[News Brief] How the AI Center of Excellence can help the Business Units
Office Hours Recap: How We Automated the research for the $100M sales pipeline, Center of Excellence new initiatives aimed at Business Units' assistance; plus the rise of markdown programming
Office Hours returned this week with a clear message: we are moving from theory to operational reality. This session showcased a live, complex orchestration built in collaboration with the Sales team, proving that AI agents can handle the critical “grunt work” of account management at scale.
Here is the breakdown of the live demos, the new content initiatives, and the technical horizon discussed in the session.
Feature Spotlight: Sales Pipeline Orchestration
The centerpiece of the session was a live demonstration by David Proctor of a new “Sales Bot” ecosystem built on Braintrust, our internal AI platform.
The challenge brought by Shannon Ramsey (SVP of Sales) was clear: How do a handful of people manage relevance, risk, and contracts for a $100M book of business with thousands of customers?. Manual review is impossible at that scale, especially during M&A intake when non-standard contracts flood the system.
The Solution: The Agent Squad
David demonstrated how a single prompt in Google Chat can trigger a squad of specialized agents to execute complex workflows.
Prospector: Enriches company data, flags risks (like layoffs or leadership changes), and monitors news.
Red Line: The contract intelligence agent. It uses RAG and Graph RAG to parse legacy PDFs and perform semantic searches for specific clauses (e.g., “uncapped liability”).
Forge: The creative engine that generates Google Sheets, Docs, and Gamma presentations on the fly.
The Algorithm: Generates and schedules high-engagement posts for LinkedIn/X.
The Demo
In real-time, the system was asked to: “Identify top 5 clients based on ARR, check contracts for risk, create a tracking sheet, draft an executive slide deck, and schedule a meeting”.
Within minutes - and asynchronously - Braintrust orchestrated the agents to:
Enrich the client data via Prospector.
Audit contracts via Red Line (flagging a geographic scope error).
Generate a strategic Google Doc and a Gamma slide deck via Forge.
Draft social media posts via The Algorithm.
Send the calendar invite via Tempo.
“This changes everything. If we can take an acquisition, bring in all the data, and have a sales team execute with speed and relevance... it makes the scaling issue a non-event.” - Shannon Ramsey
New Initiative: Three Questions with CoE
Stan Huseletov introduced a new micro-interview series: Three Questions with CoE.
This series features 15-minute, no-fluff conversations with internal Trilogy experts to extract their “spiky points of view”. The latest episode features Kathy Slowinski, who dissects the state of the industry with a memorable comparison:
[Kathy Slowinski] really digs into influencers... She calls Trilogy ‘The Jetsons’ and the rest of the industry ‘The Flintstones’.
The articles go live every Monday with shorts and videos attached. The aim is to showcase internal expertise without the bloat of standard corporate podcasts. Subscribe to not miss the Kathy-episode.
Technical Teaser: English is the New Code
Leonardo Gonzalez provided a glimpse into his upcoming deep dive on the “Text-to-App” paradigm, inspired by Andrej Karpathy’s famous quote: “English is the hot new programming language”.
He previewed three emerging tools that are shifting development from complex software stacks to simple text files:
Skills.md: A standard where agent capabilities (tools, instructions) are codified in simple Markdown files with YAML front matter, making them versionable and readable.
Ralph Wiggum: A viral “brute force” loop that decomposes projects into tasks and relentlessly iterates prompts until the code works - sometimes running for 8+ hours to solve complex problems.
Open Pros: A prompt-based orchestration layer that replaces complex frameworks (like LangChain) with structured text definitions.
Takeaway: The future of software automation isn’t just about better models; it’s about versionable, audit-able text files that direct those models to perform repeatable work.
Next Steps:
Book your place on 3 Questions w/ AI CoE via the Google Chat Group
Come to the next Office Hours to discuss the use-case we can automate for you




![[News Brief] React Sleepers, OCR Wins, and Braintrust Agents](https://substackcdn.com/image/fetch/$s_!-p3P!,w_140,h_140,c_fill,f_auto,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F47f67c3d-90c5-4786-9f9e-4a8b13fd9fb2_1024x1024.png)
![[How-To] Automate Influence via Google Chat](https://substackcdn.com/image/fetch/$s_!M0Av!,w_140,h_140,c_fill,f_auto,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F4aecf734-d714-4c96-9dd4-e8a8f53d48bb_2876x1438.png)