# Marcus Ruiz, Director of AI Automation at Celerity Systems — read of Company Brain, 2026-06-26

> 11 years building internal tooling and ops pipelines at B2B SaaS companies, currently managing 4 engineers trying to get 3 AI agents into production without them hallucinating our own pricing.

## How I got here

Googled "rag for enterprise agents without custom etl" after spending six weeks debugging a LangChain retrieval pipeline that kept returning stale policy docs to our support agent. Got a link in the second page of results. No LinkedIn ad, no referral, nobody sent this to me. That's relevant context.

## What I clicked first

The subheader problem statement landed: "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's the exact sentence I would have written to my VP last month to explain why our pilot isn't in production yet. So I kept reading.

## Where I paused

The access control section. "Row-level permissions flow from your existing identity systems. Agents respect the same access boundaries your employees do." If this is true and actually works, that is the hardest part of what we're trying to build. Not the retrieval. Not the chunking. The permissions. We have 14 different permission groups in Salesforce and our current approach is to not give agents access to Salesforce at all because it's too risky. I stopped and read that paragraph twice.

## What I distrusted

Three things, and one of them is a dealbreaker.

First: "Customer 1 / Customer 2 / Customer 3 / Customer 4." Those are placeholder logo boxes. No names, no logos, no quotes. The section is called "Trusted by Teams Who Need Agents They Can Rely On" and it's literally Lorem Ipsum for social proof. I would not put that in front of a real buyer.

Second: "Reduce support costs by 40%." No source, no customer name, no timeframe. That number came from somewhere and I don't know where.

Third, and this is the one that ended it for me: at the bottom of the page there is a section called "The Wishdeal Factory" that gives this idea a score of 68/100 and says, in plain text: "Honest disclosure: we don't have live customers on this idea yet. We shipped the strategy package; you ship the customer conversations." So this isn't a product. It's a business idea someone packaged up and is selling research decks about for $5 to $199. The pricing table, the enterprise tier, the "Start Free Trial" button -- none of that is real. The page is a concept dressed up as a launched product.

## What would convince me

If this were a real product, here is what I'd need: one named customer (doesn't have to be famous, just a real company I can look up) explaining how they wired in their Salesforce permissions and what broke before it worked. That story, in their words, would tell me more than five feature bullets. Specifically I want to know if "row-level permissions flow from your existing identity systems" works with Salesforce permission sets or just object-level roles, because that distinction is the entire problem.

## What I'd ask in an email reply

1. The permissions claim is specific enough to be testable -- can you walk me through how you handle Salesforce permission sets where the same user has different field-level access on different objects?
2. "No ETL pipelines. No manual data movement." -- what does the initial ingestion actually look like for a 5TB Snowflake warehouse? What runs, where, and who owns it?
3. The compliance bullets mention HIPAA and SOC 2 -- are those certifications in hand, or is that a roadmap item?

## Verdict: dismissive

The problem statement is real and the feature list is articulate, but the product is not a product. It is a research report about a product that could exist. Asking me to "Start Free Trial" when there are no customers and no live build is the wrong move; it creates distrust that would take three good references to undo.

---
*Memo by skeptic persona, generated 2026-06-26. Studio breaks own self-grading loop.*