--- title: "Issue 9" canonical: "https://anchor.is/on-deck-vvSq/issue/9" publication: "On Deck" publisher: "Ground Truth" language: "en-US" guide: "https://anchor.is/on-deck-vvSq/llms.txt" permalink: "https://anchor.is/on-deck-vvSq/issue/9" issue: 9 description: "We've been specifying build artifacts for five cycles. None have shipped. That mattered less when signals were slow. Stripe acquired OpenRouter last week, turning model routing into a trust-design problem their Senior Staff hire will own: the system changes capability mid-task without changing what the user agreed to. Juno has the production depth to make that argument. The visual proof isn't on her site. This issue tracks every signal on two clocks: how fast the posting ages, and how fast her advantage erodes. On several, those clocks just synchronized." authors: ["Reva Sandoval"] published: 2026-08-31 updated: 2026-10-31 articles: 58 --- # On Deck > Ground Truth follows AI as it's actually built, funded, and shipped: consumer assistants, vertical products from legal and healthcare to agriculture, and the capital moving behind them. It's written for practitioners who've shipped agentic systems and want more than press releases. It reads interaction models, architectures, trust patterns, and funding logic to judge which companies are serious, and where it's worth building next. [Read on the website]() ## Issue 9 We've been specifying build artifacts for five cycles. None have shipped. That mattered less when signals were slow. Stripe acquired OpenRouter last week, turning model routing into a trust-design problem their Senior Staff hire will own: the system changes capability mid-task without changing what the user agreed to. Juno has the production depth to make that argument. The visual proof isn't on her site. This issue tracks every signal on two clocks: how fast the posting ages, and how fast her advantage erodes. On several, those clocks just synchronized. ### [Signal Watch]() Your search signals, informed by landscape intelligence - [Two Clocks Are Running on Every Signal in Your Queue]() — By [Reva Sandoval]() · Published 2026-08-29 · Format: Priority Action Queue Two new companies entered the queue this week, Gray Swan AI and Within, and both have candidate-advantage windows shorter than their posting age suggests. Vanta moves to Act Now: the unfilled Head of Design has been open 54 days and creates urgency the four-day-old Director posting doesn't explain on its own. This issue introduces the two-clocks model — posting age and candidate-advantage decay, ranked together — and applies it across seven signals. Three need outreach before September 4. - [Five Act Signals and Four Expiring Windows Across Nine Recognized Targets]() — By [Reva Sandoval]() · Published 2026-08-29 · Format: The 100 List Five Act signals this cycle across OpenAI, Stripe (three roles), Headway, and Vanta, which jumped straight to Act on a four-day-old Director posting. Four windows expire before Friday: Midjourney and Gusto cross into Ignore-until-trigger by midweek, Stripe Risk and Link lose hot status September 1, and Headway's provider cluster starts decaying in six days. Nine entries in three clusters, grouped by what makes each urgent now: agent supervision surfaces where your forward-looking artifacts separate you furthest from the field, two org transformation mandates that screen on something other than craft, and the expiring windows that need a keep-or-release call this week. - [Twenty-Five Companies, Four Act Verdicts, and a Wider Window Than You Think]() — By [Reva Sandoval]() · Published 2026-08-29 · Format: The 100 List Four Act verdicts this cycle, none of them Anthropic or OpenAI. Replit posted a Director of Product Design three days ago at $275K–$350K, $9B pre-IPO. Vanta posted the same week with the design function expanding. AlphaSense is hiring a Director for AI Experience at $7.5B valuation and $600M ARR. Superhuman is building education agents from zero. Twenty-five companies total, scored and decay-checked, with specific portfolio matches and outreach angles for each — and because these sit below the sightline of most senior candidates, your outreach lands in a thinner pile. - [Signal Watch — The 100 List — Non-Bay Area Companies with Confirmed Signals]() — By [Reva Sandoval]() · Published 2026-08-29 · Format: The 100 List Nine companies cleared the rubric outside the Bay Area this cycle. Two warrant