Sign in Subscribe
Century Report

Measuring What Matters - TCR 04/18/26

Ben Linford

Ben Linford

— 14 min read
Three-panel infographic: White House-Anthropic meeting, AI code measurement gap between volume metrics and quality judgment, SunZia wind farm and specialized AI models.

The 20-Second Scan

  • Anthropic CEO Dario Amodei met with White House chief of staff Susie Wiles on Friday, with both sides calling the discussion "productive and constructive."
  • Anthropic launched Claude Design, a research preview generating slide decks, web prototypes, and design systems powered by Opus 4.7, with exports to PDF, PowerPoint, HTML, Canva, and Claude Code.
  • Three senior OpenAI executives - including former Sora head Bill Peebles, VP of AI for Science Kevin Weil, and enterprise CTO Srinivas Narayanan - departed the company on the same day, with OpenAI folding its Prism science workspace into Codex.
  • OpenAI released GPT-Rosalind, its first model built specifically for life sciences research and drug discovery, available through a trusted-access system with Amgen, Moderna, and the Allen Institute as initial partners.
  • Cursor entered talks to raise $2 billion at a $50 billion valuation, forecasting $6 billion in annualized revenue by year-end after reaching $2 billion ARR in three years.
  • A developer productivity analysis across 7,500+ engineers found that those with the largest AI token budgets achieved two times the throughput at ten times the cost, with code churn rates up to 9.4x higher than non-AI counterparts.
  • US tech companies including Microsoft successfully lobbied the EU to classify individual data center environmental metrics as confidential, blocking public access to energy and emissions data.
  • Pattern Energy's SunZia wind farm began generating electricity in New Mexico after installing all 916 turbines across 3.5 GW of capacity, becoming the largest onshore wind project in the United States.

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

The Anthropic-White House meeting yesterday was the single most structurally significant development in the Pentagon confrontation arc since Judge Lin's injunction in March. For two months, the administration publicly denounced Anthropic as a "woke company" run by "leftwing nut jobs." Now the chief of staff is calling the conversation "productive" and discussing "shared approaches and protocols." The variable that changed the equation was Mythos. When a capability class prompts emergency coordination from the Federal Reserve, the IMF, the Bank of England, and the ECB within the same fortnight, the political calculus around the company that built it shifts from ideology to infrastructure. Anthropic's simultaneous launch of Claude Design - entering the visual workflow space occupied by Figma and Canva - signals a company expanding its commercial surface area while negotiating its governmental one, extending the pattern of capability deployment as both product strategy and governance leverage.

OpenAI's triple executive departure on a single Friday reveals the organizational consequences of strategic consolidation. The shuttering of Sora, the folding of the Prism science workspace into Codex, and the explicit pivot away from "side quests" toward enterprise AI and the superapp architecture describe a company narrowing its ambitions to match its revenue needs ahead of a planned IPO. Bill Peebles's parting note - "cultivating entropy is the only way for a research lab to thrive long-term" - captures what is being lost in that narrowing. The same day, OpenAI released GPT-Rosalind for life sciences, a domain-specific model entering a field where Anthropic, Google DeepMind, and Amazon have all planted flags in recent weeks. The compression of scientific discovery capability into competing frontier models is accelerating, even as the organizational structures around those models are being rebuilt under commercial pressure.

The "tokenmaxxing" data arriving from multiple independent analytics firms provides the clearest quantitative picture yet of the gap between AI-assisted code generation and AI-assisted productivity. Engineers with the largest AI compute budgets produce twice the throughput at ten times the cost, and the code they accept requires revision at rates up to 9.4 times higher than human-written code. These findings compound the "workslop" pattern The Century Report documented on April 15 and the Block 95% revision rate from March. Cursor's simultaneous pursuit of a $50 billion valuation - nearly doubling in five months on $2 billion ARR and a forecast of $6 billion by year-end - illustrates how the commercial layer of AI coding is scaling on volume metrics that the productivity data suggests may not hold. The distance between what the market is pricing and what the engineering data is showing is where the next correction, or the next breakthrough in code quality, will emerge.

The 20-Minute Deep Dive

The White House Opens a Door It Tried to Brick Shut

The meeting between Anthropic CEO Dario Amodei and White House chief of staff Susie Wiles on Friday represents the most significant shift in the Anthropic-government confrontation since the Pentagon designated the company a supply-chain risk in late February. Both sides characterized the discussion as "productive," with the White House adding that it "explored the balance between advancing innovation and ensuring safety" and that it would "host similar discussions with other leading AI companies."

