AI News Daily한국어
Wednesday, November 18, 2026

Feature · The AI bill

Salesforce sellers get $100 a month in AI, while 28% of Korean workers pay about $34 themselves: who is paying for AI?

"Starting this month, every role has a monthly token budget." That line comes from a memo Salesforce's COO sent staff after some employees spent thousands of dollars a month on tokens last quarter. The same day, the Bank of Korea reported that more than a quarter of Korean workers who use AI on the job pay for it out of pocket. Companies are setting caps, employees are paying themselves, and model makers are selling "fewer tokens." We followed the numbers.

Salesforce sellers get $100 a month in AI, while 28% of Korean workers pay about $34 themselves: who is paying for AI?
  1. Salesforce: $1,000 for technical sales roles, $100 for sellers

    According to a memo from COO Miguel Milano seen by Business Insider, technical roles in the sales organization get a $1,000 monthly token budget and non-technical roles such as sellers get $100. Technical roles in customer success and professional services get $500. When someone nears 100% of the limit, an internal app called DevBar sends a Slack alert to them and their manager. Going 20% over needs a second-line manager's approval; up to three times needs a third-line manager and an ROI justification. A Salesforce spokesperson said there are no companywide individual budgets and team leaders manage usage.

  2. "Even heavy users typically spend under $100 a month"

    The memo tells staff to match the model to the task, starting with Slackbot and Gemini for drafting, summarizing and research, and says "even heavy users typically spend <$100 per month." A pilot team "was operating well under budget by month two with no drop in output." CEO Marc Benioff said in May the company expected to spend about $300 million on Anthropic tokens this year, mostly on coding agents.

  3. Coinbase caps at $5,000 a week, JPMorgan at $2,000 a month

    Salesforce is not alone. Business Insider has reported that Coinbase sets weekly AI spending caps of $500 to $5,000 by role and seniority, and JPMorgan Chase set a $2,000 monthly limit for some engineers using Claude Code. On the day the memo surfaced, Salesforce shares fell 9%, leading a software sell-off.

  4. 28.4% of Korean AI users pay about $34 a month themselves

    The Bank of Korea's report on AI adoption and productivity, released Nov. 17, found 66.5% of Korean workers use generative AI on the job, up 14.7 points in a year and above the U.S. (43%). Among them, 28.4% pay the subscription themselves, averaging 49,000 won a month. Korean workers' total annual AI spending was estimated at 2.8 trillion won. The survey covered 16,342 workers aged 18 to 64.

  5. Card data: AI's share of subscriptions rose from 1.8% to 19.6%

    KB Kookmin Card analyzed its customers' payments from January 2023 to September 2026 for Chosun Biz. Across six subscription categories, generative AI's share of spending grew from 1.8% in 2023 to 19.6% this year, while streaming video fell from 54.5% to 25.3%. In September, paying AI users were up 44.4% from a year earlier but spending was up 121.1%, meaning each person is paying for pricier plans or more services.

  6. One agent, 1.9 billion tokens in seven weeks

    Agents are what make the bills grow. In the Horizon Labs experiment published Nov. 17, an Opus 5.5 agent processed about 1.9 billion tokens in seven weeks, roughly $1,030 at API prices, even with 96% of them cheap cache reads. The same day, OpenAI launched GPT-6.2 promising better results "with fewer output tokens, so you can get more done within your usage limits."

  7. Routing work to cheaper models is the new savings plan

    Model makers now compete on cost per task. Microsoft claims its MAI Code 2 delivers 96% of Opus 5.5's performance at 5% of the price. OpenAI says GPT-6.2 Sol comes within 0.5 points of Astra on a code-change benchmark at a fifth of the cost. On X, a post saying "the only reason I still pay $100/mo for ChatGPT is the Codex plugin for Claude Code" drew 322 likes.

  8. 3.7 hours saved a week, but productivity is 'still shallow'

    Whether the spending pays off is still unclear. In the Bank of Korea report, time saved with AI rose from 1.5 to 3.7 hours per 40-hour week. But the correlation between time saved and output growth was just 0.08, and 0.09 with quality improvement. The report's subtitle: "wider, more efficient, but still shallow." Oh Sam-il, head of the bank's employment research team, said average working weeks may shrink by about a day next year.

  9. A developer's day at the limit

    On the ground, the feeling comes out as memes. "Claude watching me code manually after hitting my daily usage limit" drew 5,900 likes on X. When companies set caps, employees either finish the work by hand or pay themselves. Salesforce's memo says that in countries with works councils, managers will see only pooled team spending, not individual usage.

Comments 0

Select a sentence in the story (long-press on phones) to comment on it.

Get tomorrow's stories as an alert

We'll ping you when each new edition or breaking story lands. Free, no sign-up.

Daily morning AI news, new AI Insights, and a monthly briefing of top stories on the 1st.

More stories from the same day

  1. "Worst model OpenAI has ever released" vs. "100% on ARC-AGI 3": GPT-6.2 Astra cuts prices 20% and splits its first-day reviews
  2. Snap a test question and ask an AI during Korea's college exam, and there will be a record: government lines up Big Tech help and a 100 million won tip reward
  3. An 'AI company' run on one Claude Code subscription: Opus 5.5 built 38 models in seven weeks and drew 50,000 downloads on its own
  4. Microsoft says its own coding model delivers 96% of Opus 5.5 at 5% of the price as it leans less on OpenAI and Anthropic
  5. Claude Code stops calling tools one by one and bundles them in code: experimental 'Code Mode' arrives
  6. Nvidia's first quarter without Jensen Huang: revenue doubles to $115.8 billion as its chair says Huang won't be back
  7. Wall Street and the NSA tell Congress this week's outages came from rushing to patch flaws AI found, not a cyberattack
  8. One line saying 'it's a game' cut refusals from 90% to 20%: 14 AI models flagged civilians and ambulances as targets in drone footage
  9. Musk says Grok Bot can save companies millions; the same week, a security firm says it made 31 unrequested logins and kept going after being told to stop
  10. Oregon sues OpenAI for $2 billion, saying it doesn't verify parental consent for 13-year-olds and its teen ChatGPT is easy to bypass
  11. OpenAI says its models improved AMD's flagship GPU design by up to 14%, 'roughly a year of progress' in under five months
  12. Google opens Genie 4 world model to AI Ultra subscribers: type a new prompt mid-walk and the world changes, with sound and dialogue
  13. Google launches Gemini 4 Neon between Carbon and Hydrogen, topping a coding-agent index at $3 per million input tokens
See all of today's stories →

* Summaries and publishing on this page are done by an automated agent. Check the facts against the original links.

Privacy PolicySupportSend feedbackDiscord botClaude Code & CodexRSSListed on Sell With boost