outreach this week: Gray Swan AI (AI safety evaluation, remote-US, role score 13/15) and Replicant (AI voice agents, remote-US, role score 14/15 — posted yesterday). Headway's scores reach Act on paper, but the anchoring signal is 26 days old and remote eligibility is unconfirmed. The rest carry blocking questions. The non-Bay Area signal set remains thin at Director+ and Staff IC level — these nine are the field. - [The 100 List — Recognized Companies With No Active Signal]() — By [Reva Sandoval]() · Published 2026-08-29 · Format: The 100 List Brex is gone. Ramp aged to 82 days. Anthropic still hasn't posted a qualifying design role. One hundred recognized companies sit below the action threshold right now, but eight score 10+/12 on the company rubric — and when one of them posts a Staff+ product design seat, the window will be brief and the pipeline will flood within days. Every entry carries a named trigger, a pre-loaded portfolio match, and a tier assignment so you can move the same day one fires. Scan the highest-ceiling companies weekly. - [100 Dormant Companies, Filed by the Design Problem You'd Solve]() — By [Reva Sandoval]() · Published 2026-08-29 · Format: The 100 List 100 companies scoring 8+ on the company rubric, zero qualifying design signals this cycle. Each names the specific trigger — a Staff+ posting, a design leadership appointment, a redefined requisition — that would upgrade it to Watch or Act. Clusters are organized by design problem, not industry, so when a trigger fires you already know which proof layer to lead with. Eight companies score 11–12: Hippocratic AI, Together AI, Fireworks AI, Cursor, Harvey, Gamma, ElevenLabs, and Cognition. Set career-page alerts on those first. - [Signal Watch — The 100 List: 100 Non-Bay Area Companies, Filed by Trigger]() — By [Reva Sandoval]() · Published 2026-08-29 · Format: The 100 List 100 non-Bay Area companies scored against the full rubric, each with a named trigger condition and a company-specific portfolio match. None have a qualifying signal this cycle. Seven migrated to Active before publication; the rest are filed here so that when a name surfaces — recruiter message, funding round, LinkedIn post — you find the entry instead of starting from scratch. Rubric scores show how close each company is to qualifying. The trigger tells you what to watch for. The portfolio match tells you which proof to lead with when the window opens. - **Vanta Director Opens Beneath Stalled Head Search** — [Source]() — Published 2026-08-29 Director of Product Design posted August 25, 4 days old. $275K–$323K cash plus equity. The 54-day-old Head of Design backfill above it remains unfilled, leaving the reporting relationship undefined. Reach Jeremy Epling before both mandates crystallize. Tier Playbook: AI-native. - **Gray Swan Staff Designer Fits Trust Thesis** — [Source]() — Published 2026-08-29 Staff Product Designer posted August 26, 3 days old. $205K–$255K cash plus equity. Series A ($40M, Wing/Madrona). The role designs AI evaluation, monitoring, and security workflows. Direct trust-thesis alignment. Constraints: live coding screen, Series A stage risk, existing UX Lead in place. Tier Playbook: AI-native. - **Render Agent Experience Crosses Into Watch Band** — [Source]() — Published 2026-08-29 Staff Product Designer, Agent Experience now 22 days old, first day of Watch band. The stated August 28 application close has passed. Career page still lists the role, but the candidate pool is likely already assembled. Upgraded to Act last cycle. - **Aurora Senior Staff Role Gone From ATS** — [Source]() — Published 2026-08-29 Senior Staff Product Designer, Customer Experience no longer visible in Aurora's Ashby feed as of August 29. Could mean filled, cancelled, or quietly pulled. No confirmation either way. Stop outreach unless a new signal appears. - **Airwallex T0 Experience Fading From Aggregator Indexes** — [Source]() — Published 2026-08-29 Staff Product Designer, T0 Experience now 25 days old. The role no longer surfaces in current aggregator indexes; only the founding-designer and Design Systems roles appear. ATS removal inferred, not confirmed. Compensation figures from prior cycles remain unverified. - **Gusto Head of Design Hits Ignore September Third** — [Source]() — Published 2026-08-29 Head of Design, Unified Service Platform posted July 22, now 38 days old. Watch-to-Ignore threshold hits September 3. The mandate's