Infinite markets: We don't optimize for COGS | Christopher O’Donnell (Day.ai)

"The universe is like, you need to build this AI native CRM that's all automatic and I'm like, that sounds great"

Aug 15, 2025

Christopher O'Donnell just built what might be the most technically brutal CRM ever attempted. After two years and two million lines of code connecting Gmail APIs, Google Calendar, and meeting transcripts into a preprocessing engine that actually works with Day.ai, he's got some spicy takes on AI economics, the coding tool wars, and why the universe might be conspiring in your favor.

The "Total Piece of Cake" Preprocessing Hell

Christopher's favorite response when competitors think they can copy Day.ai's approach:

"Good luck. Have fun. I'm here for a hug when you need it”.

The technical reality of building an AI-native CRM is absolutely savage. While everyone else builds thin ChatGPT wrappers, Day.ai processes a "fire hose of incoming data" to create actual intelligence.

Instead of feeding raw database tables to LLMs, they preprocess everything into beautifully written natural language narratives. Think less "John Smith, VP Sales" and more "John is the real decision maker who's interested but you need to sell him on the core value prop, while Sarah is the champion with zero internal sway and Mike the lawyer actively doesn't want this deal."

The result? When you ask about your pipeline, you get actual strategic intelligence instead of regurgitated contact records.

Don't Optimize for COGS

While every AI startup sweats inference costs and switches to cheaper models, Christopher drops this bomb: "We should not optimize for COGS".

His framework flips conventional wisdom: When the gap between frontier models like Claude Opus and budget alternatives is massive, you optimize for customer value, not unit economics. Use the expensive model that delivers 10x better results and charge accordingly. The costs will smooth out as open source catches up and capacity increases.

"At a time when the difference between the state of the art frontier model that's going to be expensive and the best open source alternative is as big as it is today, it's not the time to optimize for that."

Ergonomic Pricing: The SKU Revolution

Christopher introduces "ergonomic pricing" - letting users create multiple instances of a paid SKU, each with their own permissions, model instructions, and even DISC personalities. His AI assistant has a whole backstory as "the child of a Swiss watchmaker" who's "super analytical."

You can literally tell your AI assistant: "Your name is Klaus, be really direct and German with me always." And it updates itself accordingly. This isn't just personalization - it's treating AI agents like actual team members with distinct roles and personalities.

Claude Code vs Cursor: The IDE Wars Get Spicy

The conversation takes a sharp turn into the current coding tool bloodbath. Christopher sees Claude Code pulling ahead not just on model quality, but on fundamental architecture. While Cursor builds a better IDE, Claude Code is building the foundation for autonomous development workflows.

"If they're going to give you the harness to do that, but you can't really, right? Because you're not really spinning up like a whole container. You don't really have MCP working in that environment."

His prediction? Claude Code extends its lead, but there's a massive "overhang" between what these tools can theoretically do and what anyone actually uses them for. We're all underutilizing the current generation while waiting for the next one.

The Transparency Trust Crisis (Claude Code Gets Real)

"The things that really damage trust are the unexpected charges. And the things that really build trust are like, hey, we're going to tell you exactly what this is going to cost you before you do it."

While Cursor hides behind vague usage limits, Claude Code is building transparent pricing that shows you exactly what each operation costs before you run it. No surprise bills, no mysterious usage spikes, no "contact sales" for overages.

"I think Anthropic and the Claude Code team specifically are like really thinking hard about this stuff in a way that I think is important for customer trust."

The Async Intelligence Breakthrough

Christopher's AI assistant Chloe watches meeting recordings overnight and emails him the top five priorities, blocking bugs, and product positioning lines that actually landed with prospects. This isn't a demo - it's working today.

"I woke up this morning with an email from my assistant Chloe and she had watched all of the meeting recordings from the previous day and came in and said, here are the issues, here are the top five things you got to worry about."

This represents the shift from reactive chat interfaces to proactive intelligence that works while you sleep.

The CRM Apocalypse: Three Scenarios

Christopher maps out potential futures with brutal honesty.

Scenario one: status quo holds, everyone builds better products.
Scenario two: a few AI-native winners emerge alongside improved incumbents.
Scenario three: one player takes everything and "Salesforce will not be a public company" by 2028.

He puts the apocalypse scenario at 20-30% probability. "Maybe somebody starts a company in 90 days with some new frontier model, with some new mindset, and they just stick the landing and get it all perfect."

The Universe Conspiracy Theory

Christopher's philosophical framework sounds like Silicon Valley mysticism until you hear the application: "The universe is conspiring in your favor if you're paying attention." When technical complexity feels impossible, when competitors emerge, when pricing gets weird - lean into the flow instead of fighting upstream.

"We do these things not because they're easy, but because we thought they would be easy. The universe is like, you need to build this AI native CRM that's all automatic and I'm like, that sounds great. And then two years in, two million lines of code, and I'm just like, okay, I'm tired."

The Infinite Market Reality

"This is an infinite market". The CRM space is so massive that multiple AI-native players can build billion-dollar companies without directly competing.

His advice for the chaos ahead?