Today's thread
Today's news runs in two directions. One is a change in how AI subscriptions add up. Anthropic has added development credits to Claude subscriptions, and OpenAI has tied Codex usage limits to a daily improvement pledge. How much you can get for the same money is changing more often than the numbers on the price page.
The other is the rules of recommendation and advertising. Kakao now lets any member earn a commission through product referral links, and Korea's National Tax Service has banned unsupported '#1' claims in tax ads, including ads posted by agencies and platforms. The faster AI produces recommendation posts and ad copy, the more important it is to check first what you are allowed to say.
Claude Max adds monthly API credits — $100 on 5x, $200 on 20x
According to Anthropic's official documentation, Claude Max and Team plans now include monthly Claude API credits. Max 5x gets $100 a month and Max 20x gets $200, while Team plans pool $20 per Standard seat and $100 per Premium seat into one balance capped at $500. To claim them, you link an organization in Claude Console, the developer management screen, from the billing menu in Claude's web settings; no separate payment method is required.
An API is the doorway through which a program you build calls Claude. Until now, the subscription for chat and Claude Code and the fees for API usage were paid separately; now a small development budget comes bundled with the subscription. The credits, however, work only for the Claude API, Managed Agents, the Agent SDK and the playground, not for Claude Code or extra usage in the apps. Subscription usage limits stay the same.
The biggest practical change is that small experiments, such as an internal document summary bot or an email triage automation, can start without a separate budget approval. Credits arrive each billing cycle and unused amounts do not roll over, so it pays to pick one experiment to run every month. Giving each project its own workspace with a spend limit keeps a single experiment from draining the credits.
Free, Pro and Enterprise plans are not eligible, and new subscribers must wait 7 days. Each plan links to only one Console organization, and you cannot change the link yourself; you have to ask support. If you rely only on these credits and they run out, API requests stop until next month's credits arrive.
OpenAI Codex: 28 days of 'improve daily or reset limits'
Thibault Sottiaux, who leads Codex at OpenAI, posted a pledge on X in early October. Codex is a coding agent that edits and runs code on its own when given instructions. The pledge: for 28 days, from October 5 to November 1, ship one clear improvement for most Codex and ChatGPT Work users each day, or fully reset paid users' usage limits on any day that doesn't happen. A usage limit is the cap on how much work you can give the AI within a set period.
On day 1, output speed for the default models GPT-6 Astra and GPT-6.1 Sol rose from about 30 tokens per second to around 50. A token is the word-fragment unit in which AI reads and writes text. According to an OpenAI developer community thread that automatically collects the pledge updates, days 6, 8 and 10 were resets, and day 11 brought changes and added compute to reduce 'Selected model is at capacity' errors.
Frequent resets also signal how strong the complaints about limits running out and outages have been. Teams that use Codex for work are safer deciding in advance which backup tool to switch to and how to hand off work when limits run out, rather than counting on resets as a bonus. Following what changes each day also helps you pick features to try this month.
The original source for each day's result is Sottiaux's posts on X; they have not been collected on an official company announcement page. What comes after day 12, and what the limit policy will look like once the 28 days end, is not yet known.
KakaoTalk Gift 'Curator' — 9% commission on referral links
Kakao opened a 'Curator' feature in KakaoTalk Gift on October 13. Any Gift member can, without special qualifications, create a dedicated link for an eligible product and post it on a blog, YouTube, Instagram and elsewhere; when a purchase through that link meets the conditions, they earn a commission. To mark the launch, Kakao is running a promotion paying a total 9% commission through December.
This setup, where you introduce someone else's product and get a share of each sale, is called affiliate marketing. Until now, Gift was a service used for giving and receiving inside KakaoTalk; now recommendation posts on outside channels become a sales route. A dedicated curator screen shows recommendation history, purchase results and earnings, and Kakao plans to add tools for finding and sharing products in stages.
Solo business owners and marketers who run blogs or social accounts can attach one more revenue stream to gift recommendation content. For brands selling on Gift, recommendation posts by ordinary members become a new source of traffic, so it is worth adding to year-end gift season content plans.
The promotion runs on a limited budget, and the commission rate may change once the budget is used up. The specific conditions for earnings to count and the regular commission rate after the promotion have not been disclosed. Because these are paid recommendations, remember to label them as advertising.
Korea NTS notice in force — unsupported '#1 refund rate' ads banned
Korea's National Tax Service promulgated the 'Regulation on Advertisements Prohibited for Tax Accountants, etc.' (NTS Notice No. 2026-32) on October 12 and put it into effect immediately. It covers tax accountants, tax firms, and lawyers and certified public accountants registered for tax agency work. Banned are ads comparing fees with other tax accountants, ads promising money or economic benefits, and ads claiming 'best, only, lowest price' or '#1 refund rate' without an objective basis.
The Certified Tax Accountant Act gained a prohibited-advertising clause, and this notice sets out its detailed standards. Notably, the same standards apply to ads posted or sent by third parties such as ad agencies and online platforms. Tax accountants cannot escape on the grounds that they did not write the ad themselves.
Marketers who run ads for tax and accounting offices, or who mass-produce ad copy with AI, need a review step that filters out comparative and superlative claims. AI copy tools readily produce phrases like 'industry best', so a person should check after generation whether there is evidence behind them. Startups running tax platforms may also fall under the same standards.
The August draft notice required anyone advertising refund or tax-saving rates to keep supporting evidence, including sample size, statistical period and calculation method, for 3 years. Whether that condition survived in the final notice, and what penalties apply to violations, did not appear in the reports checked for this issue. The Korean Association of Certified Public Tax Accountants said it will step up monitoring of online tax platform ads together with the NTS.
Open Agent Hackathon 2026 opens online on 10/22
GenAI Works' Open Agent Hackathon 2026 runs fully online for 144 hours from October 22 to 27 (UTC). The theme is 'build an agent that does real work.' Teams have 1 to 5 members, there is no entry fee, and registration closes at 23:59 UTC on October 22, which is 8:59 a.m. on October 23 in Korea.
The prize pool is up to $20,000. The three main tracks (solving fragmented intelligence, the agent that can explain why, and reasoning architecture) carry $3,600 to $3,700 each, and the wildcard track carries $2,000. The best use of NVIDIA Nemotron wins a DGX Spark (a small desktop computer built for AI) worth $4,699, and the best product built with Cursor wins Cursor API credits worth $2,301.
As vibe coding, where you describe what you want and AI writes the code, becomes familiar, non-developers can also join a team and turn a work automation idea into a prototype within six days. Track themes that sit close to real work problems, like pulling scattered information together or explaining the reasons behind a decision, also lower the barrier to entry.
Because the schedule is in UTC, the submission deadline in Korea is 8:45 a.m. on October 28. Each track has its own conditions, such as the wildcard track requiring Zetaris and Meterless technology, so read the rules before registering.
Try this today
If you use Claude Max or Team, choose the Console organization to link in the API credits section of your billing settings. Once linked, you cannot change it yourself, so pick the organization you will actually build in.
Teams using Codex at work should decide on a backup tool for when limits run out, along with a handoff note template for picking up work in progress.
Search your company's or clients' ad copy for 'best, only, lowest price, #1' and mark whether supporting evidence exists. The same check is safer even outside the tax field.
If you have been writing gift recommendation posts on a blog or social media, pick the posts that could carry a KakaoTalk Gift curator link, and label them as paid recommendations.