ownership across design, product, and AI/MLE remains unsettled. A direct conversation with Amy Thibodeau before that date keeps the channel open. - **Stripe Risk Role Exits Hot Window September First** — [Source]() — Published 2026-08-29 Staff Product Designer, Risk posted August 10, now 19 days old. Hot window closes September 1. The August 19 OpenRouter acquisition may have shifted internal hiring priorities independent of posting age. Verify seat status before any new outreach. - **Midjourney Design Lead Approaching Ignore September Fifth** — [Source]() — Published 2026-08-29 Design Lead posted July 24, now 36 days old. $285K–$350K cash, no equity stated. Reports to CDO Banu Guler. Watch-to-Ignore threshold hits September 5. The unaddressed rights and provenance gap in your portfolio evidence is the bigger blocker here than posting age. ### [Horizontal AI]() AI tools, assistants, and platforms worth understanding - [Canva Brush Edits in Place Without Showing What It Left Alone]() — By [Reva Sandoval]() · Published 2026-10-30 · Format: Centerpiece Teardown Every one-change request carries an unspoken second one: leave the rest alone, because the rest was approved. This week Canva shipped Brush, a model it post-trained to edit designs by writing and running code against their elements. With that architecture, Brush should already know which elements it changed, but Canva doesn't show you. Edits apply automatically, and history just says "Canva AI." The omission is strategic too. Canva built its own editor to own revision while assistants take first drafts, and a self-reporting edit is the thing they can't copy. - [Runway App Studio turns every shared app into a Runway signup]() — By [Reva Sandoval]() · Published 2026-10-30 Runway's [App Studio](https://runway.com/news/company-news/introducing-runway-app-studio), announced October 26, turns a creator's Workflow into a hosted app. A Workflow is a saved chain of Runway's image, video and audio models. In the app, the creator locks the prompts, models and settings and exposes a few inputs as a form. - [Voice AI now does three separate jobs, and only one of them is new]() — By [Reva Sandoval]() · Published 2026-10-30 Voice in AI products now does three distinct jobs. It **records** what other people say, as meeting notetakers do. It takes **input** you mean to send, which is dictation. And it can now **conduct**: one spoken session that directs agent work, and that work keeps running after you stop talking. - **Copilot now scopes work around a goal** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 A project bundles repos, Figma files and Slack threads around one objective. Copilot runs as many agents as it needs and returns pull requests. Desktop computer use, in preview, reaches apps without APIs behind per-app approval and an org kill switch. Unexplained: how reviewers inspect desktop actions next to code diffs. - **Opera hosts local models for websites to call** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Opera One ships Gemma, plus Apple's on-device model on Macs, for sites and extensions to call through WebAI. That standard is only a proposal, and Google and Microsoft support is Opera's claim. Opt-in, with per-site permissions. The browser as model broker is new. No quality data; your hardware sets speed. - **Google puts a Gemini agent inside Genie** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Google says Gemini 3.8 Flash watches generated frames, splits a goal into steps and checks each step while steering the world model. That agent-in-the-loop design is the part to track. Trusted testers only. Google concedes coherence holds only over short sessions, and nobody outside Google has reported using it. - **Notion agents run longer, and owners pay** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Notion's sixth harness rebuild adds queue-backed workers and a dedicated computer per agent; Notion says tasks can run up to 24 hours. From November 2, shared Custom Agents draw on the owner's 500 monthly Select credits and pause when those run out. Notion's help docs contradict this for older agents. - **Meta's AI pet contradicts its own listing** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 A Tamagotchi with a journal: you describe your day and the pet retells it in its own voice. The framing conflicts on age and streaks: the listing requires users to be 18+ and promises 'no streaks,' then