--- title: "Issue archive" canonical: "https://anchor.is/on-deck-vvSq/issues" publication: "On Deck" publisher: "Ground Truth" language: "en-US" guide: "https://anchor.is/on-deck-vvSq/llms.txt" issues: 18 --- # 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 archive - [Issue 18]() — Published 2026-11-02: We kept getting stuck on one small moment: the tap where a person approves a purchase. Shopify is opening its checkouts to shoppers who arrive with their own AI agent, and the brand that just paid for a selling agent of its own may never get a word in. So who approved what, and who can prove it afterward? That question kept turning up all over this issue, right down to the design roles worth chasing. You've shipped answers to it before. Use that. - [Issue 17]() — Published 2026-10-26: We got one thing wrong, and it changed how we read everything else in here. We'd had Brex down as a startup, with startup equity. Capital One closed on it back in April, so any stock in a Brex offer is bank stock now, and our tracker never caught it. Once we started asking who actually owns a company, the question kept turning up, right down to a radiology AI marketplace on its third owner. Before you weigh any role this week, find out who issues the shares. - [Issue 16]() — Published 2026-10-19: We got something this issue we've never had before: sworn operating data for a consumer AI company at scale. Character.AI's bankruptcy filing shows $5.3M in monthly revenue against $7M in compute, with 12 million users averaging two hours a day. The engagement was real. The business wasn't. Two sequential deals extracted the founders, the researchers, and the model rights. What remained was the users, the safety obligations, the litigation, and a compute bill the revenue couldn't cover. We're calling that a remainder company. Any consumer AI startup whose talent and IP can be separated from its user obligations is building toward the same outcome. - [Issue 15]() — Published 2026-10-12: Clio just bought a judiciary workflow platform, which means one company now touches both sides of a court filing. That's a specific bet about who controls legal work after the AI runs. We kept finding that same question everywhere this issue. Six personal agent products promise to work while you sleep, and the real cost turns out to be how many minutes you spend checking what they did. Deal structures reveal which pieces of a product the buyer thinks can be separated. The thread: everyone is shipping agents. Almost nobody has priced what it costs to be responsible for them. - [Issue 14]() — Published 2026-10-05: Databricks posted a Staff Product Designer for Agentic Coding four days ago. The product is documented well enough to map the design problem before you write a word of outreach, which almost never happens at this stage. Your Trust essay maps to their agent supervision mandate. Where you're short is developer-product proof, and a conversation will answer that faster than another build cycle. We found the same question running through fifteen roles this cycle: what's actually between you and each one? This issue names the gate for every entry. For Databricks, there isn't one. - [Issue 13]() — Published 2026-09-28: Render's Head of Design publicly asked for portfolios this cycle for a Staff Agent Experience role. The Trust essay maps directly. The engagement package is ready to send. But this issue's counter-narratives work changed the order of operations: objections that form at portfolio inspection happen before any conversation, so no verbal bridge can reach them. Right now the strongest AI evidence on junochen.com is either broken or buried. We built four engagement packages this cycle. Fix the evidence trail before sending any of them. - [Issue 12]() — Published 2026-09-21: We pulled this issue together expecting the AI-design roles on the board to sort by company type. They don't. OpenAI's growth team and Handshake's recruiting team both write "trust" in their postings, and they mean completely different things. One is afraid users won't come back. The other is afraid an agent will derail someone's career. The proof that wins one is irrelevant to the other. This issue is organized around diagnosing which fear you're actually walking into before you decide what to send. - [Issue 11]() — Published 2026-09-14: Handshake's product docs describe the moment their AI stops suggesting and starts acting on behalf of users. They posted a Staff Designer role to design that transition, and OpenAI posted an Identity role in the same territory. We built engagement packages for both. But the thing that kept surfacing as we worked through this issue: the companies are now hiring for the exact problem you've been specifying a portfolio artifact around, and the artifact doesn't exist. Six cycles of refinement, nothing shipped. This issue treats that gap as a cost. - [Issue 10]() — Published 2026-09-07: Slack posted a role this week that reads like someone wrote it from your Trust essay. Agentic handoffs, multiparty authority, designing when neither human nor agent owns the whole workflow. The buyer is named, the window is fresh, and the strongest artifact you could send him hasn't been