# Greg Halvorsen, Head of Product (AI Platform) at Vertex Supply — read of Company Brain, June 26 2026 > 9 years building B2B SaaS, 18 months trying to ship actual AI features that don't embarrass me in front of the VP of Sales. Stack is Postgres, Notion, Salesforce, Slack, OpenAI API, and a growing pile of Python scripts I keep promising to clean up. --- ## How I got here Someone dropped this link in the #ai-builders channel of a Slack community I'm in. The comment was just "is this actually real?" No endorsement, no context. That framing put me on guard before I even clicked. I've been burned by three vendors in this space in the last six months, two of whom had very similar hero sections. I gave it five minutes because the problem they're solving is genuinely my problem right now. ## What I clicked first The sub-headline under the hero pulled me in more than the headline did. "Unified knowledge platform that makes company data queryable by Claude agents" is at least a real sentence. I know what that means. Compare that to the headline "Your Company's Knowledge, Always Ready" which could describe a SharePoint upgrade from 2019. Then I read this: "Most AI agents fail not because they're not smart enough, but because they lack context. Every query, every decision, every workflow starts from zero." That is the exact frustration I described to my CTO four weeks ago, word for word. I kept reading. ## Where I paused The "Access Control That Actually Works" section. Not because it convinced me but because that name is doing a lot of lifting. "Row-level permissions flow from your existing identity systems. Agents respect the same access boundaries your employees do." I've been told this before and it was never actually true. The gap between "agents respect permissions" and the actual implementation reality of mapping Salesforce roles through a knowledge graph and then enforcing that at query time is not a copywriting problem, it's a multi-month engineering problem. This section wanted me to skim past it. I didn't. ## What I distrusted Three things, in order of how much they bothered me. First: "Customer 1. Customer 2. Customer 3. Customer 4." Those are the actual customer logos on the page. Not blurred logos. Not "names withheld." Literal placeholder text. That stopped me cold. Whatever social proof this page was trying to build ended there. Second: "Reduce support costs by 40%." Where is that from? Internal data? One customer? A Gartner report they're extrapolating from? It's sitting there with no footnote, no asterisk, no "based on a pilot with X." That number is load-bearing in their pitch and they gave it zero support. Third, and this is the big one: I scrolled to the bottom and found this buried under the pricing section. "Honest disclosure: we don't have live customers on this idea yet. We shipped the strategy package; you ship the customer conversations." This is not a product. This is a product idea for sale from something called the Wishdeal Factory. The whole page is a concept pitch dressed as a launch. The "Trusted by Teams Who Need Agents They Can Rely On" section is above this disclosure. That's a trust problem, not just a marketing problem. ## What would convince me If this were a real product: one case study from a company in a regulated industry (healthcare, financial services) that explains specifically how the permission sync works with their IdP, what the integration took, and what broke during rollout. Not a quote from a VP of Engineering saying it "transformed our workflow." The actual eng lead's account of what was hard. The compliance claims (HIPAA, SOC 2, GDPR) need more than bullet points. A link to their security page with actual attestations or at minimum a "here's our SOC 2 Type II report, request it here" form. And I'd want to see one workflow end to end. Not a diagram. A screen recording of an agent querying something it shouldn't have access to and getting blocked. ## What I'd ask in an email reply 1. The page says "no ETL pipelines, no manual data movement" for ingesting from Salesforce, Jira, Slack, and databases. What is that actually built on? Are you wrapping something like Airbyte or Fivetran, or did you build connectors from scratch? Because that claim is either very true or very misleading and I need to know which. 2. The permission model: if I connect our Salesforce and a specific user doesn't have access to Opportunity data above $500K, does the agent actually fail gracefully, or does it just not surface those records while potentially leaking aggregate patterns? How was this tested? 3. This page looks like a product but the footer says this is a Wishdeal Factory strategy package. Are you the founders who are going to build this, or are you selling the plan to someone else to build? --- ## Verdict: dismissive Not because the problem framing is bad. It's actually one of the clearer articulations of the agent grounding problem I've seen on a homepage. But this isn't a product, it's an idea with a price tag attached to the blueprint. The placeholder customer logos and the buried "no live customers yet" disclosure undercut everything above them. I'm not curious enough to pay $5 for a go-to-market doc on something that hasn't been built. --- *Memo by skeptic persona, generated 2026-06-26. Studio breaks own self-grading loop.*