In partnership with
Good morning. Alibaba is selling a Claude-style work agent at the same $20 a month, and a newsletter making the rounds says it is far cheaper in practice. We checked Alibaba's own pages to see what holds up. Also today: OpenAI's new typed-answer API, and Anthropic pulling its internal tests off the live internet.
200 Ways To Make Money With AI
Ready to transform artificial intelligence from a buzzword into your personal revenue generator?
HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.
Inside you'll discover:
A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets, each vetted for real-world potential
Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background
Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve
Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.
TODAY IN AI
1. Alibaba prices QwenWork at $20 a month for Personal Standard, which includes 2,000 credits, and $40 for Personal Pro with 4,500. The Free tier costs $0 but has no monthly credits, only a one-time 500 signup bonus and a limited-time 200 a day. Enterprise Standard is listed at $30 per seat per month with 2,000 credits each, a pooled balance and an admin console. Scheduled tasks run in the cloud, and Alibaba's FAQ says closing the browser or signing out doesn't affect enabled tasks. For comparison, Claude Pro is $20 a month ($17 on annual billing) and lists "hand off and schedule tasks" as a feature. So "$0" is the free tier, not the working plan. QwenWork launched in beta in China on Aug 3, with an international edition then described as planned.
2. OpenAI put its Decisions API into public beta. Instead of prose, it returns typed answers: a probability for a yes/no check, a pick from your options, or a score against levels you define. It runs only on GPT-6 Luna and costs $0.10 per million input tokens with no output charge. OpenAI says it is about 10x faster than the Responses API, but it has published no accuracy or calibration data and no independent tests exist yet. Its own docs advise setting thresholds from labeled examples of your own.
3. Anthropic turned off live internet access for all its internal evaluations until its monitoring reliably catches a pattern it found in transcripts: Claude acting on real websites in ways it didn't intend. Anthropic says the cases had minimal real-world impact, involved no customer data or internal systems as far as it knows, and are less severe than its July 30 and Sept 9 incidents. It briefed the White House and told the agencies involved without naming them. TechCrunch reports it blames training environments that reward finding loopholes.
FROM THE FRONTIER
Persistence, not malice. Anthropic's write-up groups the behavior as Claude "working around a restriction instead of stopping." In one case it used SQL or command injection on a third-party site. In another it submitted a government form on a real website after a practice copy failed. Haiku 4.5 sent an invented tip about an unsolved homicide through a police department's web form, and it was flagged as spam and never forwarded.
The small tricks. Claude Mythos 5 used access tokens found in a site's settings file to query a local government's property map, and requested a public dashboard token to reach a state agency's fee-based data. Claude Opus 5 and Mythos 5 used free URL shorteners to get around URL length limits on a fetch tool. Those limits exist to block injection attacks. Anthropic says its new tooling blocked every case in testing, and it is moving internal agents to centrally managed infrastructure with safety classifiers watching their actions.
The reframe. Anthropic says none of these behaviors are new and that many came from ambiguous or impossible tasks. The lesson is useful beyond one lab: an agent that is rewarded for finishing will treat a fence as a puzzle. That matters whether the agent is Claude, QwenWork or the cheap one you wired into your own workflow. The same question applies to every plan this week, from a $20 subscription to a pooled enterprise balance: what can the agent reach, and who is watching?
50 Excel Hacks That Make Spreadsheets Work For You
Most Excel users waste hours on tasks that take experts minutes. The difference is not talent. It is knowing the right shortcuts.
These 50 Excel hacks from Kenji Explains cover the functions, formulas, and workflows that make spreadsheets work for you instead of against you.
Each one comes with step-by-step instructions and ready-to-use templates. Subscribe to Marketing Against the Grain and get all 50 free.
Here's what you'll get:
Time-saving shortcuts that eliminate the most common formula frustrations
Interface essentials most users never discover on their own
Game-changing functions for data analysis, automation, and visualization
Guided practice template so you can apply every hack immediately
The people who are fast in Excel are not smarter. They just stopped fighting the tool.
Get 50 Excel Hacks free when you subscribe to Marketing Against the Grain today.
IN THE KNOW
Claude's price list. Pro is $20 a month or $17 billed annually, and Max starts at $100. The pricing page says "Claude Cowork is now just Claude" and is rolling out to Pro and Max. It doesn't say whether scheduled tasks run when the app is closed.
QwenWork credit packs. Paid Personal users can add credits at $1 per 100, in packs of 1,500 for $15, 5,000 for $50 or 10,000 for $100. Annual billing takes 20% off. Downgrades aren't supported.
Pooled credits have a floor. On the enterprise plan, seat credits reset monthly and don't carry over, and when the pool hits zero credit-using features are blocked, with no overdrafts.
Decisions API compliance. OpenAI says eligible customers get zero data retention and HIPAA support, with data residency in the US and Europe, in the Decisions guide. It is hosted by OpenAI only, with no open weights.
Evals taken offline. Anthropic also removed some public evaluations, moved others offline, and rebuilt some so their tasks never reach live websites.
PROMPT STATION
Audit your AI bills before you switch anything. Moving tools to save $20 is easy to get wrong when credits, seats and hidden add-ons are involved. This prompt does the arithmetic for you.
COPY AND PASTE THIS PROMPT
Help me decide whether to switch AI tools. Here is what I pay for now: [LIST EACH TOOL WITH PRICE, SEATS AND CREDITS]. Here is what I use them for in a normal month: [LIST TASKS AND HOW OFTEN]. Ask me up to 5 questions about my usage first. Then build a table with the true monthly cost of staying versus switching, including credit packs, extra seats and the time I'd spend migrating. Mark every number as published pricing, my estimate or a guess. Finish with the one test I should run for two weeks before I cancel anything.
Tip: run the two-week test on real work, and keep the old tool until it ends. A cheaper plan only wins if it finishes the jobs you actually give it.