Windows Puts Agents on Bugs
Plus: Northern Trust Cuts Onboarding Prep, JPMorgan Goes Beyond Coding
Good Morning, Agent Owners!
Microsoft is using AI to hunt Windows bugs. The ultimate test will be finding out why the printer only stops working when you're in a hurry.
Windows Puts Agents on Bugs

TL;DR: Microsoft says agentic scanners help find and validate potential Windows vulnerabilities. AI also assists engineers in identifying affected versions, checking related code, planning backports, and building security tests. Engineers review findings and retain responsibility for fixes. Microsoft has not disclosed how many vulnerabilities the agents independently found; its total monthly patch count also includes findings from other teams and researchers. Read more →
Northern Trust Cuts Onboarding Prep

TL;DR: Northern Trust’s onboarding agent converts years of institutional clients’ investment and transaction records into a standardized format. Analysts review its output, resolve exceptions, and validate the starting data. The bank reports data preparation fell from roughly two and a half weeks to three days, an 83% reduction measured in production cases. The figure covers preparation, not the entire onboarding process. Read more →
One Hacker, Two Weeks

TL;DR: Malbear Labs recovered more than 70 session transcripts from an exposed attacker workspace. Researchers say an agent conducted over two weeks of reconnaissance, exploitation, and credential theft against a European telecom provider and its hosted environment, with occasional human direction. The report names no victims; its list of 13 associated organizations and claimed healthcare-record exposure lack independent confirmation. Read more →
JPMorgan Goes Beyond Coding

TL;DR: JPMorganChase is extending agent use beyond engineers. Product teams turn requirements into engineering backlogs, infrastructure teams investigate alerts and root causes, and other teams assist risk assessments and program planning. Access expands in stages inside bank-controlled environments. Thousands of automated tests and approval gates check code changes, and agents cannot approve their own work. No quantified productivity gains were disclosed. Read more →
Specialist Waits Drop 30%

TL;DR: UTMB uses agents to review patient charts, prioritize referrals, and suggest care pathways across five specialties handling more than 2,000 referrals monthly. Staff decide what happens next. In the four longest-running specialties, the health system reports average waits for first visits fell from 93 to 65 days year over year. These preliminary figures do not isolate the agents’ contribution. Read more →
Veterans Test a Benefits Guide

TL;DR: VA employee and former Marine Gary Crull built an agent that helps veterans and families navigate benefits, regulations, and local services. It draws on roughly 1,000 questions and answers, federal regulations, and VA location information. Veterans, service organizations, and VA employees are field-testing it, including at Augusta’s medical center. Reported time savings remain anecdotal; a full public release has no date. Read more →
JAL Traces Flight Delays

TL;DR: Japan Airlines employees have created 741 agents as the company starts a wider November rollout. One workflow queries internal flight records to analyze delays and their effects on later flights; other uses include scheduling and document checks. The figures come from a conference presentation covered by GeekyNews. They count agents created, not active deployments, and show no measured reduction in delays. Read more →
After-Hours Emails Get Handled

TL;DR: Australian tiling business The Grout Guy uses agents to process PDF work orders, chase unpaid invoices, and handle cancellation or rescheduling emails after hours. Staff share one mobile app as the business expands nationally. The company reports quote conversion rose from 30% to 50%. The November report draws on earlier interviews; the figures are company-reported, not an independent evaluation. Read more →
Dinner Invites Take 30 Minutes

TL;DR: Merge’s dinner bot finds prospective guests, checks invitation history, prepares outreach from its CEO’s inbox, and maintains attendance records and seating suggestions. CEO Shensi Ding reviews the guest list and rewrites messages. The company says those planning tasks fell from dozens of staff hours to about 30 minutes per event over a three-month comparison. Hosting and on-site logistics remain human work. Read more →
ANZ Draws the Boundaries

TL;DR: ANZ is implementing a central control system to track AI users, models, agents, tools, costs, and decisions. CEO Nuno Matos says it will define what agents may do and stop activity outside those limits. The bank already has AI gateways and a governance council. It has not detailed the architecture, confirmed every capability is live, or published results from blocking incidents. Read more →
OpenAI Builds a Stop Button

TL;DR: OpenAI says its Control Room can halt research model output within 35 seconds of receiving a stop instruction. Authorization checks sit in the inference layer, so research code cannot simply ignore the order. Fail-closed protection covers frontier research workloads, with others following. This is an internal research safeguard, not a guarantee that every external action stops within the same window. Read more →
Banks Shut Out Customer Agents

TL;DR: In a Dallas Fed survey of 70 bankers, 11.4% said their institution or service provider had restricted customers’ AI-agent activity on online banking platforms. Another 22.9% planned restrictions within six months, while 27.1% did not know their institution’s position. Responses were collected October 27–November 4. These are regional survey responses, not a nationwide measure of banks or blocked transactions. Read more →
Most Agents Still Need Help

TL;DR: A survey of 513 managers at large U.S. companies found 71% reported agents in production. Among respondents at organizations using agents, only 30% said agents completed more than 60% of tasks without human intervention. The September survey was commissioned by Zapier and published in November. It measures managers’ reports, not observed task performance or adoption across all businesses. Read more →
Your agent cloud, everywhere — in your chat apps, browser, and tools. Always on.
- Add to Slack, Teams, Telegram or WhatsApp.
- With a clean, redesigned WebUI.
- Connect any of 100+ apps & sources in one click.
- GPT 6 Astra, DeepSeek v4.1 & more frontier AI models, built in at discounted rates.
Run your agent on MyClaw.ai in Seconds👇



