Forwarded this email? Subscribe here for more.
This week on Signal & Noise, we published two pieces on a question every company now faces: as AI takes on more of the work, who owns the advantage, and who makes the decisions behind it?
First, the new episode with Spencer Gusberg, “Own Your AI Advantage: Open-Source Models, Data Sovereignty, and When to Rent vs Buy,” is here. Spencer is Founder and CEO of e52, an independent AI engineering firm, co-founder of P2 Analytics (acquired), and an MIT MBA with 15 years of experience building data tools for startups through the Fortune 500. Watch or listen on your platform of choice!
Then, a new article from Quinn Harrington, Founder and Chief Brand Concierge at Harrington Design Company: “I’m a Supercomputer, Not a Magician.” It’s a satire, and a response to Rio Longacre’s “Is AI Alive?” Quinn goes undercover with five chatbots and the worst creative brief she could write, to find out whether the tool or the brief is the problem.
JUST RELEASED PODCAST!
Should you rent your AI or own it? Spencer Gusberg, Brett House, and Rio Longacre talk about when to rent intelligence from the frontier labs, when to build and own it, and what happens to the data and know-how you hand over along the way.
Spencer’s just published research, The Smartest Companies Are Building Their Own AI looked at what 30 top tech companies disclose about the AI they actually run, and found that many, including LinkedIn and Atlassian, build their own models for high-volume jobs and rent from the frontier labs only where they need to.
It’s candid and wide-ranging conversation, tackling topics from New Haven pizza superiority to whether AI is repeating the Solow productivity paradox (e.g., massive investments in information technology do not immediately show up in macroeconomic productivity statistics).
Key takeaways:
Split your AI spend by job: rent frontier models for broad, judgment-heavy work, and own models for high-volume, repeatable work.
A trained open weight model can beat the frontier model you rent. LinkedIn reported roughly 75x lower serving cost with an open weight model it hosts itself.
The moat has moved. Models are a commodity, and the advantage is your harness, context, and proprietary data. “Context is king.”
“No training” isn’t the whole story. Policies change, and most of the risk lives in the copies made along the way.
Spend charts miss the shift: open models are about a third of OpenRouter tokens, while the big labs collect 88 percent of API dollars.
The hard part is people. Spencer has yet to see a project fail for lack of technical capability.
Watch (or listen): Own Your AI Advantage
JUST PUBLISHED ARTICLE!
By Quinn Harrington, Founder and Chief Brand Concierge, Harrington Design Company
When AI design comes out bad, is the tool the problem, or the brief? Quinn Harrington went undercover as “Quinnie,” a junior marketing coordinator, and handed ChatGPT, Gemini, Perplexity, Copilot, and Grok the worst brief she could write: a Times Square takeover with ten taglines, Papyrus headlines, and a veto for the CMO’s book club.
In this satire, a response to Rio Longacre’s “Is AI Alive?”, the bots push back, count the mandatories, and threaten to quit before the boardroom reveal: “You don’t have an AI problem. You have a leadership problem.” Her fix: pick one goal, pick one decision-maker, and let the experts work.
Read the article: I’m a Supercomputer, Not a Magician.
Want to see more great no-BS content?
Visit https://www.signalandnoise.ai




