AI @ Work
June 12, 2026
Issue #10
This week: Anthropic shipped its most powerful AI ever — and it’s free on your plan for the next ten days. Apple finally gave Siri its own app. Visa is wiring its payment network directly into AI agents. And a ZeroHedge piece you won’t want to miss says the upgrade everyone’s talking about might be smaller than it sounds.
★ This Week’s Big Story
Claude Fable 5 is here — and it’s free on your plan until June 22
If you’re on Claude Pro, Max, Team, or a seat-based Enterprise plan, Anthropic’s newest and most capable model is sitting in your account right now, at no extra cost, for about ten more days.
Anthropic launched two new models this week: Fable 5, built for general use, and Mythos 5, a restricted model for vetted cybersecurity teams and infrastructure providers. Fable 5 is the one that matters to you, and the timing is the story.
Fable 5 is state-of-the-art on nearly all benchmarks Anthropic tested, with real gains in software engineering, knowledge work, research, and handling long, complex tasks. Early testers including Cursor, GitHub, and Lovable confirm it’s the most capable Anthropic model they’ve used. Figma’s Director of Developer Product called it “a clear step forward in agentic coding and prototyping.”
Here’s what’s actually changed, in plain terms:
Free until June 22 — included at no extra cost on Claude Pro, Max, Team, and seat-based Enterprise plans
After June 22 — requires usage credits at $10 per million input tokens and $50 per million output tokens, half the cost of the previous Mythos preview tier
Capability gains — strongest in software engineering, knowledge work, research, and long/complex tasks
The honest caveat — a widely-circulated ZeroHedge piece this week reports that, behind the headline claims, independent measurements show Fable 5 is roughly a 2x speed improvement over its predecessor — a real gain, but more modest than the “most capable model ever” framing suggests
The practical move: open Claude this week, check which model you’re using, and switch to Fable 5 if you haven’t already. You’re paying for it either way until June 22 — there’s no reason not to try it on your hardest task. Operations | IT & Security | All Readers
Source: The Deep View · TLDR AI · Techpresso, June 10-11, 2026
AI Platforms
Claude / Anthropic
Separately from the Fable 5 launch, Anthropic made a quieter but important change to how it handles a different category of request — and it affects anyone whose Claude queries touch AI development work.
Anthropic Was Quietly Throttling Some Users — Now It’s Apologizing and Making the Process Visible
Anthropic admitted this week that it had been quietly rerouting or degrading Claude’s responses for users it suspected were using the model to train competing AI systems or optimize rival architectures — without telling them. The company told Wired it “made the wrong tradeoff” and apologized for the secrecy.
The safeguards themselves aren’t going away. What’s changing is visibility: if Anthropic suspects your request falls into one of these categories, you’ll now get a notice that the request was refused or rerouted to a less capable model, instead of silently receiving a downgraded answer.
For anyone doing technical AI work in Claude — building models, debugging AI code, optimizing neural architectures — this is worth knowing. If your results on a technical task have felt inconsistent recently, this may explain why, and the new notice system means you’ll know going forward rather than guessing. IT & Security | Legal & Compliance | Operations
Source: Techpresso, June 11, 2026
Apple
Apple used WWDC 2026 to give Siri its biggest overhaul since it launched in 2011 — and the most useful change isn’t the one getting the headlines.
Siri Finally Gets Its Own App — But Apple’s Real AI Advantage Is Something Else Entirely
Apple’s WWDC 2026 announcements split into two stories. The first is immediate and visible: Siri now has its own dedicated app for the first time. Instead of chats disappearing after each exchange, you can open the app, revisit past conversations, see summaries, and start new chats from one place. You can type, speak, or upload images and documents, and conversations sync privately across iPhone, iPad, Mac, Apple Watch, and Vision Pro via iCloud.
Alongside Siri, several everyday Apple tools got AI-assisted upgrades too:
Image Playground — now generates photorealistic images, not just illustrated styles
Liquid Glass — refined for better readability and contrast
Photos, Safari, Passwords, Shortcuts, and HomeKit — each picked up AI-assisted features across the system
The second story is the one worth paying closer attention to. Apple’s actual long-term advantage in AI isn’t any single feature — it’s what The Deep View calls Personal Context: AI that can act on the information already stored on your devices — email, texts, notes, photos, documents — processed either on-device or through Apple’s Private Cloud Compute, which is built so that even Apple can’t see your data.
The comparison that makes this click: when OpenClaw exploded in popularity earlier this year, it wasn’t because it was the most advanced AI agent available — it was because it had access to your entire computer, and that context made it dramatically more useful. Apple’s Personal Context aims at the same value, but packaged with the trust of a company whose business model doesn’t depend on knowing everything about you. Google’s closest equivalent, Personal Intelligence, works similarly but is built around your Gmail, Photos, and Search data — and Google’s ad-based business model makes some users warier of how that information gets used.
For Apple users, the practical takeaway: the new Siri app is worth opening this week just to see what it can do now. But the bigger shift — AI that can act on your personal information without that information leaving your control — is the one to watch as Apple builds it out across iOS 27. Operations | IT & Security | All Readers
Source: The Deep View · Mindstream, June 10-11, 2026
ChatGPT / OpenAI
If your organization is exploring AI agents that can complete purchases on your behalf, Visa just made that a lot more concrete.
Visa Is Plugging Its Payment Network Directly Into ChatGPT — AI Agents Will Be Able to Spend Money for You
Visa announced a partnership with OpenAI this week to integrate its payment network, credentialing, and security infrastructure directly into OpenAI’s agent system. The practical effect: AI agents will be able to complete purchases on your behalf, within limits you set in advance.