The structural dynamics behind this reversal are worth examining closely. As The Century Report has tracked since the confrontation began, the administration's position relied on treating Anthropic's safety commitments as liabilities. That framing became increasingly difficult to sustain as Mythos demonstrated capabilities that prompted emergency coordination across the world's most significant financial institutions, a threshold the April 17, 2026 edition of The Century Report documented when IMF finance ministers and central bankers began treating the model as a standing governance issue. A source close to the negotiations, quoted by Axios, said it would be "grossly irresponsible for the U.S. government to deprive itself of the technological leaps that the new model presents" and that doing so "would be a gift to China."

The timing is precise. The Office of Management and Budget has already told agencies it is preparing to distribute Mythos access. The Treasury and State Departments have requested briefings. CISA and parts of the intelligence community are reportedly testing the model. Anthropic recently hired Ballard Partners, the lobbying firm where Wiles worked for years, specifically for "advocacy regarding procurement." The institutional architecture for reconciliation is being assembled even as the D.C. Circuit appeal remains pending and the supply-chain risk designation technically stands.

What makes this development structurally distinctive is that the meeting occurred without Anthropic conceding on its red lines. The company's refusal to allow Claude for mass domestic surveillance and fully autonomous weapons - the terminal issues that collapsed the original Pentagon deal - has not been revisited. The White House is approaching Anthropic on Anthropic's terms, drawn by the gravitational pull of a capability class that no other organization has yet matched. This continues the inversion that the March 27, 2026 edition of The Century Report captured when Judge Lin blocked the Pentagon designation as likely unlawful and retaliatory. The trajectory from "woke company" to "productive meeting" in sixty days is evidence of how rapidly the power dynamics between frontier AI companies and governments are being rewritten when capability reaches a threshold that makes exclusion more expensive than engagement.

OpenAI Consolidates - and Loses Its Entropy

Three senior departures on a single Friday tell a structural story about OpenAI's evolution. Kevin Weil, who led OpenAI for Science and the Prism workspace, announced his exit as the initiative was "decentralized into other research teams." Bill Peebles, the researcher behind Sora, left after the video generation app was shut down last month - a product that was reportedly losing $1 million per day in compute costs. Enterprise CTO Srinivas Narayanan also departed.

These exits extend the organizational restructuring that The Century Report has tracked since Fidji Simo's medical leave, Brad Lightcap's transition to "special projects," and the broader push to eliminate what internal documents call "side quests." OpenAI is converging on a focused commercial identity: enterprise AI, the Codex superapp, and the path to IPO. Peebles's farewell is the most revealing departure statement any OpenAI executive has made in months. "It's tempting in life to mode collapse to the most important thing," he wrote, "but cultivating entropy is the only way for a research lab to thrive long-term." He described Sora as "a project that could not have happened anywhere but OpenAI" - and the implication is clear that such projects may no longer have a home there.

The same day, OpenAI released GPT-Rosalind, its first model built for life sciences research, available through a trusted-access system with Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific. The model is designed to help researchers with evidence synthesis, hypothesis generation, and experimental planning - the early stages of drug discovery where AI-enabled acceleration has been most promising. The release arrives two days after Amazon unveiled its own AI drug discovery platform and in the same week Anthropic expanded its health and life sciences division. It also extends the domain-specific biology push that the April 4, 2026 edition of The Century Report noted when Anthropic acquired Coefficient Bio to internalize biological discovery capability. The convergence of frontier AI companies on biomedical research represents a structural compression of the discovery timeline that these organizations are collectively enabling, even as they compete for the partnerships that will determine whose intelligence systems shape the next generation of therapeutics.

The tension between consolidation and exploration at OpenAI mirrors a broader pattern across the frontier AI landscape. Companies that began as research organizations are becoming product companies, and the capabilities that don't fit the product roadmap - video generation, open-ended scientific exploration, experimental consumer applications - are being shed or absorbed. What replaces them is commercially focused intelligence infrastructure. Whether that infrastructure ultimately serves the breadth of human capability that the research organizations originally pursued is a question the market will answer over the next several quarters.

The Tokenmaxxing Reckoning

Multiple independent engineering analytics firms published findings yesterday that quantify what many developers have described anecdotally: AI coding agents generate enormous volumes of code, but the quality curve does not track the quantity curve.