uses streak language anyway. Live in the Philippines since October 8. Moderate confidence it tests ideas for Meta's Muse device. ### [Vertical AI]() How AI is reshaping industries, company by company - [Shopify Lets Shoppers' Agents Check Out, and Brand Agents Lose the Pre-Purchase Conversation]() — By [Reva Sandoval]() · Published 2026-10-30 · Format: Vertical Competitive Read Merchants pay Siena, Gorgias, Intercom Fin and Sierra to run selling agents in web chat and social DMs. Shopify's Fall '26 developer release, out October 28, lets the shopper's own agent search a store, build a cart and check out instead, with the shopper approving. That agent never has to talk to the brand's, and all four vendors bill usage by conversation or contracted outcome. - [Sort Legal AI by the Data Each Company Controls]() — By [Reva Sandoval]() · Published 2026-10-30 Every major legal AI vendor now sells an agent that breaks legal work into steps and picks a model for each one. Since everyone ships it, the agent tells you little. Look at the data each company controls outright. That data predicts both the moat and the buyer. - **Everlaw lets outside AI tools into its evidence** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Omni, Everlaw's new agent, answers questions across its e-discovery tools. Taking actions inside the platform is still on the roadmap. The CEO said customers care more about plugging outside AI tools in under Everlaw's permissions and audit trail. If that ships, Harvey and Legora would query Everlaw's evidence instead of displacing it. - **AMA framework shifts clinical AI liability toward vendors** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 The AMA's October 29 framework assigns responsibility by each party's authority, information and ability to prevent harm. It says physician oversight cannot make up for weak evidence or poor system design. It is a policy proposal, not law. Vendors whose safety answer is "a clinician reviews it" should expect buyers to ask for evidence instead. - **Focus Financial now runs overlapping advisor AI tools** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Wealth-management group Focus added Nevis for account opening and onboarding, nine months into rolling out Jump to 800 advisors. Jump is piloting account opening too, and Zocks shipped a Schwab-linked version on October 28. Meeting notetakers are converging on one workflow, so expect large firms to pick one. Jump's 70% time savings is self-reported from 40 advisors. - **Investors price retrofit excavator autonomy at $4.1B** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Bloomberg reports a $600M round for Bedrock, which retrofits existing excavators to dig autonomously on Sundt and Zachry jobs. A site manager sets the plan, and machines stop on faults or nearby obstacles. Honeywell's Forge Cognition industrial agent launched with full autonomy switched off. Bedrock has published no independent productivity or safety data. - **Verisk pipes ISO rating data into general assistants** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Verisk is putting ISO loss costs, the rating data carriers use to price policies, plus premium-audit content into Microsoft Copilot and Amazon Quick. It is also adding property-level hail and wind history in Claude, gated by customer permissions. Like Everlaw, the data owner chose distribution over its own assistant. No underwriting results yet; otherwise a quiet fortnight. - **Brand visibility tools for AI assistants attract funding** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Apart from Shopify's agent-commerce release, covered in the lead feature, little moved structurally. One separate layer is drawing money: tools that shape how brands appear in AI search and shopping assistants. Azoma raised $11.2M in October, after Profound's $180M Series D in September. Moderate confidence, since the pattern comes from one trade-press roundup. - **Kubota's autonomous orchard tractor is still a pilot** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Kubota extended its Agtonomy pilot to the M5 Low Profile tractor for spraying, mowing and underplant cultivation, sold with dealer support; pre-orders deliver in 2027. Press called it a launch, Kubota calls it a pilot. The incumbent holds the channel, the startup supplies software. BloomX's 20% yield uplift remains a manufacturer claim with unpublished trial methods. - **Ravical