Big Stories
This is the Siri people have wanted for years: personal context, on-screen awareness, app actions, and actual conversational answers.
Apple is finally turning the iPhone into the context layer for a private AI assistant.
If this works, it may become the first AI assistant normal people use every day without thinking of it as “using AI.”
Anthropic’s Fable/Mythos release turned into an export-control fight
I tried Fable for a few days and thought it was awesome, especially for long-running tasks.
When prompted well, it could build surprisingly complete things in one shot. It felt like a step change in agentic coding.
One side note is that Anthropic tightened the model hard: biology, cybersecurity, and LLM development got restricted, and the LLM research crowd was furious.
A few days later, the U.S. government moved to block foreign access to Anthropic’s most powerful models, and Anthropic took Fable and Mythos offline for everyone.
David Sacks on the Fable shutdown
Anthropic has loudly asked for regulation and has publicly argued that frontier models can be dangerous.
At the same time, I do not like the precedent of the government deciding which AI models builders can and cannot use.
This makes open source models more important.
Podcasts / Videos
This was a great builder/operator conversation on what actually changes when everyone can build.
The strongest idea: executives may need to allocate tokens like capital. AI spend becomes a management decision, not just an engineering line item.
Also liked the framing around 100x engineers, load-bearing talent, open source replacing most frontier-model work, and elite companies treating employees more like NBA athletes.
Tyler Cowen and Alex Tabarrok talk to OpenAI
The useful economist framing: do not just ask “which jobs disappear?” Ask what new coordination, capital, and demand become possible.
Growth is the master variable. If AI raises productivity, the second-order effects are bigger than the first-order job panic.
For builders, the takeaway is simple: look for places where intelligence was scarce and workflows were bottlenecked by expert time.
Tweets
Finn Mallery on how good 2026 is for starting a company
One person can now ship apps, make content, automate ops, research markets, and test distribution with a stack of AI tools.
The bar for starting is collapsing. The bar for taste, persistence, and choosing the right problem is going up.
Satya Nadella on the future of the firm
27 million views and counting.
The big idea is that companies need to compound human capital and “token capital” together.
In plain English: the winners will build learning loops where people, internal data, tools, and models keep improving each other.
Fusion is a multi-model router: different models look at the task, a judge compares the answers, and the system writes from that analysis.
OpenRouter claims Fable-level intelligence at about half the price.
The important idea is that model selection may become dynamic. The best answer might come from a small council of models, not one giant model.
A good thread on building vertical agents
The winning vertical agent is not just “ChatGPT for X.”
It needs domain context, state, tool access, evaluation, and a narrow workflow where being better actually matters.
This matches what I keep seeing: the agent wrapper is easy; the workflow and feedback loop are the hard part.
John Suh on rebuilding companies around a single timeline
His argument: AI-native companies may need one retrievable timeline across observability, product metrics, file changes, Slack, Claude Code, Codex, and decisions.
That sounds extreme until you actually try to make agents useful inside a company.
Agents need the why, not just the what. They need the history of decisions, rollbacks, exceptions, and tradeoffs.
Pankaj used Fable to build a coworker stress leaderboard
He connected WHOOP data to his work calendar to figure out which meetings and coworkers spiked his heart rate.
Funny, slightly cursed, and also a real preview of where personal data plus agents can go.
The future of quantified work is going to create some very weird product ideas.