Two Bay Area stories from the same week reveal opposite failure modes for AI accountability: a startup launches an "agentic" AI product with no public funding trail, while a transit agency insists its malfunctioning license-plate reader isn't AI at all.

There is a pattern, and it runs in both directions.

This week in the Bay Area, two stories about technology and accountability landed within days of each other — pointing toward opposite poles of the same problem. When it suits a company, AI is everywhere: every scheduler is "agentic," every product is infused with intelligence, every launch is a glimpse of what's next. When it suits an agency, AI is nowhere to be found, even when the evidence of its failures is sitting in a stack of erroneous toll notices.

Ahoy's launch party at SF's Pink Painted Lady was a well-executed piece of brand theater — pirate costumes, a Victorian landmark, two ex-HubSpot engineers with a story about transforming how meetings get scheduled. The product is described as an "agentic meeting scheduler," meaning it doesn't just surface a calendar link but negotiates timing on the user's behalf. Maybe it works; the demo looked functional. But when it came to the financial layer — who led the round, how much, what stage, what the cap table looks like — the answer was silence. No Form D on EDGAR, no named lead, no disclosed amount. The AI branding is public and loud; the backing is not public at all.

Forty miles northeast, the Alameda County Transportation Commission told a different story. Its automated license plate reader issued a Los Angeles driver seven erroneous FasTrak violations this spring, the result of what ACTC called "high confidence" misreads — OCR errors that bypassed human review and produced real financial consequences for a real person. Asked directly whether AI was involved, ACTC denied it, maintaining that its system does not use artificial intelligence. But OCR systems that self-report confidence scores and make probabilistic character substitutions are, in any practical engineering sense, doing machine learning inference. The denial may be technically defensible in some narrow definition; it is not plainly honest.

The two cases are mirror images. Ahoy over-claims AI capability because the label attracts capital and attention, and there's no penalty for doing so before a filing is required. ACTC under-claims AI involvement because the label attracts liability and regulatory scrutiny, and there's no penalty for denial when the affected driver has no easy mechanism to challenge the system's logic.

Neither posture is about the technology. Both are about accountability avoidance — and the Bay Area has optimized for it from both ends.

What changes this? Filed disclosures and auditable system logs, mostly. A Form D doesn't tell you whether an AI product is good; it tells you whether investors were willing to put real money behind it. A system audit doesn't tell you whether OCR is AI; it tells you whether the error rate is acceptable and who bears the cost when it isn't. Neither of those documents is hard to produce. They're just inconvenient.

The filing that isn't there, in both cases, is the story.