AI Summary
5 min readWhy Your Company Should Own Its AI Model
Josh Sarrotti, CEO of Aragon, argues that companies are wasting money and losing control by relying on frontier models for every internal task. His company builds custom small language models trained on a company's proprietary data, then gives the company the actual weights. The pitch: instead of paying for expensive context windows and tool calls every time an employee asks a question about internal data, train a model on that data once and let it live in the model's weights.
The problem with frontier models for internal workflows
The current approach to enterprise AI is inefficient. Companies feed their internal documents, Slack history, and email into a frontier model's context window every time they need an answer. Sarrotti points out that input tokens already outnumber output tokens by roughly 1,000 to 1, and the costs add up fast. He cites Uber's CTO, who said the company blew through its entire AI budget by April. "It turns out that even with all of these innovations in harnesses, it's actually quite expensive, depending on the workflows that you want to do, to try to pull into context everything that you need into the model."
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of This Week in Startups
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 (00:00) **Why Your Company Should Own Its AI Model** - Josh Sarrotti, CEO of Aragon, explains the thesis behind building custom, company-owned models instead of relying solely on frontier APIs.
- 2 (08:38) **The "Company World Model" and Continuous Learning** - Josh details the technical architecture, including the use of RL for continuous learning and the foundational model choice.
- 3 (11:23) **Demo: Agentic Workflow and Data Integration** - A live demo shows Aragon handling email triage and Slack data extraction, highlighting its agentic capabilities and data architecture.
- 4 (15:09) **The "Massive Lift" of In-Weight Knowledge** - The conversation pivots to the core value of having company data in the model's weights, focusing on cost and speed.
- 5 (25:33) **Pricing Model and Go-to-Market Strategy** - Josh explains the token-based pricing and the "forward deployed" motion required to onboard complex enterprise clients.
- 6 (30:13) **Founder Journey and YC / A-Cap Relationship** - Jason and Josh discuss their history, from YC rejection to A-Cap's investment, highlighting the importance of betting on the founder.
- 7 (32:27) **Future Vision: Running SLMs on Local Hardware** - Jason and Josh discuss the next frontier: running these custom models on powerful local workstations for recursive, private improvement.
+ Full timestamped outline available in the app
Guests on this episode
Show Notes
This Week In Startups is made possible by:
Quo - https://quo.com/TWiST Vanta -
https://vanta.com/twist
Pilot - https://pilot.com/twist Plaud -
https://Plaud.ai/twist
Today’s show:
*Eragon CEO Josh Sirota believes that you should “own your intelligence.” He demos a live enterprise AI operating system connecting email, Slack, calendar, and financial data into a single agentic layer that’s pre-trained to understand any business inside and out.
PLUS we’ve got IONA Drones founder Etienne Louvet, a first look at Moonshot’s Kimi K2.6, and a brand new Off Duty with JCal and Lon.
FULL SHOW NOTES ON NOTION: https://thisweekinstartups-docket.notion.site/TWiST-2278-SHOW-NOTES-34872ca918e380bd9c71e684ce2db7c9?source=copy_link
Guests :
Eragon: https://eragon.ai
Josh Sirota on X: https://x.com/joshua_sirota
IONA: https://ionadrones.com/
Etienne Louvet on X: https://x.com/etienneldva
Relevant Links
Eragon Seed Round coverage: https://techcrunch.com/2026/03/18/this-startup-wants-to-make-enterprise-software-look-more-like-a-prompt/
Neurometric: https://www.neurometric.ai/
Ollama model: https://ollama.com/
Orange Slice: https://www.orangeslice.ai/
Corgi Insurance: https://www.corgi.insure/
Slash Financial: https://www.slash.com/
The Syndicate: https://thesyndicate.com/
FAA Part 108 Explained: https://pilotinstitute.com/part-108-explained/
Zipline: https://www.zipline.com/