The Product Desk

What Actually Changed

A developer on Cursor today is paying for Claude through what amounts to a wholesale relationship Anthropic no longer wants to honor. Starting June 15, Claude subscriptions split credit pools between first-party usage (Claude app, Claude Code) and third-party tool usage (Cursor, Cline, OpenCode, Zed, T3 Code). Third-party usage gets a separate credit pool equal to the plan's dollar value. Once depleted, the user lands on API rates. Teams that built their developer workflow on Claude through third-party harnesses, at an effective 70-90% implicit discount, just absorbed a price increase of roughly an order of magnitude.

This is not a random pricing adjustment. Anthropic hired a CFO and is likely targeting an October 2026 IPO. The previous shape, where power users got enormous implicit subsidies, does not produce the revenue-per-user numbers public markets pay for. Expect at least one more adjustment before the S-1 narrative tightens.

OpenAI's Counter-Move Has a Clock

Sam Altman offered 2 months of free Codex to enterprise customers who switch within 30 days. Displacement pricing, timed to developer frustration. Theo, Jeremy Howard, Matt Pocock, and Omar Sanseviero criticized the change within hours. 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.

The era of subsidized AI inference through third-party integrations is ending. Model the cost impact now — or discover it on your invoice in July.

ServiceNow Is the Warning Shot

What product teams tell themselves: enterprise AI cost is a budgeting problem. What ServiceNow's CDIO Kellie Romack actually did: burned through the entire full-year Anthropic budget before May ended, and could not tell which users drove it because Anthropic does not ship the telemetry. PagerDuty and National Life Group report the same problem. National Life's Nimesh Mehta calls Anthropic "great for consumer usage but not great for companies." The signal underneath: AI costs are structurally unpredictable and model providers have not built the instrumentation enterprises need to govern them.

The Decision Tree for This Sprint

Axis 1: Is the Claude usage load-bearing or exploratory?

Axis 2: Is the harness replaceable with Anthropic-native tooling at similar quality?

• Load-bearing + Replaceable: Renegotiate with Anthropic inside the 30-day window while the leverage is real
• Load-bearing + 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

Three Enterprise Giants Shipped Headless This Week

SAP, ServiceNow, and Salesforce all launched autonomous agent architectures in the same week, converging on the same execution pattern: headless workflows callable over MCP. SAP put up €100M in partner funding and a Knowledge Graph that hands business context to agents. ServiceNow's Action Fabric decoupled workflow logic from the UI and exposed it for any third-party agent to call. Companies do not stand up hundred-million-euro funds for features. They stand them up for platform bets they plan to defend for years.

Vercel's production data across 200,000+ teams shows 59% of all token volume is now agentic workloads. Notion launched a Developer Platform aimed explicitly at agent tooling, with pre-built agents from Ramp, Clay, and Vercel. The convergence thesis everyone has been writing decks about — every productivity app becomes an agent-hosting platform — is what the usage data is now showing.

A procurement manager spent forty minutes clicking through a vendor onboarding flow. Next quarter, an agent will. She will review the result. If your product can't be called headlessly, the agent routes around it to the system of record directly.

Where This Shows Up First: Renewals, Not Win Rates

Here is what users are actually doing. Two of a customer success lead's top ten accounts asked the same question on the same Tuesday: "Can our agents call your product directly, or do my people have to click through your UI?" The docs did not answer it. That conversation hits renewals before it hits pipeline. A Fortune 500 procurement lead opened three demos this week asking the same question. Two vendors had no answer. The third moved to the next stage.

The Build Decision

Separate the thing being pitched from the thing being done. Shipping an MCP server against an existing API is smaller than the deck suggests: a week of scoping, two to four weeks of build, assuming the underlying API is not already a mess. That is the easy part. The harder call is whether the product's core UI should be restructured around the assumption that an agent is the primary first-touch user for a non-trivial share of sessions. That is a roadmap question, not a sprint question.

Glean's Warning

Glean benchmarked off-the-shelf MCP against an enterprise knowledge graph. Raw MCP used 30% more tokens and was preferred 2.5x less on agentic tasks. The protocol is the table stake. The intelligence layer above it is where retention will come from.

The 2x2 for Your Backlog

The Proof Point: One Person Shipped What Used to Take a Team

Elena Verna opened a Lovable project last week and shipped Lovable's enterprise pricing page to production alone. No PM scoping the requirements. No designer on mocks. No engineer on build. The same work, six months ago, would have pulled three roles and a week of calendar Tetris. Verna ran growth at Amplitude, Miro, and Dropbox before this. She now spends 90% of her time building and almost none of it in meetings. Lovable has zero product managers.

What teams tell themselves is happening: a hobbyist learned Figma. What is actually happening: a senior operator who ran growth at four unicorns chose to drop the coordination layer entirely. Lovable is hiring Growth PMs alongside her, not under her. That is a collection of autonomous operators, not a growth team.

The PM value proposition decomposes into three pillars: cross-functional coordination, customer/market judgment, and strategic prioritization. AI-enabled flat orgs are eliminating pillar one. The question is whether pillars two and three justify the role on their own.

Duolingo's Counter-Evidence: AI Mandates Fail at Scale

Duolingo acknowledged publicly that their "evaluate every employee on AI usage" policy did not work. Two findings drove the reversal: AI content at scale produced ~20% unusable output that required human QC, and the mandate produced performative adoption with no productivity gain. Amazon staff are now gaming "MeshClaw token leaderboards." Same Goodhart pattern, different wrapper.

The lesson is not that AI doesn't work. It is that measuring AI inputs (tokens, sessions, logins) instead of outputs (tasks completed, quality, cycle time) produces theater. The PM who survives this transition is the one who defines what "done" looks like. Not the one who mandates which tool the team opens in the morning.

Autonomous Coding Changes the Bottleneck

Claude Code's new /goal command lets an engineer type a measurable end state and walk away. A separate Haiku model checks completion against explicit criteria. The unit of engineering output stops being story points per sprint and becomes review throughput per senior engineer per day. If half the backlog is pattern-execution work — migrations, test generation, refactors — the sprint is mis-sized by a factor nobody will name until the retro.

The 2x2 for Your Career This Quarter