Blog

Contract review AI has shifted from copilot to autonomous agent. Thomson Reuters CoCounsel now processes 10,000 documents in agentic workflows, Spellbook redlines with 95% accuracy, and analysts predict zero-touch contracting for low-risk agreements by late 2026. But 700+ court cases already involve AI hallucinations, and the EU AI Act classifies legal AI decisions as borderline high-risk. This guide covers what works, which tools to pick, and what compliance demands.

Amazon’s AI shopping assistant Rufus reached more than 300 million users in 2025, generated $12 billion in incremental sales, and now autonomously buys products when prices drop. Customers who engage with Rufus convert at 60% higher rates. This is the first case study of agentic commerce working at scale, and it tells us more about the future of e-commerce than any protocol announcement.

ClawHub went from 2,857 skills to over 10,700 in a single month. Malicious skills grew from 341 to 1,184. The marketplace added zero new security controls during that period. This is not an attacker problem. It is a governance vacuum. AI agent skill registries are repeating the exact mistakes npm and PyPI made years ago, but with agents that have shell access, file system permissions, and credential store visibility. npm took seven years to build verified publishers, Sigstore provenance, and trusted publishing. AI skill registries have none of it. Here is what went wrong and what a properly governed registry would look like.

Cloudflare open-sourced Moltworker, an implementation that runs the viral Moltbot personal AI agent on its Developer Platform using Workers, Sandbox containers, and R2 storage. AI agent traffic on Cloudflare’s network doubled in January 2026, the stock jumped 5%, and CEO Matthew Prince declared the arrival of the ‘Agentic Internet.’ This post breaks down the Moltworker architecture, explains why self-hosted agents need edge infrastructure, and shows what it takes to deploy one yourself.

OpenAI shipped v0.8.2, v0.8.3, and v0.8.4 of its Agents SDK on three consecutive days in February 2026. The patches added Pydantic Field annotations, realtime model versioning, and a hosted container shell. But the real story is what this release cadence reveals about the framework race against LangGraph, CrewAI, and a stalling AutoGen.

Banking is not piloting AI agents anymore. Goldman Sachs, Lloyds, JPMorgan, and Deutsche Bank are running autonomous agents in production across trade accounting, compliance, credit scoring, and customer service. NVIDIA reports active AI usage in finance jumped to 65%, and Capgemini projects AI agents will deliver $450 billion in economic value by 2028. But 80% of firms are still stuck in pilot mode, and the EU AI Act classifies credit scoring as high-risk. Here is what actually works, what the numbers show, and what regulators demand.

A January 2026 class action accuses Eightfold AI of building secret candidate profiles from 1.5 billion data points without disclosure or consent. If plaintiffs win, every AI recruiting vendor that scrapes external data will need to rethink its entire product. Here is what happened, why it matters, and what employers should do now.

GitGuardian’s $50M Series C targets the fastest-growing attack surface in enterprise security: non-human identities. With machine identities outnumbering humans 144:1, 70% of leaked secrets still active after two years, and AI agents creating thousands of new credential-bearing entities, secrets management is becoming an existential enterprise concern.

The AI agents in healthcare market will grow from $1.11 billion to $6.92 billion by 2030. With 68% adoption (highest of any sector), 42% reductions in documentation time, and Google’s AMIE matching physicians on 29 of 32 clinical metrics, these systems are already in production. But the EU AI Act classifies most medical AI as high-risk, and ECRI named AI the number one health technology hazard. This guide covers what works, what fails, and what the regulations demand.

CrewAI now powers 2 billion agentic workflows across 60% of Fortune 500 companies. PwC reports 88% of teams plan budget increases for agentic AI. Yet Gartner says 40% of these projects will get canceled by 2027. Both data sets are correct, and they tell a more useful story when you read them together.