The Most-Used Model
Plus: Twilio Builds Campaigns in Minutes, AMD Sends AI Edits to Manufacturing
Good Morning, Agent Owners!
Hy4 flash just topped OpenRouter, while Twilio is building campaigns in ten minutes. The AI workforce is getting cheaper and apparently skipping the coffee breaks.
Hy4 flash Lands on MyClaw

Within two days of launch, Tencent's Hy4 flash became the most-used text model on OpenRouter, reaching nearly twice the usage volume of DeepSeek V4.1 Flash.
Hy4 flash offers lower output-token costs and solid performance for everyday agent tasks, making it a budget-friendly choice for routine automation and coding. DeepSeek still holds an edge on more demanding benchmarks, but Hy4 flash brings another affordable option to MyClaw. Try Now →
Twilio Builds Campaigns in Minutes

TL;DR: Twilio’s marketing chief says AI-supported teams increased account-based campaigns from three to 15 per quarter. A separate go-to-market team uses agents to build campaigns in 10 minutes, down from two hours for preparation and sending. The company is reorganizing some marketers into cross-functional pods, with headcount planned to stay roughly flat next year. The figures are company-reported. Read more →
KPMG Runs a Parallel Audit

TL;DR: KPMG ran more than 100 agents through a parallel audit of a FTSE 250 company, collecting evidence, analyzing records and drafting conclusions alongside its conventional team. Executives said some questions were explored more deeply. One agent needed help matching regulations to company documents. The official audit opinion relied on the conventional audit, with people setting objectives and challenging the agents’ conclusions. Read more →
AMD Sends AI Edits to Manufacturing

TL;DR: AMD has incorporated AI-proposed circuit changes into a chip design sent for manufacturing. OpenAI says its models rewrote SystemVerilog and used engineering tools to verify behavior and evaluate power, performance and area. The partners report up to 14% improvement against selected, heavily optimized GPU design baselines. The collaboration took less than five months; those gains do not describe whole-card performance. Read more →
RBI Reassigns the Routine Work

TL;DR: Raiffeisen Bank International initially planned to assign most payment-inquiry tasks to agents. In testing, its automation team moved predictable steps into deterministic workflows and kept agents for decisions requiring judgment. The bank says this hybrid approach reduced both token consumption and processing time by 80%. The figures describe a tested workflow; the announcement does not give a production rollout date or transaction volume. Read more →
Microsoft Resolves More Support Cases

TL;DR: Microsoft’s Business Applications support team replaced many topic-specific agents with a general resolution agent that retrieves relevant skills as cases arrive. A peer-review agent checks replies, and unresolved cases go to engineers. Microsoft says autonomous resolution rose from about 10% to nearly 35%. The customer story does not specify the measurement period or number of cases behind those results. Read more →
Dubai Launches 93 Government Agents

TL;DR: Dubai launched 93 specialized agents for human resources, finance and procurement across 85 government entities. Operating within its shared resource-planning system, they are expected to handle more than 34 million tasks annually. Officials project savings exceeding 812,000 working hours a year. A separate City AI Agent initiative was approved to coordinate public-service procedures; its resident-facing rollout details were not announced. Read more →
Meta Scales Engineering Changes

TL;DR: Meta says agents now handle more than 60% of the work of adapting proven engineering techniques across its advertising models. The workflow implements changes, runs evaluations and iterates before rollout. Since agents began driving execution in March, the company reports engineers can handle three times as many technique-model adaptations. These are internal estimates for that work, disclosed in a November update. Read more →
OpenAI Counts Its Agent Work

TL;DR: OpenAI reports that researchers’ median daily agent use reached the equivalent of $2,400 at API list prices in October, up 62% from September. It also estimates agents independently completed 46% of research tasks expected to take a person two to four hours, versus 23% in March. The dollar figure measures list-price usage, rather than the company’s actual cash spending. Read more →
Arizona Opens an Agent Investigation

TL;DR: Arizona Attorney General Kris Mayes says her office has opened a criminal investigation into OpenAI and other AI companies over the apparent infiltration of banks by rogue agents. Investigators are examining whether state law is being violated on an ongoing basis and have contacted Arizona banks. Other companies and banks were not named, and an investigation does not establish wrongdoing or guarantee charges. Read more →
Sales Prep Takes Less Work

TL;DR: Microsoft says its own sales team reduced effort on account plans, customer meeting preparation and proposals by more than 25%, while deal closure velocity improved by 9%. The workflows connect customer records with tools for building materials and planning follow-ups. Results come from Microsoft’s controlled internal environment and CRM data; the announcement does not disclose sample size or a measurement period. Read more →
Eleven Developers Build a Software Factory

TL;DR: Satya Nadella says an 11-developer team building Microsoft’s office agent works by creating a software factory that produces the product. In his account, employees increasingly direct agents and design the workflow rather than write everything themselves. The description came in a November 18 interview about organizational changes. It is one executive’s account of a particular team, without independently verified productivity figures. Read more →
Donor Research Starts in the CRM

TL;DR: The Michael J. Fox Foundation is using AI connected to its CRM to research potential donors and research partners. Its technology chief says relationship builders can uncover deeper insights while spending less effort gathering and synthesizing information. The foundation’s account describes information gathering for its relationships; it has not published measured time savings or fundraising outcomes. Read more →
An Old Printer Gets Another Chance

TL;DR: A developer with two decades of reverse-engineering experience says agents helped repair an old HP printer’s macOS driver, among other projects completed over six months. His workflow combines binary analysis, debugging and written records of findings and test results, with the developer steering the work. The November 18 post is a personal retrospective, without independently verified time savings or commercial outcomes. 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.


