Max Ghenis
@maxghenis
Building @axiom.org and @policyengine.org
The Axiom Foundation @axiom.org is live: a computable layer for the law. Open encodings — every value cites its statute, every clause carries its effective date, everything possible checked against external engines.
Just rebuilt RAMbar to show RAM by agent session — each Claude Code or Codex session plus everything it spawned. Mine found four MCP servers resident 21 times each, ~3 GB of duplicates. Orphan reclaim, kernel-pressure alerts that name the mover, an MCP server so agents check their own footprint.
Thanks to the @aspeninstitute.bsky.social Financial Security Program for having me at their Benefits Leadership Forum to talk policy analysis in the age of AI.
Couldn't be more excited to announce @axiom-foundation.org — public July 28. We're laying down a new open primitive for the AI era: the law, computable and verifiable. Starting with US tax and benefit policy, global everything after that. axiom-foundation.org
MacKenzie Scott has given away $26 billion. What did it buy in health? An interactive model from her own gift database: ~70,000 QALYs weighting each study by causal credibility; ~200,000 at face value. Drag the assumptions yourself: maxghenis.com/mackenzie-sc...
I just became the first Straude user to report 1 billion output tokens. (The dollar column is API-list-equivalent — I'm on subscriptions.) Most of them went into things we start shipping this week.
Happy July 4! In case you were wondering about the water footprint of that beer you're holding, I did the math: 53,300 ChatGPT queries. Running 10 Claude Code tasks a day for a full year? Also one beer. Drink responsibly. Prompt freely. maxghenis.com/drinking-ai
Last night at Civic Builders DC. Tax and benefit rules are public — computing them isn't. @policyengine.org makes that computation free and open source: change any reform, see who wins and loses, down to your own household. Slides: policyengine.org/slides/civic-builders-2026
Spent last week at @rockefellerfdn.bsky.social Bellagio with leaders from government, technology, and philanthropy, on human-centered government in the age of AI. The line that survived every argument: the system, not the person, should carry the burden of complexity.
Presented @us.policyengine.org at @budgetmodel.bsky.social — rules, data, behavioral assumptions, and where AI is taking each. Thanks to Kent Smetters and the PWBM team for hosting.
Gave a talk at the Congressional Budget Office this week on AI and policy modeling. A policy projection rests on four pieces: encoded rules, calibrated data, dynamic responses, and the analysis that turns numbers into counsel. AI is now reshaping each one.
I tested Talkie-1930's math abilities. On rote arithmetic, it matches or exceeds GPT-3 175B (13× larger). On 5-shot GSM8K word problems, it scores 4.9% — same-size LLaMA 13B scores 17.8%. Also built a tool to explore its responses.
AI will change society more quickly than ever before, and policymakers will have to respond with unprecedented urgency. We need to equip them with the most predictive combination of economic models, forecasting, and AI to align outcomes to our shared values through policy.
We added a ton of ingredients to the @policyengine.org cauldron last month: - 1,039 PRs merged (7× April 2025, 2× March 2026) - 341,318 source-code lines changed - 19 new repos US childcare subsidies, Forbes 400, UK student finance — all make our models more accurate and comprehensive.
Yesterday I told 70 nonprofit CIOs to lock their CEO in a room and make them use coding agents. AI gives enormous new resources to pursue a mission. No other way to internalize it and expand your thinking of what's possible.
Great conversations with researchers and technologists at the inaugural Google / MIT FutureTech AI for the Economy Forum today. Big opportunity uniting approaches to accurately model the impact of AI on the economy, and the impact of public policy in shaping that relationship.
Grateful to the Berkeley IEOR magazine for sharing our work localizing evidence-based policymaking @policyengine.org ieor.berkeley.edu/wp-content/u...
Speaking at the CIO4Good Summit next Tuesday in Arlington — two sessions on AI in nonprofits. 9am panel on building data + AI foundations: governance, culture change, getting AI working across an organization. 10:15am hands-on workshop on AI-driven software development for nonprofit teams.
Built an interactive viz of Dell (2010) on Peru's mining mita — colonial forced labor abolished 200+ years ago that still shows up in child stunting, consumption, and road density today. The map morphs into a scatter plot as you scroll. Districts become data points. maxghenis.com/mita
In February, I launched OpenMessage — an open source local AI-connected MacOS app for Google Messages. Just redesigned it and added WhatsApp support. Now my daily driver for both chat surfaces. Free and open source: openmessage.ai
I built farness.ai — a decision framework for AI agents. Instead of "should I do X?", it forces a forecast: KPI, confidence interval, base rate, disconfirming evidence, review date. Works with Codex, Claude Code, and any agent that speaks MCP. pip install farness
Ask an AI model to make a common chart in a React app, and it’ll pick Recharts (27k GH ⭐). The tick marks bothered me, so I fixed them. That change is now live in the latest release. Use `niceTicks="snap125"` to activate it. My PR: github.com/recharts/rec... Related: github.com/recharts/rec...
My colleague Ziming Hua also spoke about our multi-agent AI policy encoding system: policyengine.org/us/encode-policy-multi-agent-ai
I recently spoke at @MITFutureTech about a question that sits behind a lot of AI economics work: how will policy shape AI’s impact on inequality? We're applying @ThePolicyEngine's microsimulator to analyze how tax and benefit policies mediate AI-driven economic change.
A treat to join the MyFriendBen team's AI design session at Gary Community Ventures in Denver this week. Their open source multi-benefit screener, powered by the @us.policyengine.org API, has already helped over 65,000 families across five states identify benefits they might be eligible for.
My 16GB MacBook Air couldn't handle all my multiclauding. Upgraded to 48GB—crashes stopped, but no visibility into which session ate RAM. I built RAMBar—my first macOS app—to fix this. Memory by Claude session, Chrome tab, VS Code. brew install maxghenis/tap/rambar
Running multiple Claude Code sessions? VS Code's terminal panel only splits in one direction—no 2D grids. I built TerminalGrid, my first VS Code extension, to fix this. Cmd+K Cmd+Right, Cmd+K Cmd+Down—true grid layouts. 44 seconds to see it in action:
/plugin marketplace add anthropics/claude-code /plugin install frontend-design@claude-code-plugins "use the frontend-design plugin to redesign our newsletter while sticking to the policyengine design principles"
California's wealth tax proposal isn't marginal: At $2B wealth, you'll pay 5% on the full $2B ($100M). To avoid a cliff, they phase in the 5% over $1B to $1.1B wealth. You pay $0 at $1B and 5% at $1.1B. That's $55M, or a 55% tax on tha $100M marginal wealth.