bentolabs-ai/platform-demo): Schedule a platform walkthrough with the founders.