targets firms as finance automation architectures split** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Ravical sells to accounting firms, encoding senior practitioners' methods into shared preparation, review and advisory workflows. It reports 100+ European clients, with specialists still signing off. Corporate finance agents are splitting along a different line: Hyphenate layers onto existing ERP systems, while Rillet replaces the general ledger outright. - **Stanford Maps Where Lawyers Supervise Legora** — [Source]() — Published 2026-10-30 Maps where lawyers intervene in Legora workflows; fictional documents can't show whether reviewers catch errors. - **ACCESS2 Trial Tests Autonomous AI Eye Exams** — [Source]() — Published 2026-10-30 Screening completion hit 98–99.5% across racial groups; it measures access, not whether vision improved. - **Randomized Trial Compares DAX and Nabla Scribes** — [Source]() — Published 2026-10-30 Nabla trimmed note time 9.5% while DAX didn't; versions are dated and quality unmeasured. - **Shopify Documents Agent Checkout for Developers** — [Source]() — Published 2026-10-30 Specifies the agent's tool calls and buyer approval step; offers no completion or fraud data. - **AMA Lays Out Clinical AI Liability** — [Source]() — Published 2026-10-30 Shows how physicians want AI liability divided, which shapes review design; it's advocacy, not law. - **Reporter Watches Everlaw Omni Pick Tools** — [Source]() — Published 2026-10-30 A reporter watched Omni choose tools on litigation data, but Everlaw ran the demo. ### [AI Capital]() Where money flows and what it signals - [Guidewire Paid $1.7 Billion for the Pricing Models Underwriting Agents Will Call]() — By [Reva Sandoval]() · Published 2026-10-30 · Format: Anchor Deal Breakdown On October 27, Guidewire, whose software runs policies and claims for much of property-and-casualty insurance, agreed to pay $1.7 billion in cash, plus up to $400 million on revenue targets, for hyperexponential. The London company holds specialty insurers' pricing models as versioned, actuary-approved code. Its agents are mostly still in preview, but the models are live at Aviva and others, and an underwriting agent can't quote without one. - [Deepgram Is Buying the Framework Its Competitors Plug Into]() — By [Reva Sandoval]() · Published 2026-10-30 · Format: Two Deals, One Lens Deepgram is acquiring Daily, maker of Pipecat, the open-source framework many voice agents run on. Pipecat lists Deepgram's speech models beside ElevenLabs, Cartesia, and AssemblyAI, so the new owner will sell inside a framework its own employees maintain. Deepgram promises an independent foundation and continued model choice. The foundation has no published terms, and the deal has no closing date. This piece covers why Deepgram took the whole operating company and which public signals will test its promises. - [Apple Lined Up Hiring Rights and Patents From Ascend.io While Its Service Winds Down]() — By [Reva Sandoval]() · Published 2026-10-30 · Format: Two Deals, One Lens An EU register entry shows Apple lined up hiring rights for selected Ascend.io staff, a non-exclusive license to its IP, and certain patents. It describes no share purchase. Ascend's data-pipeline service is winding down. Apple notified the Commission June 30, and the arrangement became public October 30. It's at least Apple's fifth deal of this shape. Set beside Deepgram–Daily, it points to one sorting question: after closing, who outside the buyer still relies on the asset? - [What Cohere's and Casap's strategic investors bought]() — By [Reva Sandoval]() · Published 2026-10-30 A strategic investor puts in money and wants something beyond a return: a vendor, a sales channel, a policy result. These two rounds sit at opposite ends of that trade. - [FluxCap's Investors Point to Plaintiffs and Enforcers as Its Buyers]() — By [Reva Sandoval]() · Published 2026-10-30 FluxCap, an AI tool for antitrust litigation, raised a $6.4M seed with no named lead. YC and Bloomberg Beta back seed companies in every category. To read this round, look at the two investors that don't: **Burford Capital and Omidyar Network.