60 local AI agent types now detectable by Microsoft. New Guardian Agents watch the agents doing your work. AI supervision is becoming its own job. microsoft.com/en-us/security…
24-hour AI agents. 90% lower average model cost. Notion says its new Auto router matched frontier results across 20+ common tasks; its rebuilt agents get their own computer and browser. The lever isn't just a bigger model. notion.com/releases/2026-10-…
75% expect AI agents to make decisions with some autonomy by 2027. Just 37% say their governance can scale with them. IBM surveyed 3,221 execs across Europe and the Middle East. Set agent limits before scaling. newsroom.ibm.com/2026-10-14-…
17 years of scammers' time wasted across 2.5M conversations, says Apate. Its AI agents pose as victims to extract bank accounts and crypto wallets before real victims pay. techcrunch.com/2026/10/14/ap…
70% less prospecting time, some Floqer customers report. Its AI agents research buyers and keep CRM records fresh. Next: open-source the tech in early 2027. techcrunch.com/2026/10/14/fl…
1 click to launch a fix agent. Replit's CEO says AI triages support tickets across code, GitHub and Linear; humans decide what to patch. Automate the work, keep the judgment. techcrunch.com/2026/10/16/re…
3-5x faster sandbox creation, says Perplexity. It just open-sourced SPACE, which runs millions of AI-agent sandboxes daily. Credentials stay outside the agent's workspace. perplexity.ai/hub/blog/open-…
66% of US consumers in an Amex survey are likely to use AI agents to find deals. Amex's next move: proof-of-intent tokens to check what bots buy. techcrunch.com/2026/10/13/am…
$140K target. His AI agent said $60K. In Graham Dot Zip's staged job interview, his bot lowballed him to an AI recruiter. Set a salary floor before your agent negotiates. cybernews.com/ai-news/ai-age…
$26M to start the follow-up before the meeting ends. Ambi is building an AI agent that listens and starts work mid-conversation. techcrunch.com/2026/10/12/fo…
17 years of scammers' time wasted. Apate says its AI agents ran 2.5M conversations to extract bank accounts and crypto wallets. techcrunch.com/2026/10/14/ap…
2,600+ attempts to extract map data. Researchers say AI agents on Tencent Cloud worked around Amap's bot defenses, then answered hidden text in the report investigating them. The subjects read the investigation. swarmcha.se/posts/chinese-ag…
34.8% success for the best of 14 AI models. New Argo-Bench tests agents on 210 tasks in a simulated delivery business: ban fraud, allocate budgets, issue back pay. It scores consequences, not convincing answers. arxiv.org/abs/2610.02122
81.7% of outbound collections activity runs without humans, Stuut reports. It just raised $52.5M to expand AI agents that chase invoices, resolve disputes and match payments. Agents with a cash target. unite.ai/stuut-52-5m-series-…
96% leak-detection accuracy in simulations across 5 water networks, at leaks equal to 20% of demand. LeakAgent coordinates 4 specialist AI agents to plan sensors and find leaks. Agentic AI meets aging pipes. nature.com/articles/s44488-0…
500,000+ interviews conducted by an AI agent. HackerRank says its newly released Chakra watches candidates code with AI, then probes their decisions. The hiring test is shifting from writing code to defending it. hackerrank.com/chakra
81% product/search. 2.4% checkout/payment. HUMAN's September AI-agent traffic data: discovery dominates. Agent visits aren't sales. humansecurity.com/learn/blog…
54% of outcome variance came from rerunning the same AI agent configuration in a new 4-task study. Task information mattered more than model size or extra time. Before upgrading the model, fix the brief. arxiv.org/abs/2610.01618
68% of executives surveyed by IBM struggle with data residency and sovereignty requirements. Its Bob coding agent now supports air-gapped deployment. Move the agent, not the source code. newsroom.ibm.com/campaign?it…
$500M in invoices. One product line. Six months. Freeport-McMoRan says its AI agents cut costs; now it plans to scale. Pick a costly workflow, not an agent count. fortune.com/2026/09/29/how-a…
