This Company Spends $32m Per Year On Claude Code!?

Eric opens with Jacob’s company Spawn, which has only 10 employees but reportedly spends about $32 million per year—roughly 30 times payroll—on AI tokens. Jacob says the company shipped 77,000 substantial changes in 30 days, at about 108 per hour, using permanent AI agents with their own areas, permissions, and memory. Spawn coordinates through its living code and product rather than pull requests, while five humans focus on taste and entropy. Eric says this shows how radically AI-native companies can operate.

Eric says marketers used to be told not to touch code or ship pull requests, but AI agents now let them request work directly: 20 iterations, five landing-page tests, 20,000 ads, or 10 trial reels at once. He recommends asking every team member, and even the agent itself, what unreasonable task could be attempted next. The demand for AI work is effectively limitless, but speed should not mean making dozens of changes simultaneously without knowing which change improved or damaged conversions.

Eric describes Money Prompt, a tool that analyzes code and the entire funnel to identify drop-offs and suggest fixes. It found missing activation emails, absent reminders and incentives for users running low on credits, and a page where 20% of users abandoned. Engineers told Eric they could not have inferred those business patterns from the code alone. He argues that experience still matters: marketers may understand revenue behavior, while engineers understand robust implementation, and neither should assume AI eliminates the other’s judgment.

Eric and Neil review Jen Abel’s startup-sales list. Abel says outbound sales often produces the highest average contract value, particularly for enterprise SaaS and RFP-driven deals. She also argues that demos should focus less on product features and more on increasing the buyer’s influence, while Neil says his strongest-performing demos show how the product solves problems and ties them to money made or, secondarily, money saved. They also agree that founders should handle most early sales because they can answer objections and change the product quickly.

Neil says founder involvement improves sales less through superior pitching than through discovering product objections and fixing them. Eric recalls Larry Page returning to Google’s AI efforts, cutting through internal resistance, and helping the company start shipping competitive AI products. He also cites Jason Cohen of WP Engine describing a founder’s “invisible hammer”: unlike a hired CEO, the founder can rapidly unblock decisions. Both hosts say founders should eventually hand off sales, but remain deeply involved while sales is the startup’s main bottleneck.

Neil agrees with Abel that a win rate above 50% is often a warning sign. It may indicate deceptive or unclear expectations, underpricing, or a product that is too easy to sell because buyers do not understand what they are receiving. He recalls NP Digital selling SEO services for $250 per month to a sombrero business; the price produced an extremely high close rate but could not support a profitable result. Neil refunded the money, helped the owner on several calls, and improved the business before declining to take her on as a client.

Eric discusses an X audio session claiming that “GPT-6 Astra” had been distilled, meaning its model was copied, and notes that Anthropic and OpenAI have themselves distilled information from the internet and search. He connects this to AI-created drugs reportedly entering phase-three trials and argues that software parity is arriving quickly. Since intellectual property is becoming easier to reproduce, he recommends giving more away, delivering customer value, and shipping faster. For sales, the hosts add that a no-brainer offer must combine clear value, appropriate pricing, relevant experience, and a solution to the buyer’s real problem; tools such as Gong can reveal when salespeople talk too much instead of listening.