built yet. We kept hitting that same wall. Three new mandates landed this cycle, and the constraint on every one is identical: your published evidence hasn't caught up to what you actually know how to do. This issue is about what to ship first, what to send now, and which conversations can start before the portfolio catches up. - [Issue 9]() — Published 2026-08-31: 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. - [Issue 8]() — Published 2026-08-24: We should have built this earlier. Capital One posted a design leadership role in May, let it run three months, then renewed it under a fresh requisition number last Thursday. That sequence means one of three things, and two of them change your whole approach. It's a small example of a larger problem: every company in your pipeline says "AI design," and none have been pressed to prove what they mean. You've been providing evidence while they provide adjectives. This issue reverses that. - [Issue 7]() — Published 2026-08-10: We mapped every objection you'll face this cycle. The answers came together. The boundaries took longer: where your evidence genuinely ends and you need to say so before the room finds the edge for you. One company card in here, Fieldguide, we built to open with the comp gate that might kill the opportunity before we wrote a word of positioning. That felt wrong while building it. Turns out it's the most useful thing in the issue. Every brief and dossier in here carries the same discipline: here is where your proof stops, and here is what you do there. - [Issue 6]() — Published 2026-08-03: We added a field to every role on the board: how to enter. The reason is a checkbox on the Gusto application. It asks a management-scale question whose honest answer needs a paragraph, not a yes or no. That form will filter you before anyone reads the portfolio. Most senior applications work this way: the screening question is the simplest version of something that deserves a conversation. Every listing now carries a route. Find the decision-maker first, apply directly, or call to resolve an unknown before you commit. - [Issue 5]() — Published 2026-07-27: We found something this cycle that reframes three listings on the board. Brex's careers page now identifies it as a wholly owned subsidiary of Capital One. The equity thesis doesn't hold. The role itself, designing trust and supervision for AI that touches money, is still one of the strongest mandate matches in the scan. That kind of split kept surfacing: what a company calls itself versus what's true underneath. So we scored every role by the design problem, not the brand. The portfolio section turns the same lens inward. Your surfaces have trust failures that contradict the essay you're leading with. Fix those first. - [Issue 4]() — Published 2026-07-20: We read eight live postings this cycle and kept finding the same question underneath different language. Amplitude wants someone to "define how agents and humans work together in the same workflows." Gusto wants someone for "AI wrongness" and "exception movement." Completely different industries, same fear: who designs the moment a person decides whether to trust what an agent just did. You have that answer. What you don't yet have is packaging that lets a buyer see it in 30 seconds. That's what this issue is for. - [Issue 3]() — Published 2026-07-12: We pulled apart eight target companies this cycle. Every one is hiring for the same problem it can't quite name: what happens when the system acts and a person has to decide whether to trust it. Ramp's posting calls it "designing memory." Headway wants it solved across clinical documentation and billing. You've been building for that moment since Thermo Fisher. The positioning is the sharpest it's been. One thing in the way: your site serves a private essay in full HTML before the password check runs. Close that, then move. - [Issue 2]() — Published 2026-07-06: We swept every design leadership posting across your targets this cycle for one signal: does the company name who's accountable when the AI is wrong? Ramp does, describing agents handling $200B in spend and asking who verifies. Headway names clinical consequence when AI drafts a provider's session notes. Most companies don't. That gap changes your whole approach. It's the difference between answering a question the buyer already has and teaching them to ask it. The proof sequence is different for each. Battlecards are ready if you need to move today. - [Issue 1]() — Published 2026-06-30: We pulled apart eight live signals this week and kept landing on the same word in the job copy: trust. Showing up as operational requirements, the kind you get evaluated against. Gusto's posting asks candidates to design what happens when AI gets something wrong. That is your Thermo Fisher case restated in their language. Capital One wants someone to change how 200 designers work with AI. Suno needs a design leader while an artist-rights controversy defines the product's trust problem in public. All three are live, all three map to work already published at junochen.com, and a message sent before the July 4 weekend lands in a thinner inbox. ## More - [Publication guide]() - [Issue archive]()