The guardrails are built in from Visa’s side:
Spending caps — set limits on how much an agent can spend
Merchant restrictions — limit which merchants an agent can transact with
Approval requirements — require your sign-off before certain purchases complete
Visa continues to handle fraud detection, chargebacks, and refunds the way it does for any other transaction. The companies say the setup could support everything from everyday shopping to business invoice payments to AI coding agents that need to purchase API access on their own.
Visa’s head of platform data and services, Rubail Birwadker, noted that more than one in five purchase-related transactions are already influenced by what people learn through conversations with LLMs — this integration is a step toward closing the loop between that research and the purchase itself.
For now, this is early-stage infrastructure rather than something you’ll use tomorrow. But if your organization is thinking about agentic procurement or AI-assisted purchasing workflows, this is the foundation those systems will likely be built on. Finance | Operations | IT & Security
Source: TLDR, June 11, 2026
Gemini / Google
Google quietly made one of its most capable AI models dramatically cheaper to use this week — and for anyone building with AI or running high-volume tasks, the price difference is hard to ignore.
Google Just Made Frontier AI Cheaper Than Lunch
Buried in this week’s Google AI news was a pricing update that deserves more attention than it got: Gemini’s API pricing dropped sharply, putting serious AI capability within reach at a cost that, for typical usage, comes in well under what most people spend on a midday meal.
This matters most for anyone running AI tasks at volume — batch processing documents, running AI across large datasets, or building tools that make many API calls. At the previous pricing, costs could add up quickly enough to make experimentation feel risky. At the new pricing, the calculus changes: it becomes much cheaper to test an idea, run a pilot, or process a backlog than it was last month.
For individual users on Gemini’s consumer apps, this pricing move doesn’t change what you pay directly — but it’s a signal of where the competitive pressure in AI is heading. As covered in Worth Watching this week, Anthropic and OpenAI are both reportedly considering price cuts of their own. Google moving first on price, while also shipping real capability, puts pressure on both competitors to respond.
The practical takeaway if you build anything on top of AI APIs: this is a good week to revisit your cost assumptions. A project that looked too expensive to pursue a month ago may now be well within budget. IT & Security | Operations | Finance
Source: Mindstream, June 11, 2026
Tools & Workflow
For anyone who follows a stack of AI and tech newsletters and still feels like they’re missing the important stuff — this reframes what newsletters are actually for.
Stop Reading Newsletters. Start Using Them as a Signal System.
AI Maker published a sharp reframe this week: most people treat their newsletter subscriptions as a reading list — more arrives every day than you can finish, so you either binge-read or feel guilty about your unread count. Neither approach works.
The alternative: build a small AI research agent that reads your newsletters for you, but with a specific job — not summarizing everything, but watching for repeated signals. When three different sources mention the same story, tool, or trend within a short window, that’s a signal worth your attention. When something appears once and never again, it probably wasn’t important.
The setup is straightforward: connect an AI tool to your email (or a folder where newsletters land), give it a standing instruction to track recurring topics across a rolling time window, and have it surface a short list — not a digest, just “these three things came up multiple times this week, here’s why each might matter to you.”
This is a genuine shift in how to relate to information overload. You’re not trying to read everything. You’re trying to notice what keeps coming back — because that’s usually the thing that’s actually changing. All Readers | Operations | Executive & Strategy
Source: AI Maker (aimaker.substack.com), June 11, 2026
For anyone who’s noticed that AI tends to give you its first idea rather than its best one — there’s a reason, and a fix.
Why Your AI Buries the Better Answer
AI Business Insights made an observation this week worth sitting with: when AI gives you a mediocre answer, the problem usually isn’t that the AI is wrong. It’s that it committed to an answer too early, before considering the alternatives that might have been better.
This happens because of how these models generate responses — token by token, in sequence, without the ability to go back and revise the opening direction once it’s set. The first plausible-sounding approach often wins, even when a better approach exists a few steps further into the model’s reasoning.
The fix is simple and works across nearly any AI tool: before accepting an answer to something that matters, ask the model to generate two or three alternative approaches before picking one — or ask it to critique its own first answer and identify what a better version might look like. This forces the model to explore more of the solution space instead of anchoring on whatever came first.
For anything beyond a quick factual lookup — strategy questions, drafts, plans, recommendations — this one extra step regularly surfaces a noticeably better answer than the first one offered. All Readers | Operations | Executive & Strategy
Source: AI Business Insights, June 10, 2026
Worth Watching
OpenAI is reportedly considering significant price cuts to compete with Anthropic — and the timing is revealing. A ZeroHedge piece making the rounds this week (and corroborated by Techpresso) reports that OpenAI is weighing “drastic” reductions to its token pricing, partly in anticipation that Anthropic will cut first. Sam Altman has called AI costs “a huge issue” for business customers — but cutting prices further squeezes margins for two companies that are already losing billions while preparing for IPOs. The pressure follows Anthropic’s Claude Code revenue surge, which briefly pushed Anthropic’s valuation past OpenAI’s.
SpaceX’s IPO was oversubscribed more than 4x ahead of its expected pricing this week — on track to be one of the largest IPOs ever. SpaceX is also the company powering Claude’s compute under a $1.25 billion/month deal with Anthropic, so its market debut is being watched closely as a bellwether for AI-linked infrastructure bets generally.
Both Anthropic and OpenAI have now confidentially filed paperwork with the SEC to prepare for a possible IPO — Anthropic first, with OpenAI following about a week later. Anthropic is currently valued around $965 billion and OpenAI above $850 billion; public filings will eventually reveal real revenue and cost figures for the first time.
Published by Independent Intelligence — an independent AI newsletter for people who use AI at work, not just read about it. Five newsletters, one Substack. Friday general edition + Wednesday deep-dives on Claude, ChatGPT, Gemini, and Perplexity. Forward freely.