The Week AI Stopped Feeling Free

Written By Chantal Emmanuel

AI just became a metered cost, an advertising channel, and your new front desk, and each one changes how you should run your business.


The Bill Came Due: What Runaway AI Costs Actually Look Like

Nicolas Brilliante did not set out to become a cautionary tale. He is head of strategic verticals at Slash, a fintech valued at $1.4 billion, and when leadership encouraged the team to build faster with AI, he spent a day prompting Claude to build a browser game. Here is the mechanic that matters for you: AI tools bill by the token, the small usage units a model processes, and you pay separately for what you send in and what it sends back. Every time Brilliante fed the AI his growing codebase to work on, the meter ran. By the end of the week the total was $81,267.

That is the visible version of a quiet trend. Uber rolled out AI coding tools to thousands of engineers and exhausted its entire annual budget in roughly four months. Uber, Coinbase, and Walmart have all set caps on how much employees can spend, with Walmart pointing specifically to repetitive, low-value usage. One AI startup, Lindy, moved all of its traffic off its premium model to a cheaper alternative and watched its cost curve, in the CEO's words, crash to the ground. On June 24, Gartner forecast that AI coding token costs will overtake the average developer's salary by 2028. The practitioner takeaway is not "use less AI." It is "govern it." Set account-level spending caps. Tier your tools so simple, repetitive tasks run on cheaper models and you reserve the premium ones for work that truly needs them. Assign an owner who reviews the bill monthly, the same way you would review any other operating line.

ChatGPT Becomes an Ad Platform: The Next Search Box Has Sponsored Slots

OpenAI said this week that advertising has become a core part of its business strategy, framing sponsored experiences around usefulness rather than traditional attention metrics. The company reported more than 900 million weekly active users, with roughly one fifth of queries expressing direct commercial intent, and said it will keep user conversations unavailable to advertisers.

Sit with that one-in-five number. It means a meaningful share of the people opening ChatGPT are effectively shopping. Today, when someone asks for a recommendation, the answer is generated organically. That is the same window small businesses had in the early days of Google, before paid results reshaped how customers discovered local services. The lesson from that era is simple: the businesses that learned to show up early, while it was still cheap or free, built an advantage that compounded. My read is that answer-engine visibility, whether your business surfaces when someone asks an AI assistant for help, is about to become as important as search ranking was fifteen years ago. The time to understand how you appear is now, while it is still organic.

AI Picks Up the Phone: Agentic Customer Service Goes Mainstream

A Salesforce survey of more than 3,000 service professionals found AI agent adoption rose from 39 to 66 percent over the past year, with 70 percent of organizations reporting measurable returns within 60 days. These agents now handle interactions across chat, email, messaging, and voice, while letting customers reach a human whenever they want, and organizations reported gains in satisfaction, resolution speed, and productivity, with many expanding training so staff can supervise the AI.

Treat that survey with a grain of salt, since the company selling the tools ran it, but the direction is real and worth acting on deliberately. The design detail that matters is the human handoff. The strongest deployments are not "AI does everything." They route the repeatable and low-stakes to automation and escalate the rest. Draw the line yourself. Hours, pricing, order status, and appointment changes are strong automation candidates. First complaints, sensitive situations, and high-value decisions are better left analog. And notice the quiet opportunity in that last survey point: the businesses getting this right are not cutting their people, they are retraining them to oversee the systems. That is the version of this shift worth building toward.

What This Means

Three unrelated headlines, one pattern. For two years, AI felt free and effectively infinite, so the smart move looked like "use more." This week the strings became visible. AI is a metered cost that can spike without warning. It is an advertising channel that will soon decide whether customers find you. And it is a front-line worker that touches your customer relationships directly. None of this is a reason to retreat. It is a reason to stop treating AI as a magic subscription that quietly renews and start treating it as infrastructure you budget, meter, and manage.

What Business Owners Should Actually Do

  1. Put a hard spending cap on every AI account before you expand usage, not after the bill arrives.

  2. Tier your tools. Run simple, repetitive tasks on cheaper models and reserve premium ones for work that genuinely needs them.

  3. Test your visibility. Ask ChatGPT the questions your customers would ask and see whether you come up. Start improving it while it is still organic.

  4. Draw your customer service line. Automate the repeatable, keep the emotional and high-stakes moments human.

  5. Give AI an owner and a budget. Review the spend monthly like any other operating cost, and retrain staff to supervise rather than compete with the tools.

BAMPT has free guides in its library to help you be more efficient with your token uses, as well as showing up better in chatbots.

Chantal Emmanuel is co-founder of BAMPT, where she builds AI automation systems for service businesses, and CTO of LimeLoop. This Week in AI is her weekly, hype-free read on what actually mattered and what to do about it.

SOURCES

Chantal Emmanuel
Previous

A look at AI imagery, one year apart.
Next

When to Respond, And When to Lead