Superstar researchers at frontier labs can earn over a hundred times more than the average AI postdoc, and the gap is only widening. OpenAI and Anthropic now cluster above $300K in base salary alone, before equity that can double or triple total packages. Researcher quantity doesn't easily make up for quality in AI. Frontier labs are compute-constrained, they can only run so many experiments, so two "merely very good" researchers can't replicate one superstar with deep intuition about which experiments are worth running.

Even a 2x researcher can earn far more than the median because their contributions scale to billions of users. At the very top, individual deals have reportedly reached $1.5 billion over six years, a figure that makes NBA max contracts look modest. When the prize is automating trillions of dollars of human labor, paying $50M for the right mind isn't irrational, it's basic math.

Google I/O 2026 kicks off May 19, and the headline act is almost certainly Gemini 4, Google's answer to GPT-4.5. Expect a massive context window, native multimodal processing, and a heavy focus on agentic coding — a direct shot at Claude Code and OpenAI Codex. Google is also pushing Gemini beyond the chatbox: an internal agent codenamed "Remy" is already being tested as a "24/7 personal agent" that can take actions on your behalf and handle complex tasks proactively. This is Google's biggest AI moment in years — and a clear signal they're done playing catch-up.

Higgsfield just turned "describe what you want" into a finished ad. Supercomputer is a chat-based agent that takes a plain-language brief, plans the job, picks the right model, and delivers the asset, no prompts, no presets, no tool-hopping. The proof is already in the wild: using the same underlying stack, a small creative team produced a 23-minute sci-fi pilot in just 96 hours, work that would traditionally take a team of 50 people six months. The agent connects to 30+ tools including Slack, Notion, and Figma, stores your brand guidelines in long-term memory, and gets smarter with every task it completes. For marketers drowning in creative production, this is less a tool upgrade and more a full role replacement.

Ineffable Intelligence, the London startup founded by AlphaGo architect David Silver, just landed an engineering-level partnership with Nvidia, weeks after closing a $1.1B seed round at a $5.1B valuation. The core bet: AI that learns from experience, not human data. Engineers from both companies will co-build a large-scale reinforcement learning pipeline running on Nvidia's Grace Blackwell chips and the upcoming Vera Rubin platform.

Silver's argument is that LLMs can remix human knowledge but can never discover something genuinely new, AlphaGo's Move 37 against Lee Sedol was never in any human game record; it was found by a machine reasoning beyond human intuition. If he's right, this is a very different kind of AI race.

Anthropic just made its biggest push into real-world work. Claude for Small Business plugs Cowork into QuickBooks, PayPal, HubSpot, and DocuSign, with 15 ready-to-run workflows covering payroll, invoicing, and month-end close, at no extra charge beyond your existing subscriptions. Small businesses account for 44% of U.S. GDP but have consistently lagged larger enterprises in AI adoption.

On the legal side, Anthropic dropped 12 practice-area plugins, covering M&A, employment, IP, and litigation, plus 20+ connectors to platforms like Westlaw, iManage, and Relativity. One clear signal: Claude is coming for your entire workflow, not just your chat window.

Lovable just solved a quiet problem every vibe-coder has: you build an app, but nobody finds it. New apps now ship with full server-side rendering, real HTML that AI crawlers can read immediately, with pre-rendered snapshots auto-generated for older apps too, free on all plans. Your app now surfaces across Google, ChatGPT, Claude, and Perplexity without touching an SEO setting. Building was never the hard part. Getting found is.

Building a physical product used to mean weeks of back-and-forth with designers before you could even talk to a factory. Genpire skips that entirely — describe your product or upload a sketch, and it generates a factory-ready tech pack in minutes, complete with measurements, materials, construction details, and technical drawings. No design degree required.

Traditional CRMs are glorified spreadsheets, you do all the work, they just store it. Lightfield flips that: every email, call, and meeting gets captured automatically, and you build agents with plain-language prompts to handle prospecting, follow-ups, and deal coaching. You sell, the CRM does the rest.

That's a wrap for this week.
Thanks for reading, we appreciate you being here.

See you next week.

P.S. Hooked is brought to you by Reloop, the AI marketing creative platform we're building on the side.