Finally finished this article. As a data practitioner, I couldn’t agree more on the system of records concept. Internal organization transaction costs depend on information flow speed, which drives decision velocity. AI coding enables integrating info from various systems and sources much faster than ever. More ambitiously, one can build a new system or application holding all systems of records, greatly reducing firm communication costs. That cost reduction and value creation is massive with a lesser magnitude of people if the leader does it methodically.
That is what I’m doing in my team: building a new data Infra and application to capture all data points and information, including past human decisions for operational management, and provide insights and recommendations for decision makers.
Key notes to ruminate:
• A shockingly large component of “what is someone’s day-to-day job” is constrained by the employer’s system of record and decision-making, itself constrained by the database technology used and what it makes possible.
• Everyone is a manager.
• Today, thanks to new technology, front-line craft practitioners are regaining responsibility—and we’re all better for it.
• For agents to act autonomously, much of the ontology, authority, and evaluation currently in people’s heads must move into the firm’s system of record.
• His answer came down to transaction costs. Finding the right person, negotiating tasks and prices, and monitoring all take time and money.
• A continuously updated representation of the organization’s operations, decisions, capabilities, and priorities.
• Civilization advances by extending the number of important operations we can perform without thinking of them.
The article is gold. my takeaways:
As a solo developer:
1. Before opening a PR, run a handful of internal skills and verification steps (tests, builds, diff reduction, end-to-end checks, UI vs design guidelines, etc.).
2. Run an AI review on every pull request using a new agent with fresh context focused on the diff.
As a team:
3. Define risk tiers and classifiers based on the change, not the author. One classifier sets the risk level; another checks whether extra approval is still required even if the tier doesn’t call for it. Some paths always need another approval (sensitive code flagged by owners, CI workflow files, etc.).
4. Ask Claude to generate a mini explainer of the change. It’s faster and more useful for understanding context than reading every line of a trivial diff.
5. With agents it’s cheap to start new work — but sometimes you shouldn’t. That’s another reason for keeping humans in the loop beyond just security and ownership.
Sounds like Zuck got the cheat book from @elonmusk to succeed: make a compelling case, attract the best talent, and secure large—even humongous—investments.
Cheeky Pint is back! Mark Zuckerberg @finkd and Alexandr Wang @alexandr_wang came to the pub to discuss their new frontier model, Muse Wonder.
00:00:24 Muse’s breakout use cases
00:07:28 The model is the product
00:14:35 Muse Wonder
00:24:56 How frontier models are built
00:48:26 Betting $100B
01:07:21 Agentic commerce
01:19:34 OpenRouter
01:20:06 Beyond the smartphone
01:27:00 Funding
01:34:06 Why AI won’t be winner-take-all
01:40:46 Meta Superintelligence Labs (MSL)
02:08:11 Working for Mark Zuckerberg