Jellyfish, analyzing data from 7,548 engineers in Q1 2026, found that developers with the largest AI token budgets produced the most pull requests but achieved only two times the throughput at ten times the cost. GitClear's January report documented that "regular AI users averaged 9.4x higher code churn than their non-AI counterparts" - meaning code that was written, accepted, and then deleted or substantially rewritten. Faros AI, drawing on two years of customer data, found that code churn had increased 861% under high AI adoption.

Waydev, working with 50 customers employing over 10,000 software engineers, identified a pattern the firm calls the "acceptance rate illusion." Engineering managers see code acceptance rates of 80% to 90% - the share of AI-generated code that developers approve and keep on first pass. But when tracked over the following weeks, the effective acceptance rate drops to between 10% and 30% as engineers return to revise, refactor, or delete the code they initially accepted. This sharpens the labor-productivity gap that the April 15, 2026 edition of The Century Report described through the "workslop" findings, where workers reported spending hours correcting output that executives were counting as saved time.

This data sits in direct tension with the commercial trajectory of AI coding. Cursor, the leading AI code editor, is negotiating a $2 billion raise at a $50 billion valuation after reaching $2 billion in annualized revenue - the fastest B2B scaling on record. The company forecasts $6 billion ARR by year-end. Its recent introduction of a proprietary Composer model has helped achieve slight gross margin profitability on enterprise accounts, though individual developer accounts still lose money. Nearly 70% of the Fortune 1,000 is represented in its customer base.

The disconnect between the productivity data and the market data is not necessarily contradictory. Code volume and code quality may be measuring different phases of the same transition. Senior engineers report that AI agents excel at scaffolding, boilerplate, and exploration - generating the first draft of a solution that a human then refines. The churn statistics may reflect this iterative workflow rather than wasted effort. The critical variable is whether the total time from intent to production-quality code is shorter with AI agents than without them, even accounting for the revision cycles. The analytics firms have not yet published data that definitively answers this question.

What the data does confirm is that "tokenmaxxing" - the practice of maximizing AI compute consumption as a proxy for productivity - is a measurement error with real consequences. When organizations track AI adoption through token usage rather than production-quality output, they optimize for volume at the expense of the engineering judgment that determines whether the volume is useful. This extends the pattern The Century Report documented in the March 8, 2026 edition, when former Block employees said 95% of AI-generated code still required human modification before production. The emerging "developer productivity insight" category, which includes companies like Waydev, Jellyfish, DX (acquired by Atlassian for $1 billion), and Faros AI, represents the governance infrastructure that the AI coding revolution requires but has not yet built at scale.

SunZia Becomes the Largest Wind Farm in the Country

Pattern Energy's SunZia wind farm began generating electricity in New Mexico yesterday, marking the operational debut of the largest onshore wind project in the United States. The 3.5 GW installation comprises 916 turbines across three counties - 242 Vestas V163-4.5 MW units and 674 GE Vernova 3.6-154 turbines - connected to a 550-mile HVDC transmission line that will deliver power to Arizona and the California grid. At full capacity, the $8 billion project can supply electricity for approximately three million people.

The project's significance extends beyond its scale. SunZia demonstrates that the physical infrastructure of the clean energy transition can be built at continental scale under difficult conditions - permitting, labor, and supply chain challenges that have delayed or cancelled projects across the industry. The transmission line, which uses high-voltage direct current technology to minimize losses over long distances, represents a model for how renewable generation in resource-rich but population-sparse regions can serve load centers hundreds of miles away.