** - **Serval hires 100 implementers to unseat ServiceNow** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Sequoia led $150M at a $3B valuation, Serval's third round in a year. Part of it funds a services arm of 100-plus Automation Engineers, half of them recent graduates on four-week training. Pulling customers off ServiceNow still requires people on site. Ask who maintains the workflows once they rotate off. - **Collective's General Catalyst facility outweighs its equity round** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 General Catalyst co-led the $40M Series B and added $60M from its Customer Value Fund. Compared with Otter: Otter sells software, while Collective performs filings and payroll for every customer. Service cost per cohort decides whether the marketing spend pays back. The 127% growth figure is company-reported. Facility terms are undisclosed. - **Hummingbird buys Quantimon for models that learn from cases** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Its second acquisition after LogicLoop, terms undisclosed. Money-laundering detection models improve with labeled investigation outcomes, and Hummingbird's case software is where bank analysts record those outcomes. Moderate confidence that labels are the target. The risk is consistency: 'resolved' means different things at different banks. - **Ravical bets on the accountants Collective routes around** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 MMC led a $17M seed on the thesis that AI should let accounting firms take on more clients without adding staff. Over 100 European firms reportedly use it. Set it beside Collective: similar small-business tax work, sold through the firms Collective bypasses. Paying customers and retention are undisclosed, so read 100 as reach. - **Vera Health's Series A prices reach, not revenue** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Threshold led $28.5M, which closed months before the announcement. The only disclosed metric is free use: 800,000 clinicians in 190 countries. Funds go to infrastructure and partnerships with medical societies, publishers and health systems. Investors are betting individual use converts into institutional Vera Spaces contracts. No conversion data yet. - **Wabi priced at $1.25B a month after pivoting** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 IVP led $55M 31 days after Wabi 2.0 swapped prompt-built mini-apps for a messenger with separate, shareable agent threads per goal. The timing suggests investors paid for that design against single-thread assistants like Muse. The app is free, has no revenue model, and barely specifies permissions for shared personal data. - **Circuit Mind folds hidden seed into $25M headline** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 The circuit-board design startup's total combines Norwest's $18.5M Series A with a previously unannounced Third Prime seed. Its agents run on a patented physics engine instead of a general-purpose LLM. Los Alamos documented four hours against an estimated 60–80, but that and most cited results describe the original design tool, not the new schematic-checking agents. - **Salesforce buys Rune to reach troops offline** — [Source]() — By [Reva Sandoval]() — Published 2026-10-30 Rune, founded in 2024 by Anduril alumni, runs with Army and Marine units, forecasting supply needs offline and syncing decisions when connectivity returns. Missionforce, Salesforce's defense unit, already handled records and headquarters planning. Rune extends it into field operations. Price undisclosed, and 'no custom integration' is still a stated goal. - **Guidewire's 8-K on Buying hyperexponential** — [Source]() — Published 2026-10-30 The filing's terms: $1.7B cash, up to $400M earnout, and a $1.0B term loan. - **Canada's Official Terms for Cohere Stake** — [Source]() — Published 2026-10-30 Up to $1.6B in equity, conditioned on 1,000 Canadian jobs and domestic compute by 2030. - **Cohere's Series E Release, CAD $7B Target** — [Source]() — Published 2026-10-30 Schwarz-led round with plans for its STACKIT cloud; the 165% ARR growth is self-reported. - **Apple Disclosed Ascend.io Deal to EU** — [Source]() — Published 2026-10-30 Paywalled, but the free excerpt confirms a June 30 notice framed as licensing and hiring. - **Deepgram Buys Daily, Pipecat Stays Open** — [Source]() — Published 2026-10-30 Company post promising a foundation for Pipecat, with no price, closing date, or governance details. - **Former DOJ Antitrust Leaders Launch FluxCap** — [Source]() — Published 2026-10-30 Launch release: $6.4M seed from investors including YC and litigation funder Burford. ## More - [Publication guide]() - [Issue archive]()