The Product Desk

The Arbitrage Is Closing

A developer opened her Cursor billing page this week and saw the same number she has seen for months. Next month she will see a number roughly ten times larger. Anthropic announced that every Claude subscription now includes API credits equal to the plan's dollar amount — the $200 plan gets $200 in credits. The pitch is generosity. What it actually does, for the large cohort using Claude through third-party harnesses like Cursor, Cline, OpenCode, and Aider at effective 70-90% discounts to direct API pricing, is end the implicit subsidy. Starting June 15, third-party tool usage gets a separate credit pool. Once that pool is burned, the meter runs at full API rates.

The timing is not a coincidence. Anthropic hired a CFO and is targeting an October 2026 IPO. Power users absorbing enormous implicit discounts do not produce the revenue-per-user numbers public investors want to see on the S-1. Plan for at least one more pricing adjustment before that filing.

OpenAI's Displacement Counter

Sam Altman responded within hours with 2 months of free Codex for enterprise customers who switch within 30 days. This is displacement pricing aimed at the exact week developer frustration peaks. The Ramp data showing Anthropic at 34.4% versus OpenAI's 32.3% — the first time Anthropic has led in business adoption — explains why the counter shipped on the same news cycle.

The interesting question is not whether Codex is as good as Claude for a given workflow. It's what the real switching cost looks like once the harness discount is gone on one side and two months of runway is on the table on the other.

The Decision Framework

Two axes this sprint. First: is the Claude usage load-bearing for a specific production workflow, or exploratory? Second: is the harness replaceable with Anthropic-native tooling at similar quality, or not?

• Load-bearing and replaceable: renegotiate with Anthropic inside the 30-day window while the leverage is real.
• Load-bearing and not replaceable: pilot Codex on the 2-month-free offer this week, not next month.
• Exploratory in either cell: stop paying metered rates for exploration. Move to whichever vendor is currently subsidizing it.

Sources Disagree on Duration

One read frames this as IPO preparation, which implies pricing stabilizes after October 2026. Another reads the same facts and points out that Anthropic's ARR grew from $9B to $30B in roughly 4 months and customers are absorbing price increases without churning, which implies the pricing power outlasts any IPO calendar. The conservative plan models at least one more adjustment before October regardless.

The Budget Blow Nobody Saw Coming

Kellie Romack, ServiceNow's CDIO, opened her finance dashboard one morning and saw her team's full-year Anthropic budget get consumed before the middle of 2026. She cannot tell you which users drove it, 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's Nimesh Mehta called Anthropic 'great for consumer usage but not great for companies.'

The signal here is not that AI is expensive. The signal is that AI costs are structurally unpredictable, and model providers have not built the instrumentation enterprise customers need to govern them.

Tokenmaxxing: The Goodhart's Law Problem

Here is what teams tell themselves users do: adopt AI tools, get more productive, generate value. Here is what users actually do when usage becomes a metric. Amazon mandated 80%+ weekly AI tool usage and staff are gaming leaderboards. Duolingo's CEO publicly acknowledged their blanket 'evaluate all employees on AI usage' policy failed. AI content at scale produces approximately 20% unusable output, and mandating usage produced performative adoption. They reversed the policy.

The cost model is now the product risk. Token spend sat in a finance spreadsheet and got reviewed quarterly. That arrangement worked when AI features were pilots with capped traffic. It does not work when the feature is embedded in the workflow and usage scales with customer success.

The Missing Infrastructure Layer

ServiceNow built an AI Control Tower internally and staffed it with a dedicated person to watch Anthropic consumption. It now sells the tool to its own customers. That is the tell. Two product categories are being pulled into existence by this gap:

1. AI cost governance: usage monitoring and cost attribution has moved from nice-to-have to procurement blocker
2. Multi-model abstraction: strategic infrastructure the moment a provider can raise prices without SLAs or usage transparency

The Stable Cell

One forcing function resolves the sprint question. On one axis: is inference cost fixed per seat or variable per call. On the other: does the customer pay per seat or per outcome. The only stable cell is variable cost matched to variable pricing. Every other cell is a bet that usage will not grow, which is a strange bet to place on a feature the same deck tells the board is working.

SAP, ServiceNow, and Salesforce Converged on Headless Execution

SAP shipped a Knowledge Graph for agent context alongside a €100M partner fund for Autonomous Enterprise. ServiceNow's Action Fabric decoupled workflow logic from UI and exposed it to any third-party agent over MCP. Salesforce added native WhatsApp voice to Agentforce Contact Center. The shipping notes look unrelated. The architecture underneath is identical: headless, API-first execution addressable through MCP servers.

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

The Procurement Question Has Already Changed

A Fortune 500 procurement manager is now asking in demos: '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. That is the question RFP language will calcify around inside two to three quarters.

If your APIs are not agent-consumable, agents will route around you to a competitor whose APIs are. The window before this shows up in RFPs is two to three quarters.

What 59% of Token Volume Actually Says

Vercel's AI Gateway production data across 200,000+ teams shows 59% of all token volume is now agentic workloads. Anthropic captures 61% of AI spend, driven by Opus for heavy reasoning. Google captures 38% of token volume, driven by Flash for cheap, fast tasks. Most large teams route across multiple providers rather than committing to one. That is what buyers are doing, not what the vendor decks claim they will do.

The Forcing Function for Product Teams

The decision is narrower than 'build an agent strategy.' It is whether the workflows a product owns can be invoked by an agent that is not yours, without a human clicking through the UI, by Q4. If not, the agent living in the buyer's stack will route around the product to one whose surface is callable.

The honest scope is a week of design and two to four weeks of build for an MCP server against existing APIs — assuming the underlying API is not already a mess. That is the estimate from teams who have actually shipped it.

Lovable Has Zero Product Managers

Elena Verna — former head of growth at Amplitude, Miro, and Dropbox — took a pure IC seat at Lovable in December 2025. She now spends 90% of her time building, has almost no meetings, and personally shipped Lovable's enterprise pricing page to production. In a traditional org, that project needed a PM, a designer, engineers, and a week of calendar time. She did it in hours.

Lovable is not a thought experiment. It is growing fast enough that the missing PM function is not an oversight. It is the operating model. Engineers talk to users, write specs, ship code, and read feedback directly.

What's Actually Being Unbundled

The PM role decomposes into four jobs: user research, prioritization, spec-writing, and cross-functional coordination. AI tools collapse the first three into a single high-context operator when that operator talks to users directly. The fourth disappears when the org is small enough or flat enough that coordination lives in a shared channel.

AI doesn't make a PM world-class at design or engineering. It makes them average-to-good at everything at once. For a PM who already thinks across functions, that is an opening, but only if the time goes into shipping instead of coordinating other people shipping.

The Diagnostic

Two questions to carry into Monday:

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

If the second answer is yes, the job gets better. If no, the job gets done by someone like Verna.

The Structural Risk for Established Companies

Senior builders who can get autonomy and impact density at a flat org will leave to get it. The companies that ungate information access, the blocker Verna names explicitly, will pull disproportionate talent. The ones that protect management layers end up with coordinators and no builders. Not every PM role disappears. The roles that survive look less like project managers and more like mini-GMs who happen to prototype and iterate directly.

Caveat: this model works at Lovable's current size. It breaks at a different size, and nobody knows exactly where. Teams with regulatory requirements, complex multi-team dependencies, or enterprise sales motions still need coordination. The real question is whether that coordination is creating value or just filling a calendar.