The Product Desk

What Actually Changed

A developer opened Cursor on Monday, ran the same Claude-powered workflow she has run every week for six months, and got the same output. Her bill next month will be roughly ten times higher. Anthropic announced that every Claude subscription now includes API credits equal to the plan's dollar value. The $200 plan gets $200 in API credits. Pitched as generous. For teams using Claude through third-party harnesses (Cursor, Cline, OpenCode, Zed), it is a price increase of roughly an order of magnitude. The 70-90% implicit discount those tools quietly passed through is being collapsed. Starting June 15, third-party usage draws from a separate credit pool and overage bills at full API rates.

The era of subsidized AI inference through integrations is ending. If your product's AI capabilities route through third-party tools, model the cost impact now — not after the bill arrives.

Why It's Happening Now

Anthropic hired a CFO and is likely targeting an October 2026 IPO. The old model, where power users absorbed enormous implicit subsidies, does not produce the revenue-per-user numbers public market investors want to see. Expect at least one more pricing adjustment before October as the S-1 narrative tightens. The simultaneous 50% rate limit increase for two months is a grace period, not a concession.

OpenAI's Counter-Move

Sam Altman offered 2 months of free Codex to enterprise customers who switch within 30 days. That is displacement pricing, timed to Anthropic's moment of developer frustration. Ramp data showing Anthropic at 34.4% versus OpenAI's 32.3% explains the urgency. OpenAI lost the business adoption lead for the first time and is fighting to reclaim it before the switching window closes.

The Decision Framework

What Sources Disagree On

Multiple sources confirm Anthropic leads in business spend per Ramp, but diverge on stickiness. Some argue switching costs are approximately zero at the model layer, with Vercel data showing heavy multi-model routing already in production. Others note that teams with fine-tuned Claude models or Claude-specific prompt engineering face real migration costs measured in weeks, not days. Which story applies depends on your specific cell in the framework above. Read your own logs before you read the analyst note.

The Convergence Signal

Three of the five largest enterprise software companies shipped agent frameworks in the same week, and they converged on the same execution layer: headless workflows callable over MCP. ServiceNow's Action Fabric decouples workflow logic from UI and exposes it for any third-party agent to call. SAP shipped a Knowledge Graph for agent context plus a €100M partner fund to incentivize ecosystem development. Notion launched a full developer platform enabling code sync and agent tool building, with pre-built agents from Ramp, Clay, and Vercel.

Companies do not stand up hundred-million-euro funds for features. They stand them up for platform bets they intend to defend for years.

Production Data Confirms the Shift

Vercel's AI Gateway data across 200,000+ production teams shows 59% of all token volume now flows through agentic workloads. Anthropic captures 61% of spend (Opus for heavy reasoning), Google captures 38% of volume (Flash for cheap, fast tasks), and most large teams route across multiple providers. This is not experimental traffic — it is the primary consumption pattern.

What This Means for Your Product

A procurement manager at a Fortune 500 opened three demos this week and asked each vendor the same question: "Can our agents call this directly, or do my people have to click through your UI?" Two vendors had no answer. The third moved forward. The window before this shows up in RFPs is 2-3 quarters.

The Sizing Decision

Shipping an MCP server against an existing API is smaller than decks suggest — a week of scoping, 2-4 weeks of build assuming the underlying API is not already a mess. The larger question (restructuring the product around agents as primary users) is a roadmap question, not a sprint question.

The Every-App-Becomes-Platform Pattern

Notion, Cursor, Airtable (Hyperagent with $10M inference credits), and meeting tools are all launching developer platforms simultaneously. The convergence thesis — that every productivity app becomes an agent-hosting platform — is playing out in real time. Your competitive landscape is no longer bounded by product category. Your defensible moat needs to be upstream (proprietary data, unique user relationships) or downstream (specialized execution), not in the middle layer where everyone is converging.

The ServiceNow Wake-Up Call

ServiceNow's CDIO Kellie Romack watched her team's full-year Anthropic budget get consumed before the middle of 2026. She cannot answer which users drove the burn, or which workloads, because Anthropic does not ship the telemetry to answer those questions. PagerDuty and National Life Group describe the same problem. National Life Group's Nimesh Mehta calls Anthropic "great for consumer usage but not great for companies."

AI costs are structurally unpredictable and the model providers have not built the instrumentation customers need to govern them.

The 'Tokenmaxxing' Problem

A pattern now called tokenmaxxing shows up across multiple sources: employees running AI tools constantly without productivity gains, inflating the consumption metric the leadership deck is reading. Duolingo's CEO publicly acknowledged their blanket "evaluate all employees on AI usage" policy failed, producing ~20% unusable output at scale and performative adoption with no productivity to show for it. They reversed the policy. Amazon mandated 80%+ weekly AI tool usage and staff are gaming leaderboards. Goodhart's Law in production.

The Telemetry Gap Creates Two Product Categories

ServiceNow built AI Control Tower internally and now sells it to customers. The gap between what enterprises need (per-user, per-feature cost attribution) and what Anthropic provides (aggregate billing) is pulling a governance product category into existence. For teams building enterprise AI features:

• Per-customer, per-feature inference cost telemetry is no longer optional. It is a finance prerequisite for the next AI feature launch
• Usage governance features (rate limiting, admin dashboards, spend alerts) are becoming procurement blockers
• Multi-model abstraction stops being an engineering convenience and becomes strategic infrastructure the moment a provider can raise prices without SLAs

The Unit Economics Trap

Here is what teams tell themselves: we priced the feature with healthy margin over inference cost. Here is what users actually do: they find the workflow that saves two hours and run it 11 times a day instead of the 3 the pricing model assumed. Retention improves. Gross margin goes sideways, then down. The only stable pricing cell is variable cost matched to variable pricing. Everything else is a bet that usage will not grow, which is a strange bet on a feature the same deck is telling the board is working.

Why This Gets Worse Before Better

Anthropic can raise prices and customers absorb it. OpenAI committed $20B to Cerebras for compute. GPU demand runs at 4+ customers per GPU brought online. The common assumption that AI costs decline may be wrong for the next 4-6 quarters. Model AI feature costs with a 30% escalation factor. If that breaks unit economics, the choices are three: pass costs to customers, reduce inference through better caching and routing, or build multi-provider fallback before the next renewal cycle.