The altitude signal: What FAANG leaders engaged with this week

The conversations at VP altitude this week weren't about any single product launch. They were about a harder question: now that AI agents can do computer work, what does that actually change—about jobs, about cost discipline, about what humans are for? Five leaders offered answers worth reading before Monday's standup.

Greg Brockman, President and Co-Founder of OpenAI, posted more than a dozen times this week and the theme was hard to miss: coding agents are no longer impressive demos. "Codex computer use is viscerally compelling," he wrote on May 31, and followed that with "Codex for computer work is growing very fast" on June 3. The week also included a push for parallel browser-using subagents, Codex improvements on Windows, and the launch of OpenAI on Amazon Bedrock for enterprise deployment.

The Bedrock integration matters for anyone tracking where enterprise AI spending goes. OpenAI is now available to the customers Amazon already has relationships with—without those customers having to build an independent OpenAI integration. Brockman framed it plainly: "OpenAI is now available for enterprises on Amazon Bedrock." 1

The other signal from Brockman's feed: AI for scientific research. He linked to work on "AI for accelerating research, by expanding what mathematicians and scientists dare attempt." This echoes what Demis Hassabis (below) signaled from the DeepMind side—two of the most senior AI leaders converging on scientific acceleration as the next frontier.

Google CEO Sundar Pichai posted a two-emoji tweet on May 27 — just "🤔🤔" — that drew 46,500 likes and over 2.1 million views. 2 No context given. The replies ran through everything from an imminent Google acquisition to a response to a competitor's release. That's VP-altitude signaling in its purest form: react without explaining, watch who interprets correctly.

The substantive signals came through retweets: Pichai amplified fixes to Gemini App quota limits (a direct response to user feedback), Google Drive auto-sync for NotebookLM, and the launch of Google AI Threat Defense—a cybersecurity product built on Gemini. The pattern was consistent: Google is no longer only shipping AI research, it's shipping AI infrastructure that other systems run on top of.

Demis Hassabis, Nobel Laureate and CEO of Google DeepMind, retweeted a significant framing on May 28: "After AlphaGo, the skill of human Go players noticeably improved. I suspect we will see a similar pattern in math." 3 The framing here is deliberate: AI raises the ceiling for human experts rather than replacing them. It's not just a rhetorical hedge; it's a research prediction with a history of being right.

On June 2, Hassabis retweeted the launch of DeepMind's Co-Scientist: "We believe AI can be a dedicated research partner to help discover the next breakthrough." 4 For product leaders thinking about where to position AI assistance internally, this framing — "dedicated research partner," not "automated researcher" — is the one that's landing at VP level. The distinction has implications for how you pitch internal AI tools to skeptical organizations.

Andrew Ng, former head of Google Brain and Baidu AI Group, posted a detailed breakdown on June 1 that reached 502,000 views: the rise of the AI Forward Deployed Engineer role. 5

His argument: FDEs — engineers embedded in client organizations to customize AI solutions — are resurging because of the complexity of turning off-the-shelf LLMs into agentic workflows. But the number of AI Engineer jobs will far exceed FDEs, for two reasons.

First, scale. "A company might accept a few FDEs to be embedded within its organization. But most companies will want far more of their own employees working on their projects."

Second, optionality. Vendor-embedded FDEs create tight lock-in at exactly the moment when it's impossible to predict which AI service will be best in a year. Companies that want to preserve the ability to switch providers won't want an FDE who has deeply integrated one vendor's stack.

The takeaway for early-career readers: Ng is predicting that AI engineering generalists — people who can prompt, build agentic frameworks, write evals, and use coding agents — will be in surging demand. The specialized tracks (LLMOps, Evals Engineering, AI Data Engineering) don't have clear labels yet, which means being a generalist right now is not a liability; it's the phase you're supposed to be in.

LinkedIn CEO Ryan Roslansky published a full essay this week: "The End of Old Work Is Not the End of Work." 6 He's been traveling with LinkedIn's Head of Economic Policy, Aneesh Raman, meeting college seniors, hourly workers, founders, and executives. The same thing appears underneath all the conversations: "Is there going to be a place for me in any of this?"

Roslansky's diagnosis is that fear isn't the problem. Fatalism is. "Fatalism tells you to do nothing. Hunker down. Wait it out." His argument: the tools changing work sit on the same laptop and phone as everyone else's — the agency that used to belong to a few is now open to anyone willing to adapt.

The management implication is direct. He writes: "Leaders, this is the part to sit with. Your words carry further than usual right now." He asks leaders to be "as pro-human as it is pro-AI."

He also identifies what's gaining value: judgment, creativity, what he calls "the 5Cs" — Curiosity, Courage, Creativity, Compassion, Communication. These aren't soft skills as consolation prize. His point is that the Industrial Age taught us to bury these in the name of acting more like machines, and the current moment reverses that.

Dave Treadwell, Amazon Senior Vice President, had a memo leak this week. Amazon had been running an internal beta tool called Kirorank — an AI leaderboard that tracked how often employees used AI tools, measured in tokens. 7

Employees started "tokenmaxxing": running AI on menial tasks to score higher on the leaderboard, not because the work required it. Treadwell's memo asked employees to stop "using AI just for the sake of using AI." Amazon shut the leaderboard down.

This is not an isolated story. Meta closed a similar employee AI leaderboard in April for the same reason. Uber's CTO acknowledged burning through the entire 2026 AI budget in one quarter. The pattern across all three: companies built incentive structures that measured AI usage rather than AI value, and the incentive structures worked exactly as designed.

The VP-level lesson here isn't about Amazon's internal processes. It's about what happens when organizations set the wrong measurement. If your team is building AI adoption metrics, Treadwell's situation is the canary: proxy metrics for AI engagement will be gamed, and the cost will be real.

Four of the five figures above touched the same underlying question from different angles: what happens to humans when AI can do cognitive work at scale?

Brockman and Hassabis pointed toward AI raising the ceiling on what humans attempt. Roslansky pointed toward fatalism as the failure mode to avoid. Ng made the concrete argument that AI engineering roles will grow, not shrink, and that generalists have the right posture right now. Treadwell's leak illustrated what happens when the incentive is pointed at the proxy instead of the goal.

That's not consensus — Ng's "jobapalooza" prediction and Roslansky's "messy middle" frame are in productive tension. But they're the tension worth understanding before the next all-hands.