July 29, 2026
|
|
Greetings, this week: Anthropic ships its most capable model yet at half the old price, Google's own usage data draws a hard line between where AI shows up and where it's actually doing the work, ChatGPT Health goes live for every US user, a practical fix for why your long Claude threads get slow and expensive, and Cohort 2 of our AI Workflow Accelerator opens for individuals.
|
|
In Today's Signal
|
| Signal |
Claude Opus 5 launches at half the price of Fable 5, beating GPT-5.6 Sol on FrontierBench |
|
| AI News |
Google's real AI usage data, and ChatGPT Health goes live for every US user |
|
| Steal This |
Why your long Claude threads get slow and expensive, and 5 fixes |
|
| Futurelab |
AI Workflow Accelerator: Cohort 2 opens for individuals |
|
|
|
| Signal of the Week |
 |
Anthropic launches Claude Opus 5, its most capable model yet, at half the price of Fable 5
| Opus 5 landed July 24 at $5 in / $25 out per million tokens in standard mode, roughly half what Fable 5 costs, with a faster mode at $10/$50 for teams that need speed over savings. |
| It carries a 1-million-token context window and a new low, medium, high effort toggle, so a quick task and a genuinely hard one no longer cost the same to run. |
| On Anthropic's own FrontierBench v0.1, it scored 43.3%, four points clear of OpenAI's GPT-5.6 Sol at 37.5%. |
| It ships three days after Kimi K3's weights went fully open and free, so for the first time this year, a CXO evaluating models has a real choice between paying for the best closed option or running a frontier-grade one for nothing. |
Futurelab Take
Half the price and a built-in dial for how hard the model should think, that's Anthropic pricing for a market where free frontier models now exist. This isn't a benchmark story. It's Anthropic conceding that cost, not capability, is what decides the next round of enterprise contracts.
Source: Claude Opus 5 launch coverage, July 24-25, 2026
|
|
| AI News That Matters |
| AI Adoption Data |
 |
Google's own data says AI has touched 68% of occupations, but only a fifth of the actual tasks within them
| Google's own ATLAS report, published July 23, found AI activity present in over 68% of occupations, occupations that together represent 88.4% of the US civilian workforce. |
| But the median occupation with any AI use shows that activity touching just 21% of its actual tasks, not the whole job, a slice of it. |
| Only 3% of occupations show AI touching more than three-quarters of their tasks, and full end-to-end automation attempts show up in fewer than 10% of conversations involving non-routine cognitive work. |
| The report is built on 14.7 million anonymised conversations across the Gemini app, AI Mode, and the Gemini API, spanning 800 occupations and 4,000 tasks across 150 countries, real usage data. |
Futurelab Take
Adoption looks enormous from a distance and looks ordinary up close, and both readings are true at once. If your team is planning around "AI is replacing roles," this data says plan around the 21%, that's where the real implementation work actually sits.
Source: "Google study finds broad AI use, but little evidence of job automation," Axios
|
| Consumer AI and Regulated Categories |
 |
OpenAI starts rolling out a dedicated Health feature inside ChatGPT
| OpenAI rolled Health out to every US user 18 and older, across Free, Go, Plus, and Pro, on web and iOS, starting July 23. |
| It lets people connect Apple Health, supported US hospital systems, One Medical, and Function Health, then compare lab results, track medications, and summarise what changed since a prior appointment. |
| The feature formalises something already happening at scale: ChatGPT was fielding roughly 300 million health-related queries a week, up from 230 million in January, before this rollout even started. |
| The launch landed one day after a Florida pastor sued OpenAI, alleging ChatGPT failed to tell him to see a doctor, timing OpenAI didn't choose but can't avoid being read against. |
Futurelab Take
Every regulated category a consumer AI product enters raises the compliance bar for everyone already operating in it. 300 million weekly health queries were happening with or without a dedicated feature, so the real story isn't the launch, it's OpenAI finally building guardrails around scale that already existed.
Source: "OpenAI makes ChatGPT Health available to all US users," TechCrunch
|
|
| Practical Tool - Steal This |
 |
| Productivity - Claude Context Management |
Claude is not getting slower. Your chat history is.
Every message in a long Claude conversation gets reread before the model answers your next one, old drafts, pasted research, screenshots, previous answers, all of it. That's why long threads feel sluggish and expensive, not because the model got worse.
| 1. | Start a new chat when the task, file, or audience changes, not just when the old one starts feeling slow. A stale thread does more damage than it looks like. |
| 2. | Before a thread gets long, ask Claude for a clean handoff summary, goal, decisions, open questions, next step, then continue from that instead of the full history. |
| 3. | Paste only the excerpt Claude actually needs. A two-page section with a clear ask beats a forty-page file with a vague one. |
| 4. | Split the work across chats: research in one, writing in another, review in a third, so Claude isn't carrying every rough note into the polished output. |
| 5. | Tell Claude the shape of the answer you want, short reply, bullet summary, table, 500 words, before you hit send. One line up front saves a much longer answer you didn't need. |
Why it works: none of this is about writing shorter prompts. It's about not making Claude reread work it's already finished with.
|
|
| From Futurelab |
 |
|
Cohort
AI Workflow Accelerator: Cohort 2 is Open
We're opening a second cohort of the AI Workflow Accelerator, a 4-week live program for individuals who want to move past casual prompting into workflows they'll actually reuse. It starts August 15, runs Saturday mornings, and every session is built around your own work, not sample exercises. Past cohorts have included people from PepsiCo, Coca-Cola, ITC, and Tata Capital, and you leave with one real, working AI system, not a certificate. Seats are limited, and referring someone who joins gets you a 10% bonus.
|
|
| Join the Conversation |
5,000+ professionals
Founders, business leaders, and practitioners navigating the AI transition together. No noise. No self-promotion. Just people serious about getting it right.
Join the Community
|
|
© 2026 Futurelab Studios. All rights reserved.
To unsubscribe please click here.
|
|