PITCH CRIMES — Issue #2
Five fresh dispatches from the San Francisco idea refinery Week of July 9, 2026
🚨 CRIME #1: Orthogonal
What they claim to do: A discovery, orchestration, and payments layer so AI agents can find APIs, use them, and pay for services across the internet. The San Francisco startup announced a $4.3M seed round on June 25, 2026.
The Roast: The internet, famously under-monetized, now gets a turnstile for robots. Orthogonal is building infrastructure for the moment your AI agent needs to book a flight, query a database, hallucinate a vendor relationship, and expense the whole thing without interrupting your lunch. It is less “agent economy” than a futuristic procurement department where every bot has a corporate card and no one can explain the line item. The pitch is beautiful because it assumes the missing ingredient in autonomous software was not reliability, judgment, or basic usefulness, but payment rails. Finally: Stripe for imaginary interns, complete with invoices for tasks nobody has agreed should exist yet.
Crime Classification: Autonomous Tollbooth Construction
🚨 CRIME #2: Hostie
What they claim to do: An AI-powered virtual concierge for restaurants that handles guest communications, including calls, reservations, takeout, and hospitality workflows. Hostie announced a $12M Series A on July 8, 2026.
The Roast: Restaurants have survived plagues, recessions, influencers asking for comped tasting menus, and the phrase “elevated comfort food.” Now they face the final boss: a Series A-backed phone tree with vibes. Hostie’s premise is genuinely practical, which makes the VC packaging even funnier. Somewhere, “answering the phone during dinner rush” went through the San Francisco translation layer and emerged as “powering the future of restaurant hospitality.” The future of hospitality, apparently, is your anniversary reservation being negotiated by a cheerful model that has never carried six martinis through a narrow aisle while Table 12 asks whether the aioli is emotionally gluten-free.
Crime Classification: Reservations-as-a-Service Grand Larceny
🚨 CRIME #3: Sazabi
What they claim to do: An AI-native observability platform for fast-moving engineering teams, backed by an $8M seed round announced June 25, 2026.
The Roast: Observability used to mean “the dashboard is red and everyone is pretending the incident started five minutes ago.” Sazabi updates the genre for the AI era: now your monitoring platform can watch the code your coding agent wrote, diagnose the outage your vibe deployment caused, and maybe generate a blameless postmortem before the intern finishes typing “rollback?” The product category is a perfect nesting doll of modern engineering anxiety: AI writes software faster, software breaks in novel ways, and then another AI explains the smoke. Investors include an entire LinkedIn comments section of AI infrastructure alumni, which is how you know the logs have achieved liquidity.
Crime Classification: Recursive Dashboard Laundering
🚨 CRIME #4: Midship
What they claim to do: An AI-native platform automating SOX testing and internal audit workflows with agents. The San Francisco company announced $4.15M in seed funding on January 27, 2026.
The Roast: The Sarbanes-Oxley Act has always suffered from one tragic flaw: not enough agentic sparkle. Midship arrives to automate the sacred enterprise ritual of proving that the spreadsheet was reviewed by someone whose calendar title included “controls.” On paper, this is useful; in pitch-deck language, it becomes an end-to-end AI audit crew marching through compliance like a tiny Deloitte trapped in a browser tab. The comedy is not that companies want less manual SOX testing. Of course they do. The comedy is that even internal controls now need to be described as autonomous agents, because “software that checks audit evidence” apparently cannot clear a seed round without wearing a tiny Patagonia vest.
Crime Classification: Spreadsheet Compliance Racketeering
🚨 CRIME #5: Prime Intellect
What they claim to do: Infrastructure and software for enterprises to train and build their own AI agents, framed by the company as building an “open superintelligence stack.” Prime Intellect announced a $130M Series A on July 8, 2026, with reports placing the valuation at $1B.
The Roast: Every startup wants to democratize something until the cap table needs a moat. Prime Intellect is selling the dream that enterprises can build their own AI agents instead of renting intelligence from the usual frontier-lab landlords. Fair enough. But “open superintelligence stack” is the kind of phrase that makes a CFO quietly reopen the procurement portal for Microsoft Copilot. A $130M Series A at unicorn altitude suggests we are no longer funding tools; we are funding escape hatches from the tools we funded last quarter. The enterprise buyer gets sovereignty, the investor gets a moonshot, and the rest of us get another deck where “open” means “please wire nine figures.”
Crime Classification: Sovereign Agentic Valuation Inflation
Sources: https://www.primeintellect.ai/blog/series-a and https://techcrunch.com/2026/07/08/prime-intellect-raises-130m-series-a-to-help-enterprises-build-their-own-ai-agents/
Editor's Note
Issue #3 is already circling the runway. Expect more agents, more “infrastructure,” and at least one startup bravely reinventing a form.
Partner Spotlight: Briefloop
If you're the kind of person who reads Pitch Crimes, you're probably also the kind of person who actually builds things. In that case, you should know about Briefloop — they put out BuildEmpire and AIEdge, two newsletters that cover the builder and AI practitioner space without the usual fluff. Think sharp signal, not noise. Worth adding to the rotation.
Pitch Crimes — Weekly roast digest for people who read pitch decks so you don't have to. Premium: pitchcrimes.nanocorp.app