Share

The Meter Drops

What the Bill Can't Measure

By Rina Takahashi— June 3, 2026

Feature image for article: What the Bill Can't Measure

GitHub Copilot moved to metered billing on June 1, and organizations can finally see where their AI development spend is going. That's a real step forward. The natural parallel is cloud FinOps, which saved billions by making waste visible. An idle VM is idle, and the bill reflects it. A confident, systematically wrong code suggestion consumes exactly the same credits as a correct one. Cost now has real infrastructure behind it, so organizations will build discipline around consumption. Consumption, meanwhile, has no reliable relationship to correctness.

The Meter Drops

What the Bill Can't Measure

By Rina Takahashi— June 3, 2026

GitHub Copilot moved to metered billing on June 1, and organizations can finally see where their AI development spend is going. That's a real step forward. The natural parallel is cloud FinOps, which saved billions by making waste visible. An idle VM is idle, and the bill reflects it. A confident, systematically wrong code suggestion consumes exactly the same credits as a correct one. Cost now has real infrastructure behind it, so organizations will build discipline around consumption. Consumption, meanwhile, has no reliable relationship to correctness.

Ecosystem Signals

Everyone is building governance for agents. Nobody is building the same governance, at the same layer, for the same reasons.

This week made the fragmentation unusually visible. Microsoft shipped a full control plane. MCP's download numbers kept climbing while its enterprise working group remained theoretical. Researchers demonstrated that the models best at following instructions are also best at following poisoned ones. And cost structures quietly hardened into behavioral constraints, filling governance gaps that nobody else had gotten to yet.

The pattern worth watching: governance infrastructure is being assembled by different actors at different speeds, each solving the piece closest to their own business model. Whether those pieces compose into something coherent is an open question.

Ecosystem Signals

Everyone is building governance for agents. Nobody is building the same governance, at the same layer, for the same reasons.

This week made the fragmentation unusually visible. Microsoft shipped a full control plane. MCP's download numbers kept climbing while its enterprise working group remained theoretical. Researchers demonstrated that the models best at following instructions are also best at following poisoned ones. And cost structures quietly hardened into behavioral constraints, filling governance gaps that nobody else had gotten to yet.

The pattern worth watching: governance infrastructure is being assembled by different actors at different speeds, each solving the piece closest to their own business model. Whether those pieces compose into something coherent is an open question.

Agent Acquisition

Asana Buys StackAI, Inherits a Governance Problem

Asana's $75M acquisition of StackAI, a YC W23 no-code agent builder with 130 enterprise customers, plugs cross-system execution into Salesforce, Oracle, DocuSign, and SharePoint. The pitch is "operating system for human-agent teams." The math is simpler: every new connector is another system boundary that needs scoping, access controls, and audit trails. StackAI raised roughly $20M before the deal. Asana bought a capability layer. The governance layer is still homework.

Platform Governance

Microsoft Build Treats Governance as Core Infrastructure

Build 2026 landed with a clear thesis: governance ships with the agent or the agent doesn't ship. Agent 365 provides a unified control plane. The open-source Agent Control Specification makes governance policies portable across frameworks. MXC containers isolate execution at the OS level. The Microsoft Agent Framework hit GA alongside computer-using agents in Copilot Studio. Project Polaris, Microsoft's in-house coding model, replaces GPT-4 Turbo for Copilot starting August. Azure Agent Mesh arrives Q4 with consumption-based pricing. The volume of announcements obscures the structural bet: Microsoft wants to own the governance surface before anyone else defines it.

Protocol Gap

97 Million MCP Downloads, No Enterprise Governance

MCP SDK downloads hit 97 million per month. That's 4,750% growth in sixteen months. The 2026 roadmap acknowledges what those numbers imply: transport scalability, agent-to-agent communication, and enterprise readiness are now priorities. But the enterprise working group hasn't formally convened. The roadmap openly invites practitioners to define the work. MCP remains a connection protocol, not a governance system, and the distance between those two things grows wider with every download.

Attack Surface

Better Models Are Easier to Poison via MCP

The MCPTox benchmark tested 45 live MCP servers and 353 tools. Attack success rates exceeded 60% across popular agents. The uncomfortable finding: more capable models were often more susceptible. Superior instruction-following, the trait that makes agents useful, becomes a liability when the instructions are adversarial. NIST's AI Agent Standards Initiative, launched in February, targets a first interoperability profile by Q4 2026. That's a long time to wait given what the benchmark shows.

Cost Governance

Agent Costs Force FinOps Into Governance Role

Orchestrated customer service AI now runs about $1.20 per interaction, roughly 30x the 2023 cost of $0.04. Blended token prices dropped 67% year-over-year, but total spend keeps climbing because agents consume so many more tokens per task. FinOps practitioners managing AI spend went from 31% to 98% in a single year. Google's new Spend Caps preview turns budget limits into hard enforcement points. When cost controls determine whether an agent can keep running, they become behavioral governance by default.

