The Product Desk

The Pricing Move

A developer opened Cursor on Tuesday morning and ran the same Claude-powered loop she runs every day. It cost her nothing extra, because her $200 Claude subscription was effectively subsidizing her API usage through the third-party harness. Anthropic announced this week that every Claude subscription now includes API credits equal to the plan's dollar amount ($200 plan = $200 API credits). Pitched as generous. For the cohort running Claude through Cursor, Cline, Zed, and OpenCode at effective 70-90% discounts to API pricing, it is a price increase of roughly an order of magnitude. Starting June 15, third-party tool usage moves to a separate credit pool. Once burned, API rates apply.

The era of subsidized AI inference through integrations is ending. The 50% rate limit increase for two months is a grace period, not a permanent concession.

The Competitive Response

OpenAI replied inside hours. Sam Altman offered two months of free Codex to enterprise customers who switch within 30 days. That is displacement pricing, timed to the exact moment of developer frustration. The Ramp data explains the urgency: Anthropic at 34.4% of business customers versus OpenAI at 32.3%, the first time Anthropic has led. OpenAI lost the business adoption lead and is fighting to reclaim it with a shot clock.

Why This Happened Now

Anthropic hired a CFO and is likely targeting an October 2026 IPO. The prior model, where power users captured large implicit subsidies, does not produce the revenue-per-user metrics public markets reward. Revenue moved from $9B to $30B+ ARR in roughly four months. At a $900B valuation, they need $40-50B annually. Product teams should model at least one more pricing adjustment before October as the S-1 narrative tightens.

The 2x2 for This Sprint

The teams that wrote a one-page memo defining "what price change would make us reverse course" will move in 72 hours. Teams without that memo will spend the quarter in a Slack thread while the window closes.

The ServiceNow Canary

ServiceNow's CDIO Kellie Romack watched her team's full-year Anthropic budget get consumed before mid-2026. She cannot say which users drove it, or which workloads, because Anthropic does not ship the telemetry. PagerDuty and National Life Group describe the same posture. The signal is not that AI is expensive. The signal is that AI costs are structurally unpredictable and the model providers have not built the instrumentation a buyer actually needs to run the feature in production.

A feature whose success creates unpredictable costs is not a shipped feature. It is a finance conversation waiting to happen.

Three Failure Patterns Converging

1. Tokenmaxxing: Amazon mandated AI usage and staff gamed the consumption leaderboards. Duolingo mandated AI across all roles, got 20% unusable output, and reversed the policy. Tokens consumed and sessions opened measure whether a user pokes the feature. They do not measure whether the feature produced value.
2. Power-user concentration: A team ships an AI feature without per-user cost visibility. Three months later they discover a handful of power users are consuming most of the budget. Then finance calls.
3. Data readiness gap: Only 15% of organizations have the data foundation for agentic AI. Companies are spending millions anyway. Nearly half cite data quality as the primary blocker. What customers actually do is buy, fail to activate, churn, and blame the vendor.

What Instrumentation Looks Like

ServiceNow built AI Control Tower internally and staffed a dedicated person to watch consumption, then turned around and sold the tool to its own customers. The category gap is the product opportunity: per-customer, per-feature inference cost telemetry does not exist in most production stacks. The metrics that show whether the surface is working are time-to-first-useful-output and week-four retention on the AI surface. Token volume tells you nothing about either.

The Pricing Model Test

The only stable cell on the pricing 2x2 is variable cost matched to variable pricing. Fixed cost per seat against variable inference is a bet that usage will not grow, which is a strange bet to place on a feature the roadmap deck tells the board is working. Palantir's value-based pricing, charging against cost savings delivered, is becoming the template for AI deployment companies. OpenAI's DeployCo is hiring Palantir alumni to execute it.

The Convergence Signal

Three of the five largest enterprise software companies shipped agent-callable workflow architectures in the same week. They converged on the same execution layer: headless workflows exposed over MCP. SAP put up a €100M partner fund and a Knowledge Graph that hands business context to agents. ServiceNow separated workflow logic from UI via Action Fabric and exposed it for any third-party agent to call. Notion shipped a full developer platform with agent tool building and plans to host Claude and Codex as "teammates."

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

What Changed for Your Product

A procurement manager at a Fortune 500 opened three enterprise software 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 did not have an answer. The third did, and moved to the next stage. The window before that question lands in RFPs is 2-3 quarters.

The Work Is Smaller Than the Deck Suggests

For most teams the build is a week of scoping and 2-4 weeks of implementation, assuming the underlying API is not already a mess. The sprint decision is whether to ship an MCP server against the existing API this quarter. The second decision, whether the product's core UI should be restructured around the assumption that an agent is the primary first-touch user, is a roadmap question and does not need to be answered this month.

The Diagnostic

Separately, Glean benchmarked raw MCP against enterprise knowledge graphs and reported raw MCP used 30% more tokens and was preferred 2.5x less on agentic tasks. The intelligence layer above MCP, not MCP alone, is where differentiation lives. Exposing endpoints is table stakes. Curating the context those endpoints return is the actual product.

The Case Study

Elena Verna — former head of growth at Amplitude, Miro, Dropbox, and SurveyMonkey — pushed Lovable's enterprise pricing page to production alone last week. No PM writing requirements. No designer on mocks. No engineer on the build. In a traditional org, that page needed all three people and a week of calendar coordination to land. Verna says she spent 90% of her time building, with almost no meetings.

Lovable employs zero product managers. The company is growing fast enough that the absence reads as the operating model, not an oversight.

What's Actually Being Unbundled

The PM role decomposes into four jobs: user research, prioritization, spec-writing, and cross-functional coordination. Lovable's bet is that the first three collapse into the engineer or operator the moment they talk to users directly, and the fourth disappears when the org is small enough to coordinate in one shared channel. Ravi Mehta's framing is the cleanest version of this: AI makes one person average-to-good at everything at once, which is enough when the alternative was coordinating three specialists.

The PMs who survive this shift look less like project managers and more like mini-GMs who happen to prototype and iterate directly.

The Parallel in Regulated Verticals

Abridge (clinical AI, $5.3B valuation, 250+ health systems) staffs 'clinician scientists' — MDs who are also full-stack engineers. Their release cadence moved from quarterly to monthly, a 4-6x acceleration, not by stripping coordination but by putting domain judgment in the same seat as the code. The pattern ports cleanly. The highest-velocity teams put the person who understands the problem in the chair where the artifact gets built.

The Diagnostic for Monday

Two questions, in order:

1. Of the work shipped last quarter, how much of the PM contribution was judgment about what to build versus coordination of the people building it?
2. If the coordination half went to zero tomorrow, does the judgment half still justify the role?

If yes to the second, the job gets better. If no, the job gets done by someone shaped like Verna. The companies that ungate information access — the blocker Verna names — pull disproportionate talent density. The ones protecting management layers end up with coordinators and no builders.