Shaun Modi

Made in ⁦@capitolai⁩

Visualization on quantum entanglement, built on Capitol AI. Our team is cooking on a Friday night, 11:15 p.m. Eastern Standard Time.

If there’s one thing I hope you’ll take from my recent piece in Battle Updates is that defense procurement is overdue for modernization. There is a great need for structured intelligence in defense procurement and for AI in high-stakes environments to be governed from day one as a baseline requirement. I hope you’ll take just a few minutes to read it: battle-updates.com/ai-procurement…

🤓

Here is the Composer workflow that powered it. Let me know what data you'd like to see.

Most AI is a chatbot. This Persian Gulf Crisis Global Energy Trade Impact Dashboard was built in a fully automated way using @capitolai's Composer product. It maps a 97% collapse in traffic through the Strait of Hormuz, freight insurance shocks, fertilizer + food security trends, and country exposure... amongst other key trends that matter. Play around with the link (on a desktop). Would love to hear your feedback on what's missing. Can be easily updated by simply dropping datasets into the Composer. Raw, but that's how we like it. vercel-deploy2-khaki.vercel.app

Cool statue of General George Thomas in front of @capitolai’s new DC office. He was known as “The Rock of Chickamauga”. civilwarmonitor.com/rock-of-chicka…

Consulting firms partnering with OpenAI or Anthropic makes sense right now. Enterprises want speed, and model companies want distribution. But there’s a longer term issue. When your core capability is tied to a single model provider, you inherit their roadmap, pricing power, and policy decisions. That’s manageable in experimentation but becomes very risky in defense, finance, and other regulated industries. In high stakes environments, model choice is a requirement. The firms that separate orchestration from any single model will have more leverage and more durability because infrastructure outlasts models. wsj.com/tech/ai/ai-nee…

Appreciated the chance to speak with Pulse 2.0's @amitchowdhry about what we’re building at @capitolai and how we got here. I’ve always been drawn to the space between information and understanding and creating Capitol has been my answer to fixing the bottleneck issues I’ve seen over my career in the public and private sector. This piece dives into why we believe the next phase of AI is about model choice, data sovereignty, infrastructure, and governance inside real workflows. You can read the full interview here: pulse2.com/capitol-ai-pro…

The case for sovereign AI is very real

First day in @capitolai’s new office in Washington DC!

2027: >_ the world’s most critical organizations are facing an intelligence paradox: they have more data than ever, but less control over the outcomes. they are currently renting intelligence from west coast monoliths. creating a monopolistic lock-in that turns their proprietary IP into someone else's training data. this isn't just a technical hurdle; it’s a security liability. @capitolai

Nearly 2/3 of companies say they don’t know where all their data lives while simultaneously giving AI broad access to it. I see this tension constantly and it’s clear the issue is a sovereignty problem rather than an AI problem. Shared drives, data rooms, internal messages, research archives and machine speed wired into the organization before leadership even fully understands what’s inside. This scenario invites risk to compound which, for institutions, shows up later in audits, litigation, and regulatory reviews. If a conclusion can’t be traced, defended, and repeated across teams under the same conditions pulling from the same source material, that becomes improvisation rather than infrastructure. Sovereignty keeps control intact and repeatability makes decisions durable. The winners in this phase of enterprise AI will understand their data well enough to let intelligence run and structured well enough to reproduce every outcome when challenged. Otherwise you’re just accelerating uncertainty. fortune.com/2026/02/25/tha…

Consultancies are entering a new phase of the AI cycle. The era of open-ended experimentation is closing. Clients are no longer impressed by pilots. They expect measurable financial and operational impact. In advisory work – diligence, integration, compliance, transactions – that pressure shows up fast. Teams are consolidating thousands of documents under compressed timelines, and every recommendation has to stand up to board-level and regulatory scrutiny. This isn’t a chatbot problem. It’s an infrastructure problem. The firms pulling ahead are embedding governed, model-agnostic intelligence directly into their workflows. Not to replace expertise, but to strengthen it. Advisory teams can synthesize proprietary information into clear, attributable, decision-grade artifacts that clients can actually act on. Our work with organizations like EY reflects this shift. By operating inside secure environments, teams generate structured deliverables from sensitive data while preserving sovereignty, attribution and control. When AI lives where expertise already lives, it doesn’t create noise. It sharpens judgment. That’s the phase we’re in now. giftarticle.ft.com/giftarticle/ac…

Abstraction of engineering is the surest bet.

New York State of mind

In high stakes environments, KYC succeeds when systems adapt to context. Risk isn’t static. Clients differ. Jurisdictions differ. Context matters. Yet too many KYC systems still force every relationship through the same rigid process, regardless of what’s actually at stake. This is where AI earns its place: by supporting human judgment inside real workflows. Adaptive KYC lets teams calibrate depth, scrutiny, and explanation to the situation in front of them. Low risk cases move quickly. Complex ones receive deeper analysis, with reasoning preserved and decisions defensible over time. That approach only works if the data never leaves its secure environment. KYC lives on sensitive, regulated information. Data sovereignty and auditability form the foundation of credible intelligence. The goal is better artifacts: clear, attributable outputs that show why a client was cleared, flagged, or escalated. Regulators can review them, teams can stand behind them, and institutions can rely on them with confidence. When KYC systems align with human judgment and adapt to context, trust scales.

Honored to gather at Ambassador Shigeo Yamada’s residence to talk about the future of the US-Japan space partnership / making sense of space data— and to welcome @NASA Administrator @rookisaacman. @JapanEmbDC @JAXA_en @InstituteForEdu 🇺🇸🛰️🇯🇵

Unemployment in the UK has climbed to 5.2% and AI sits at the center of conversations around that. I recently spoke with @CityAM's @SaskiaLKoopman about how this narrative is taking shape in her recent story. Every major technology shift begins with compression. Workflows tighten. Hiring slows while companies rethink how work gets done. Markets reprice quickly long before business models fully evolve but then, eventually, reinvention follows and new roles emerge. We’ve seen this before and there’s no reason to believe AI would follow a different arc of compression followed by reinvention. The real question is how businesses channel productivity gains into growth, expansion, and new categories of work. cityam.com/is-ai-really-t…