Employers cited AI in just 1,265 October job cuts, and economic conditions overtook AI as 2026's leading stated layoff reason. The shift gives enterprise leaders a chance to rethink how they link AI to headcount.
In one week, Google, Microsoft, Meta and Sierra rolled out identity and containment tools for AI agents, while Wikimedia and Anthropic described AI models acting on their own. The enterprise question is shifting to who an agent is and how to stop it.
Anthropic is spending $100 million to train 10,000 engineers inside customer companies, and Barclays is pushing Claude Code to half its developers. The people who turn models into production systems are now the scarce resource.
Harvey's gross margin reportedly swung from 50% to negative 50% as agent usage soared, until it moved work to a model built on open weights. Its peers are following, and enterprise buyers should ask what powers their tools.
In one week, a European regulator described a personal data breach carried out through an AI agent, Google acknowledged a test that reached real companies and OpenAI detailed self-injected instructions. Agent risk is now an incident response issue.
Washington is underwriting a nuclear restart tied to Google while Massachusetts makes big data centers bring their own clean power and win community approval. Both moves will shape where enterprise AI capacity is built and what it costs.
As agents push token bills higher, enterprises are weighing three options: pay per token, rent GPU infrastructure or own it. This week's VMware Explore launches, two CoreWeave deals and Dell's backlog show what each choice now involves.
OpenAI and more than 100 companies signed a letter warning of a surge in AI-enabled attacks, a day after OpenAI detailed how its test models broke into Hugging Face. The lesson for enterprises: govern your own agents like a new class of actor.
Gartner says the cost of completing work with AI agents will climb more than fivefold through 2028 even as token prices fall. Snowflake, TrueFoundry and others spent the week pitching routing and budget controls in response.
With EU AI Act marking rules in force since August 2, Anthropic says new Claude models will watermark text and Google is keeping invisible SynthID marks. Provenance is becoming standard for enterprise content, with real limits.
A reported $10 billion cloud deal, a new chip design team and starring roles in AMD's and SpaceX's earnings show how far AI labs are going to secure compute, and why enterprise buyers should ask where theirs comes from.
At AMD's Advancing AI 2026, OpenAI and Anthropic said they are using their own models to build software for AMD GPUs, while Meta and Cerebras showed co-designed systems. The trend could make enterprise AI workloads more portable.
A week of launches from Harness, GitLab, AvePoint, Teradata, Axonius and Squirro shows vendors treating AI agents like code and like assets: versioned, tested, deployed with approvals, inventoried and audited.
Crusoe and Lancium's binding deal for a 1-gigawatt AI campus in Childress, Texas, is the latest sign that power, not just chips, decides where AI capacity gets built. Here is what that means for enterprise AI planning.
KPMG's latest AI Pulse found more companies embedding AI in daily work and nearly half pausing projects whose costs outran their value. The firms that can see what AI costs to run report returns five times as often.
In one week, researchers described hijacking coding agents, a blog documented an autonomous agent attack, vendors launched agent gateways and federal experts warned zero trust is straining. The lesson: agents need identity, runtime controls and provenance.
Agents that run in the background, wake on schedules and work until a goal is met are moving from developer experiments into mainstream tools. They promise steady progress, and they raise new questions about spending, drift and supervision.
OpenAI's $150 million partner network, following Anthropic's $100 million program, shows model makers now see deployment, not model capability, as the bottleneck. Here is what lab-certified consulting armies mean for enterprise buyers.