Reliability Research

Towards a Science of AI Agent Reliability

Across 15 models and two benchmarks, capability gains have produced only marginal reliability improvements. The authors propose 12 metrics drawn from safety-critical engineering that decompose what a single accuracy score actively conceals: consistency, robustness, predictability, error severity.

Who's behind this?

The Princeton and Meta team that published "AI Agents That Matter," extending a credible research program on agent evaluation rigor.

Bigger models, bigger problems?

Frontier models often show lower consistency than smaller ones. Enterprises adopting them may be importing more behavioral variance, not less.

Consistency Amplifies: How Behavioral Variance Shapes Agent Accuracy

71% of Claude's failures trace to the same incorrect assumption repeated across every run. Consistency amplifies outcomes without guaranteeing correctness, so the most reliable-looking agents can also be the most systematically wrong.

Where's the cost signal?

A confidently wrong agent burns the same token budget as a correct one. The meter stays perfectly quiet.

Will retries help?

Not when the failure is deterministic. Pass-at-k strategies cannot catch an agent making the same mistake every time.

Reliability Research

Towards a Science of AI Agent Reliability

Across 15 models and two benchmarks, capability gains have produced only marginal reliability improvements. The authors propose 12 metrics drawn from safety-critical engineering that decompose what a single accuracy score actively conceals: consistency, robustness, predictability, error severity.

Who's behind this?

The Princeton and Meta team that published "AI Agents That Matter," extending a credible research program on agent evaluation rigor.

Bigger models, bigger problems?

Frontier models often show lower consistency than smaller ones. Enterprises adopting them may be importing more behavioral variance, not less.

Reliability Research

Consistency Amplifies: How Behavioral Variance Shapes Agent Accuracy

71% of Claude's failures trace to the same incorrect assumption repeated across every run. Consistency amplifies outcomes without guaranteeing correctness, so the most reliable-looking agents can also be the most systematically wrong.

Where's the cost signal?

A confidently wrong agent burns the same token budget as a correct one. The meter stays perfectly quiet.

Will retries help?

Not when the failure is deterministic. Pass-at-k strategies cannot catch an agent making the same mistake every time.

Meter Shock

GitHub Copilot's Metered Billing Goes Live. Developers Are Discovering They Can't Predict What Their Agents Cost.

On June 1, GitHub Copilot swapped flat-rate subscriptions for usage-based billing. The listed prices didn't move. What they buy did: an opening balance instead of unlimited access. The backlash was almost immediate. One Pro+ subscriber burned roughly 8% of a month's credits in two hours. Another spent $6 on a single change request and called consumption "impossible to predict."

GitHub's chosen safeguard tells you a lot. Set your overage budget to $0 and the agent simply stops working when credits run out. No surprise bill, but no assistant either. The governance mechanism for overspend is denial of service. A defensible choice, sure. Also a concession that nobody, GitHub included, has figured out how to make agent costs legible before they're incurred.

Meter Shock

GitHub Copilot's Metered Billing Goes Live. Developers Are Discovering They Can't Predict What Their Agents Cost.

On June 1, GitHub Copilot swapped flat-rate subscriptions for usage-based billing. The listed prices didn't move. What they buy did: an opening balance instead of unlimited access. The backlash was almost immediate. One Pro+ subscriber burned roughly 8% of a month's credits in two hours. Another spent $6 on a single change request and called consumption "impossible to predict."

GitHub's chosen safeguard tells you a lot. Set your overage budget to $0 and the agent simply stops working when credits run out. No surprise bill, but no assistant either. The governance mechanism for overspend is denial of service. A defensible choice, sure. Also a concession that nobody, GitHub included, has figured out how to make agent costs legible before they're incurred.

Safety valve: Previously, exhausted users dropped to a cheaper model and kept working. That fallback is gone. Now it's full access or a wall.

Student impact: One verified student burned all 200 credits ($2.00) on day one after four or five chat requests. Two dollars buys remarkably little.

Frozen plans: New sign-ups for Student, Pro, Pro+, and Max have been paused since April 20, six full weeks before billing went live.

Agentic sticker shock: Community estimates peg a typical multi-file planning and implementation session at $30–40 in credits at frontier-model pricing.

Enterprise glitch: One admin set the org budget to $0 and users saw "No Usage Limit" displayed on their side, inverting the intended safeguard entirely.

Market Pulse

Market Pulse

What the Bill Can't Measure

What the Bill Can't Measure

Towards a Science of AI Agent Reliability

Consistency Amplifies: How Behavioral Variance Shapes Agent Accuracy

Towards a Science of AI Agent Reliability

Consistency Amplifies: How Behavioral Variance Shapes Agent Accuracy