Friday, June 26 June 26, 2026

Good morning. It's Friday, June 26th, 2026, and today's briefing is about accountability — who's watching AI, who's funding it, and who's quietly controlling the hardware stack that makes it all possible.

Let's start with the biggest regulatory move of the week. California just became the first state in the nation to launch an official AI job-loss tracker. Governor Gavin Newsom signed the executive order and unveiled the California AI-Unemployment Tracker — a real-time, public dashboard that monitors AI-related unemployment claims across the state. This is not a warning shot. This is a state government telling businesses: we're watching, we're measuring, and eventually we'll regulate based on what we see. Expect this model to spread to other states fast.

On the infrastructure side, the New York Times is out with a deep dive on advanced chip packaging — a technology called CoWoS — and it's raising serious supply-chain alarms. Advanced packaging is what lets GPU dies communicate fast enough to power today's frontier AI models. The problem? TSMC in Taiwan has a near-monopoly on this process. Counterpoint Research confirmed this week that global foundry revenue grew 23 percent year-over-year in Q1 2026, driven almost entirely by AI chip demand. But that growth is funneling through a single geographic chokepoint. If anything disrupts Taiwan's capacity, the entire AI build-out stalls. This is the hidden bottleneck most people aren't talking about.

Now let's talk about where the money is going. Trase, a McLean, Virginia startup, just raised a staggering $107 million in a seed round — that is not a typo — led by ARCH Venture Partners. What does Trase do? They build an agentic operating system for regulated industries: healthcare, financial services, government. Their platform is already being tested at Duke Health. At $107 million in seed funding, they're betting that enterprises in highly regulated sectors are done piloting AI and are ready to deploy it at scale. Duke Health testing them already signals this is real traction.

In enterprise AI, Salesforce launched its Agentforce Help Agent this week with a provocative new pricing model: pay-per-resolution. Not per seat, not per month — per resolved customer issue. Salesforce is essentially making a performance guarantee, which forces every competitor to answer the same question: can your AI actually close tickets? This signals a broader shift in how enterprise AI will be bought and sold. Subscription pricing is giving way to outcome-based billing across the industry, with OpenAI, Anthropic, and Microsoft all reportedly moving in this direction.

Cohere dropped a quiet bombshell yesterday with their North Mini Code model — a 30-billion-parameter model that they claim beats models four times its size on coding benchmarks. For enterprises that want frontier coding capabilities without the infrastructure cost of a 120-billion-parameter behemoth, this is a big deal. Smaller, specialized models are eating the bottom of the market that the big labs thought was safely theirs.

And there's one more signal worth noting: Mistral released OCR 4, their enterprise-grade document processing model. It includes bounding boxes, block classification, and inline confidence scores. Document automation is the killer use case for enterprise AI right now — the one that proves ROI fastest. Mistral is quietly building a serious enterprise stack with each of these incremental releases.

Finally, a quick policy note: a GAO report out this week found the IRS is deploying AI without adequate governance frameworks. A federal agency using AI at scale with no formal oversight structure — that's the kind of story that ends up as a congressional hearing. Watch this one.

The thread connecting all of it today: AI is scaling into critical infrastructure — healthcare, taxation, semiconductor supply chains — and governments and enterprises are only now starting to build the guardrails. The next 18 months are going to be defined by who builds those guardrails well and who gets caught without them.

Here's today's business idea. Build a regulatory AI readiness auditing service targeting mid-market enterprises in healthcare and finance. As California's job tracker and federal oversight gaps signal increased AI scrutiny, companies that can't document their AI governance will face mounting liability. A 90-day audit-to-certification product that benchmarks an enterprise's AI stack against emerging state and federal standards could command $50,000 to $150,000 per engagement and position you perfectly ahead of the compliance wave.

That's your Morning AI briefing for Friday, June 26th. Stay sharp, and we'll see you Monday.