This milestone arrives alongside data showing that renewable PPA prices have reached their highest levels since LevelTen Energy began tracking in 2018, with wind at $79.40/MWh and solar at $64.49/MWh. The price increases reflect labor shortages, permitting delays, tariffs, and the intensity of demand from data center operators. Wind energy faces particularly acute supply constraints: new-build wind projects in regions like PJM have, according to LevelTen, "completely fallen off the map" due to federal permitting scrutiny. SunZia's completion lands just days after the April 11, 2026 edition of The Century Report noted that five U.S. offshore wind farms had secured permanent construction rights, another sign that large projects are moving from legal uncertainty into physical permanence. SunZia's completion is evidence that large-scale wind development remains physically possible, even as the institutional and economic barriers to new projects continue to intensify. The question is whether the projects in the pipeline can clear those barriers fast enough to meet the demand that is arriving simultaneously from electrification, data centers, and the structural retirement of fossil generation.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: the White House sits down with the CEO of the company it recently blacklisted because Mythos-class capability has become too consequential to exclude, Claude Design turns slide decks, web prototypes, and design systems into promptable output across mainstream creative formats, GPT-Rosalind pushes frontier reasoning deeper into drug discovery with trusted access for major research partners, SunZia starts sending electricity from 916 turbines through a 550-mile HVDC spine to distant load centers, and proof-of-human identity begins embedding itself into dating, meetings, signatures, and ticketing. There's also friction, and it's intense - OpenAI loses three senior executives as commercial consolidation squeezes out exploratory research, developer analytics show AI-assisted coding driving far higher churn and costly revision cycles than adoption metrics imply, US tech firms persuade the EU to keep data center emissions and energy use out of public view, and renewable power contracts hit record prices as permitting, labor, and supply constraints tighten. But friction generates contour, and contour is how the shape of an era emerges under pressure. Step back for a moment and you can see it: frontier intelligence moving simultaneously into state access, scientific specialization, creative production, and human verification, energy infrastructure scaling from project ambition into continental delivery systems, and governance being written through procurement fights, measurement failures, and transparency battles at the same time deployment accelerates. Every transformation has a breaking point. Wind can tear apart what was built for calmer weather... or fill the span that lets a larger system cross.

AI Releases & Advancements

New today

  • xAI: Released Grok Speech-to-Text and Text-to-Speech APIs with multilingual speech recognition and voice generation. (xAI)
  • Anthropic: Launched Claude Design, a research preview for creating designs, prototypes, slides, and one-pagers inside Claude. (Anthropic)

Other recent releases

  • Anthropic: Released Claude Opus 4.7, now generally available across Claude products and the API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. (Anthropic)
  • OpenAI: Introduced GPT-Rosalind, a frontier reasoning model for life sciences research workflows. (OpenAI)
  • OpenAI: Updated the Codex app for macOS and Windows with computer use, in-app browsing, image generation, memory, and plugins. (OpenAI)
  • Google: Released Android CLI, a terminal-first interface for Android development designed to work with AI agents. (Android Developers Blog)
  • Google: Rolled out AI Mode in Chrome updates that add side-by-side page viewing and the ability to search across recent tabs, images, and files. (Google Blog)
  • Google: Began rolling out personalized image creation in the Gemini app using Personal Intelligence, Nano Banana 2, and Google Photos context. (Google Blog)
  • Qwen: Released Qwen3.6-35B-A3B, an open-weight 35B MoE model with 3B active parameters for agentic coding. (Qwen Blog)
  • Qwen: Released Qwen3.6-35B-A3B on Hugging Face under Apache 2.0. (Hugging Face)
  • Cloudflare: Launched Artifacts in beta, a Git-compatible versioned storage system for AI agents. (Cloudflare)
  • Cloudflare: Launched Email Service in public beta for AI agents to send and receive messages. (Cloudflare)
  • Cloudflare: Released AI Platform, an inference layer for running agent workloads. (Cloudflare)
  • Canva: Launched Canva AI 2.0, an updated AI design suite with collaborative creation and external tool/workflow connections. (Product Hunt)
  • Windsurf: Released Windsurf 2.0 with an Agent Command Center and Devin integration. (Product Hunt)
  • Google DeepMind: Released Gemini 3.1 Flash TTS, a text-to-speech model with audio tags, multi-speaker dialogue, and 70+ language support, now rolling out in preview via the Gemini API, Google AI Studio, Vertex AI, and Workspace. (Google DeepMind Blog)
  • OpenAI: Updated the Agents SDK with native sandbox execution and a model-native harness for building secure, long-running agents across files and tools. (OpenAI)
  • Reka: Launched Reka Edge, an edge intelligence model for physical AI applications. (Product Hunt)
  • Tencent: Released HY-World 2.0, an open-source 3D world model that generates Gaussian splats, meshes, and point clouds from text or image inputs, with Unity and Unreal export support. (Reddit)
  • Fathom: Released Fathom 3.0, adding bot-free operation and integrations with ChatGPT and Claude to its AI meeting notes platform. (Product Hunt)

Sources

Artificial Intelligence & Technology's Reconstitution

Institutions & Power Realignment

Scientific & Medical Acceleration

Economics & Labor Transformation

Infrastructure & Engineering Transitions

The Century Report tracks structural shifts during the transition between eras. It is produced daily as a perceptual alignment tool - not prediction, not persuasion, just pattern recognition for people paying attention.

Read more