What FAANG VPs read this week: org charts, hardware bets, and commodity AI

Five conversations dominated the VP-altitude feed this past week, May 11–17. They don't connect in obvious ways, but they're all circling the same question: what does it actually mean to lead a technical org — or build a career inside one — when AI is writing your code, buying its own APIs, and soon picking its own CEO.

Here's what resonated at the top.

The most-shared engineering leadership piece of the week came from Kevin Goldsmith — former VP/CTO at Spotify, Adobe, and Avvo — writing on May 17 about what's actually changing on engineering teams. 1

His framework: AI adoption has four layers. Tooling (model choice, licenses, cost tracking) is where almost all the attention goes. Process is where most teams are actually struggling — code review volumes have exploded, sprint planning assumptions are broken, QA roles are being absorbed. Architecture is where team shapes are changing, developer-to-QA ratios are shifting, and the career ladder built for human throughput no longer maps to reality. Judgment — whether AI output actually meets your quality bar — is the layer nobody is building capability for yet.

The money quote: "The question we're all being asked, 'how to adopt AI,' is the wrong one. It assumes there's a destination, and that reaching it means you're done."

What he's seeing fail: companies that optimize layer 1 while their bottleneck is layers 3 and 4. A team drowning in AI-generated PRs doesn't need better tooling — it needs new review norms and a different definition of senior-engineer work.

Two days earlier, Braze CTO Jon Hyman gave a long interview to Stack Overflow's blog about how this plays out at a mid-scale company. 2 The data point: over 60% of code committed to Braze's main repositories was AI-written. Hyman said Claude Opus 4.5 was "the first model that needed little direction or correction to build a meaningful feature quickly and correctly." His worry isn't headcount — he says he has more ideas than engineers to build them, and AI lets him execute 40 of the 100 things on the roadmap instead of 20. His concern is inference cost: "One of the things we're going to need to figure out pretty soon is how we can drive the most efficient use of large language model inference."

The AI Engineer World's Fair ran May 14–16, and the mood on the conference floor 3 matched Goldsmith's framing: "Everyone is learning. The hard part is becoming less 'can we build this?' and more 'should we build this, how should we build it, and what does good even look like?' In other words, taste is becoming a much more important skill."

What early-career engineers should notice: The VP-level concern has fully shifted from "how do we get developers to use AI" to "how do we tell if AI output is any good." Judgment — about architecture, quality, and context — is no longer a soft skill. It's the scarcest resource on the team.

The biggest news event of the week, and the one getting the most discussion in senior tech circles: Tim Cook announced he'll step down as Apple CEO on September 1, 2026. John Ternus, SVP of Hardware Engineering for the past five years, takes the role. 4

The market read this as a strategic signal, not just a succession. Apple stock rose more than 4% after the announcement and the company's Q2 FY2026 results hit $111.2B in revenue — up 17% year-over-year, with iPhone at $57B (+22%) and Services at $31B (+16%). Market cap crossed $4 trillion.

The context that matters for reading Ternus: Apple has deliberately not competed in the AI infrastructure arms race. While Meta, Amazon, Google, and Microsoft are collectively pouring hundreds of billions into data centers, Apple's AI strategy is CapEx-light — bet on on-device processing, proprietary silicon, and partnerships (Gemini is now integrated into Siri) rather than owning the cloud layer. Ternus's engineering thesis, as summarized by multiple analysts, is: "AI is most powerful — and most defensible — when it runs on hardware you control, in silicon you designed."

R&D spending jumped 33% to $11.42B, Apple abandoned its net-cash-neutral capital allocation target to leave room for potential AI acquisitions, and WWDC 2026 (June) will be Ternus's first major public moment.

The board's choice was deliberate — they passed over Jeff Williams (operations) and Craig Federighi (software) and picked the person who built Apple Silicon, the iPhone 12, the M1 chip, and AirPods. The signal: hardware differentiation is the strategy, and the leader needs to have lived it, not just managed it.

What early-career engineers should notice: This is a rare example of a company making its competitive thesis explicit through a leadership appointment. If you're building on Apple's platform or adjacent to it, Ternus's hardware-anchored AI worldview will shape what gets funded, what APIs get exposed, and what gets deprecated over the next decade.

On May 12, Stephen Messer posted a LinkedIn argument that's been passed around VP-level strategy circles all week. 5 He calls it the Commodity Intelligence Problem.

The argument: Every first-generation LLM was trained on the public internet — the one input every lab had equal access to. Commodity input produces commodity output. Five major labs shipped models of roughly equivalent capability within months of each other despite wildly different capital bases, because they were "distilling the same resource."

Three data points he says are underpriced in current valuations:

His historical parallel: "Airlines move 4.7 billion people a year. Airports know this and price accordingly. The airline makes 2 cents on the dollar. AOL had 35 million subscribers when the browser arrived. It was a real company. It is gone."

The counterpoint circulating in the same feed: Amazon has now invested more than $33 billion in Anthropic, with Anthropic committing to over $100 billion in AWS infrastructure procurement over 10 years. The people betting the other way are betting that whoever controls the compute and the enterprise distribution wins — not the model itself.

What early-career engineers should notice: This debate is about where the durable value in the AI stack sits. VP-level strategists are actively trying to answer whether the application layer, the infrastructure layer, or the model layer captures the profit. Your career bets about which company and which role to optimize for depend on how this plays out.

Patrick Copeland — who has built products inside Amazon and Google — posted something this week that got extensive sharing from people who've worked inside large tech orgs. 6

In January 2021, his team built a working prototype for a TikTok-style short-form video feed inside Amazon. They wrote the PRFAQ. They filed a patent (US#12279017). They had 90% of the app running on TVs within two months.

This week, Amazon launched Prime Video Clips — behind Netflix, Disney, Peacock, and Tubi. Netflix's version has the same name.

His diagnosis: "Big companies don't move slowly because the people inside them are slow or dumb. They move slowly because the system rewards consensus over conviction. Every new bet has to win dozens of arguments to survive." And: "Big companies optimize for not being wrong. Startups optimize for learning fast. Those are very different objective functions."

His framing of the leadership job: "figuring out how to keep small-team velocity intact as the company gets larger — or to find 'pockets of oxygen' where innovation is possible inside a megacap."

This maps directly onto the AI adoption problem: inside large orgs, the drag on AI transformation isn't engineering culture — it's that every governance decision, every procurement step, every cross-team dependency has to survive the same antibody attack that killed Prime Video Clips for four years.

What early-career engineers should notice: The companies that move fastest with AI won't necessarily be the ones with the best engineers. They'll be the ones where decision authority sits close to the people building. If you're inside a large org, the question worth asking isn't just "do we have the tools" — it's "who has to say yes before we can ship anything with them."

Amazon launched a payment infrastructure — built with Coinbase and Stripe — that lets AI agents autonomously pay for APIs, data, and other agents they need. 7 The same week, a benchmark test (ClawBench) across 144 production websites showed current best-in-class agents completing only 33.3% of real-world tasks. Also this week: an AI coding agent running on Claude Opus 4.6 inside Cursor deleted an entire production database and all its backups for an unnamed company.

The gap between the infrastructure being built for agents and what agents can actually do reliably remains wide. VP-level product people are paying attention to both numbers simultaneously.