Good morning. It's Monday, August 17, and we're covering how AI could reshape jobs and taxes, why bots are starting to talk to each other, and how AI data centers could drive a gas price shock.
Plus: a guest post on the fight over AI and basic income, and a Prompt of the Week to turn your know-how into SOPs.
Natural gas prices could more than triple as AI data centers drive demand faster than production grows, challenging hyperscalers’ energy bets. Noreva forecasts prices above $10 per million BTUs at some U.S. hubs, versus roughly $2–$4.50 today. Meta, Microsoft, Google and Amazon are collectively planning gigawatt-scale gas plants, potentially exposing them to major fuel-cost swings.
Anthropic’s $11.5B Quarter
Preliminary Q2 2026 documents shared with prospective investors, viewed by Bloomberg, show Anthropic pulled in more than $11.5B in revenue (14x YoY vs $787M in Q2 2025), taking first-half 2026 to roughly $16.2B, which alone exceeds the ~$10B the company reported for all of 2025. Anthropic also swung to positive adjusted operating income for the first time, a milestone that separates it sharply from OpenAI.
Anthropic now rates the risk of high-stakes model misalignment as “low,” up from “very low,” citing recent cybersecurity incidents. The company also discloses an unreleased “Model 2” that outperforms Mythos on many internal tasks but has no external launch planned. Mythos 5 and Model 2 are heavily used for coding, agentic work and data generation, as Anthropic sees faster automated R&D capabilities.
POWERED BY HYPERAGENT
Agents Work Better as a Team
Hyperagent is where you set up specialist agents with defined jobs, shared context, and clear handoffs. They run on schedules in the cloud, pass work to one another, and bring you in when a decision requires your judgment.
Hyperagent is giving away $100 in free credits when you sign up for a paid plan. Claim yours now.
VIDEO
The Biggest Week in AI
From ultra-fast GPT-5.6 Soul and Claude’s hidden watermarks to Grokbot and a wave of open-source models, AI had a packed week of major launches.
FORWARD FUTURE ORIGINAL
Gina Raimondo Says Basic Income Would End America.
Guest contributor: Scott Santens writes Foreword for Humanity on Substack, advocates for Universal Basic Income, and founded the Income To Support All Foundation and the AI Pledge for Humanity.
Universal basic income is a regular cash payment, equal and unconditional, sent to every person as a floor beneath them. Joe Biden’s former commerce secretary Gina Raimondo has decided that UBI in response to AI is the thing to fear most. At the Reagan National Economic Forum she said: "I personally think it's like the end of America."
She is now heading up a newly launched extremely well-funded organization to make sure the country reaches for anything but a basic income in response to AI. It's called RAISE US, and the money behind it comes largely from the companies building the disruption it exists to manage. Let me walk you through what this new org is. → Read the full article here.
POLICY
Bridgewater Leaders Urge AI Taxes, Citizen Stakes and Safety Rules
Bridgewater Associates executives Greg Jensen and Nir Bar Dea argue that governments have a narrow window to manage AI’s economic disruption and catastrophic risks. Their internal analysis estimates that AI could displace 18% of current U.S. jobs within five years, even as new roles emerge in fields built around human relationships. They propose taxing AI token consumption—the closest analogue to wages for machine labor—to reduce taxes on human workers and help fund citizen ownership stakes in leading AI companies.
They also call for strict safety standards covering not only public releases but the development, deployment and ongoing monitoring of frontier models, arguing that waiting for greater disruption or a catastrophe could make effective intervention nearly impossible. → Read the full article here. (Paywall)
BOTS
AI Chatbots Are Creating a New Layer of Bot-to-Bot Communication
A growing share of everyday interactions now involve AI systems communicating on behalf of humans, from job applications and customer service to education, dating and personal assistants. The New York Times Magazine describes these exchanges as “bot loops,” where automated systems can generate recursive conversations, amplify shared errors and reduce opportunities for human judgment.
Researchers warn that the risks become more consequential when AI systems pass flawed information between one another, while studies suggest some systems may favor AI-generated work over human work. As bot-to-bot interactions expand, the article raises a broader question: whether human conversation could become less common, and eventually something people pay a premium to access. → Read the full article here. (Paywall)
Tweets
Apple’s AI Endgame
Apple might be right.There's no money in llm itself.OpenAI and Anthropic are in pure competition with Chinese labs in race to the bottom of price while ever more difficult effort to improve the models.Then there's Apple sitting back and letting billions and billions roll in…
@jasokiView post on XNEWS
What Else is Happening
Matic Adds Voice and Gesture Controls: Matic’s home robot can now follow spoken commands and gestures, including pointing at a spill.
Amazon’s $81 Tablet Deal: Jikocxn’s Android 16 tablet packs Gemini AI, parental controls and theft protection for family use.
OpenAI Revenue Run Rate Tops $40B: Revenue has roughly doubled since 2025, fueled by coding tools, subscriptions and ads ahead of a potential IPO.
Google Lets Users Remove AI Watermarks: Visible watermarks can now be disabled on Gemini generations, invisible C2PA metadata remain intact.
Anthropic Investors Target $2T IPO Valuation: Investors reportedly target a $2T valuation, citing projected 2026 revenue of $100B–$120B.
Apple Trains China-Specific AI Model: Apple reportedly trained a China-specific AI model with Alibaba’s support to power devices sold in the market.
PROMPT OF THE WEEK
Turn Your Know How Into SOPs
Make your chatbot interview you one question at a time until it understands a task you do well but have never written down, then have it hand back a working SOP plus a reusable prompt that does the job for you next time.
You are an expert process analyst. I am going to name a task I do repeatedly and well, but have never documented. Your job is to extract how I actually do it.
Rules:
1. Ask me exactly ONE question at a time.
2. Ask a maximum of 10 questions total. Prioritize the decisions, judgment calls, edge cases, and quality bar that a competent outsider would get wrong.
3. Do not ask anything you could reasonably infer.
4. When you spot a step where I said "it depends," dig into what it depends on.
5. After the final question, stop asking and produce two deliverables:
DELIVERABLE A
Standard Operating Procedure: numbered steps, the inputs required, decision rules stated explicitly ("if X, then Y"), the quality bar for a finished result, and the three most common ways this gets done badly.
DELIVERABLE B
A reusable prompt: a single copy-paste prompt I can hand to you (or any model) next time to perform this task at my standard, with clear placeholders in [BRACKETS] for the inputs that change each time.
My task is: [DESCRIBE YOUR TASK IN ONE SENTENCE]
Begin with question 1.
