AI Got Cheaper. Your AI Bill Is About to Triple.
The Signal: Anthropic cut Opus 4.8 to 3x cheaper per token this week. The same week, Uber publicly burned through its entire 2026 AI budget in four months. Microsoft reports its own AI agents now cost more than the humans they were supposed to replace. Per-token is the lie. Per-workflow is the truth. The coaches who priced offers assuming AI stays cheap are about to feel the floor move.
You opened a new Claude window this morning.
Drafted a sales email in 90 seconds. Pulled three insights from a client transcript. Spun up a content outline for next week.
You felt like a wizard.
Then your monthly Anthropic bill landed.
$340.
Last month was $90.
"But everyone keeps saying it's getting cheaper..."
Yes. Per token.
And that is the entire trap.
Two Stories That Ran the Same Week
Almost nobody in the coaching world connected these two headlines.
Story one. Anthropic shipped Opus 4.8 on May 28. Fast mode 2.5x faster. Pricing dropped 3x per token. Subagents in parallel. Headlines screamed "AI just got 3x cheaper."
Story two. Fortune published the internal Microsoft data on May 22. Running their AI agents at scale now costs more than paying the human employees the agents were supposed to replace. Microsoft started quietly pulling back on its "use AI on everything" mandate.
And then Uber.
Uber's CTO disclosed they burned through their entire 2026 AI coding budget by April. Their engineers were spending $500 to $2,000 a month each on Claude Code and Cursor API calls. Their COO Andrew Macdonald called it a "head-exploding moment" on Fortune. His exact quote: the link between their Claude Code use and consumer value "is not there yet."
Read that middle number again.
Goldman Sachs published a forecast this month. Agentic AI workflows use up to 24x more tokens than traditional chat sessions. Tom's Hardware ran a tighter number. For some agent loops, the multiplier is 1000x.
The price per token went down.
The tokens per task went up faster.
Net bill? Going up. Hard.
What This Actually Means for Your Coaching Business
Here's the trap I'm watching coaches walk into in real time.
You added AI to your workflow when it was cheap. You priced your offer when AI was a $20 a month curiosity. Your margins quietly assume AI stays in that lane.
It will not.
You are not paying per token. You are paying per workflow. And in 2026, every "AI feature" you ship is an agent workflow, not a chat session.
Let me give you the math your bookkeeping is hiding.
| Feature you might ship | 2025 cost | 2026 cost | Change |
|---|---|---|---|
| Brand voice AI (per client per session) | $0.04 | $0.18 | +350% |
| Weekly insights brief (per client) | $0.12 | $0.95 | +690% |
| "Ask my coach AI" loop (per question) | $0.03 | $0.22 | +630% |
| 40 clients, three uses per week | $19/mo | $155/mo | +715% |
That "scalable" feature you were going to ship next week? It costs $155 a month before a single client clicks anything. At a hundred clients it's $387.
And those numbers are based on per-token pricing dropping the way Anthropic just dropped Opus.
If you priced your $2,000 cohort assuming AI was free, you're not running a coaching business. You're running an unfunded AI startup with extra steps.
Per token is what the model company shows you. Per workflow is what your bookkeeping shows you. The gap between those two numbers is where coaching margins go to die in 2026.
The Four Moves That Keep Your Margin Alive
Here's what the coaches who survive this reprice are doing differently. Right now.
1. Model the token cost into the offer before you ship.
Don't price a $2,000 cohort assuming AI is a freebie. Open a spreadsheet. Sum the agentic workflows you plan to give clients. Multiply by 4x for safety. That's your floor. Anything below it, you cut or you reprice.
2. Cap inference per user. On purpose.
Token budgets. Daily quotas. Hard limits in the product. Not because it feels elegant. Because Uber didn't and now their COO is on Fortune asking if AI is worth it. You don't get to skip that lesson by being smaller.
3. Choose the cheap model for the boring work.
Haiku for sorting and classifying. Sonnet for drafting and short reasoning. Opus only for the work that genuinely needs the genius. Most coaches default Opus on every task. That's a 90% margin leak hiding in plain sight.
4. Price the ROI loop, not the AI access.
Your client isn't buying tokens. They're buying the outcome. The 30 leads. The shipped offer. The first $10K month. If you can't charge 10x the token cost of that outcome, the offer isn't ready for an AI feature yet. Sell the human version, build the AI version on the back, ship it when the unit economics work.
That last bar is the one to sit with.
84 percent of coaching offers I look at are priced as if AI is still $20 a month. They are not.
The Real Question
Cheaper AI sounds like a coach gift.
It's actually a sorting mechanism.
The coaches who priced their offers when AI was a curiosity, ran agentic loops without caps, and defaulted Opus on every task...
They're about to feel the floor move under them in the next two quarters.
The coaches who built ROI loops, capped usage, and chose the right model for the right job...
Best year of their life.
Same week. Same news. Two completely different outcomes.
Your Move
Open your bookkeeping app right now.
Sum your AI subscriptions and API costs for the last 90 days. Be honest. Include the Claude. The ChatGPT Pro. The Cursor. The Granola. The little Replicate bills you forgot about.
Now compare that to the actual revenue those tools moved this quarter.
If the ratio doesn't make you smile, you have 60 days to fix it.
And if you have no idea where to start, that's not a tool problem.
That's a clarity problem.
Ready to build a coaching business that survives the reprice?
Book a free Brand OS session. We'll map your offer, your funnel, and your AI stack so the numbers actually work in 2026.
Book Your Free Call →