Blog

Rivet’s open-source Sandbox Agent SDK normalizes the APIs of Claude Code, Codex, OpenCode, Amp, and Cursor behind a single HTTP/SSE interface. A 15MB Rust binary runs inside any sandbox provider (E2B, Daytona, Docker, Vercel) and exposes standardized session management, event streaming, and permission handling. This post covers the architecture, supported agents, real-world deployment patterns, and how it compares to building direct integrations.

In February 2026, the US Department of Labor released its AI Literacy Framework. The same month, Stepstone reported 77% of German companies would prioritize skills-based recruiting. Two continents, two announcements, one conclusion: hiring criteria that worked in 2024 are obsolete. This post covers why skills-based assessment and AI literacy screening converged simultaneously, how the EU AI Act turned AI literacy from a preference into a legal obligation, and what practical assessment methods work for both criteria.

Your agents work inside your company. But your business does not. The agent web, a network where AI agents from different organizations discover, authenticate, and collaborate with each other, is being built right now with Google’s A2A protocol, Anthropic’s MCP, and a discovery mechanism borrowed straight from the early web. Here is what it looks like, who is building it, and why it matters more than any single agent you deploy.

Traditional IVR systems automate only 7-15% of customer interactions. Voice AI agents connected to live data hit 3-5x those rates while cutting per-call costs from $6-12 to under $0.50. A Metrigy study of 656 companies found 37.6% plan full IVR replacement, rising to 62.5% among top performers. Here is what the transition actually looks like, what it costs, and which platforms deliver.

MIT’s NANDA report found that 95% of AI pilots fail not because of bad models but because of a learning gap: organizations do not understand how to design workflows, empower the right people, or build feedback loops. This post digs into the MIT diagnosis, real-world agent catastrophes from Vectara’s awesome-agent-failures repo, the context engineering problem Inkeep identified, and the community response through standards like FAILURE.md.

The web’s security architecture took 30 years to build. Same-origin policy, Content Security Policy, CORS, sandboxed iframes. All of it assumes one thing: a browser sits between the user and untrusted content, mediating every interaction. AI agents bypass that assumption entirely. They fetch web content, treat it as instructions, and execute actions with the user’s full permissions. This is the confused deputy problem at a scale the web was never designed to handle.

Deloitte’s 17th annual Tech Trends report maps the enterprise agentic AI adoption funnel: 30% exploring, 38% piloting, 14% deployment-ready, 11% in production. The report’s CTO Bill Briggs warns that applying AI to broken processes ‘weaponizes inefficiency.’ This post breaks down the five strategic questions every enterprise needs to answer, why governance enables speed rather than blocking it, and what HPE, Toyota, and Dell did differently.

Gartner predicts an 8x jump in AI agent embedding across enterprise applications in a single year. But with 40% of agentic AI projects also headed for cancellation, the real question is which platforms are delivering and which are agent-washing.

Traditional search still drives 48.5% of website traffic, but the shift is accelerating. ChatGPT has 800 million weekly users, Google AI Overviews appear in 60% of U.S. searches, and AI visitors convert at 4.4x the rate of organic traffic. GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization) are two distinct approaches to earning visibility inside AI-generated responses. This post breaks down how they work, how they differ from SEO, and what you can do right now to show up when someone asks ChatGPT about your industry.

Coordinated AI fraud attacks rose 180% in 2025. Know Your Agent (KYA) is the emerging framework that binds every AI agent action to a verified human identity, closing the accountability gap that KYC and traditional IAM leave wide open. This post covers the Sumsub implementation, the four KYA components, and how the framework maps to EU AI Act compliance.