Here is a number worth sitting with. 63% of Americans say AI is moving too fast, and yet about half of us now use it regularly, per new Pew data this week. So we are leaning on the very thing we say is racing ahead of us. I keep landing on this: the unease is not really about speed. It is about using a powerful tool without ever feeling like you are steering it. When was the last time AI made you feel like a passenger instead of the driver, and what was the one small thing that put your hand back on the wheel? Curious what has actually worked for you. Full thoughts on the blog: https://www.harvestkernel.com/most-americans-say-ai-moves-too-fast-fluency/

There is a difference between saving time and buying it back, and it is the whole ballgame.

Saving time means doing the same task faster. You still own it. Buying back time means the task leaves your plate for good. Most of us aim our AI tools at the first one. We write the email quicker, grade the stack quicker, build the deck quicker, and then quietly cram more into the same hours.

The shift that actually frees up a Sunday is small. Stop re-explaining your context every morning. Set it once. A Claude Project or a Gemini Gem can hold your role, your audience, your voice, and your standards so you edit instead of starting from a blank page. A tool like Zapier can quietly handle the copy-paste-move-file handoffs between your apps. You build it once, it runs forever.

One rule though: do not try to automate ten things. Stack one. And before you reclaim the hour, decide what it is for, because unassigned time always gets eaten.

So I am curious. What is the one repeating task in your week you would hand off first, and which tool would you point at it?

There is a difference between saving time and buying it back, and it is the whole ballgame.

Saving time means doing the same task faster. You still own it. Buying back time means the task leaves your plate for good. Most of us aim our AI tools at the first one. We write the email quicker, grade the stack quicker, build the deck quicker, and then quietly cram more into the same hours.

The shift that actually frees up a Sunday is small. Stop re-explaining your context every morning. Set it once. A Claude Project or a Gemini Gem can hold your role, your audience, your voice, and your standards so you edit instead of starting from a blank page. A tool like Zapier can quietly handle the copy-paste-move-file handoffs between your apps.

One rule though: do not try to automate ten things. Stack one. And before you reclaim the hour, decide what it is for, because unassigned time always gets eaten.

So I am curious. What is the one repeating task in your week you would hand off first, and which tool would you point at it?

Today's question. When you pick up a new AI tool, how do you actually check it before you trust it? Or do you just dive in and find out the hard way?

Here is the honest version most people never do: test it on your messy real work, not the demo. Watch what it does when it does not know. Decide how you would catch a quiet, confident, wrong answer.

Today's post digs into why that habit matters: the AI skill nobody talks about. And the new field test walks you through it in four steps: The AI Evaluation Field Test.

Drop one tool you rely on and your number-one way to catch it when it is wrong.

Something happened this past Friday that we will be referencing for a long time. The US government, through a Commerce Department export control directive, restricted foreign-national access to Anthropic's two most powerful models, Fable 5 and Mythos 5. To comply, Anthropic took both offline worldwide. It is the first time the government has forced a deployed AI model offline.

I want us to practice reading this calmly, the way we read any AI moment here.

The facts, neutral and sourced: Fable 5 launched June 9 as the most capable public model by independent benchmarks. The directive landed June 12. The order targeted foreign-national access, but the practical result was a global pause. The dispute centers on whether a jailbreak of the model's safeguards is serious. The government says yes. Anthropic says the evidence it saw was narrow and minor.

Here is the lens I use, and I would love your take. Three levers: Authority (what legal tool was used, here export control), Trigger (who set it in motion, reportedly an investor rather than a regulator), and Off-ramp (how it resolves).

So my question for the community: of those three levers, which one do you think matters most for the future of the tools we all use, and why? No wrong answers. I want to see how you are thinking about it.

---

Quick gut check for the group. PwC dropped a billion ad jobs barometer today showing the market splitting into two tracks, one that rewards human judgment with faster pay, one that does not.

In your own work right now, do you feel AI making you more valuable because you can do the judgment heavy part faster, or do you feel it quietly making your tasks easier for anyone to copy?

No wrong answer, and plenty of us are in both at once. Tell me which track a specific part of your week falls on.

Context: https://www.harvestkernel.com/ai-job-market-two-track-pwc-barometer/

Quick one for the lounge today. We spend a lot of energy on what AI does well. Let's flip it.

What is one task you handed to AI and then realized you gave away something you actually needed to keep? For me it is anything where the thinking WAS the point. A reflection, a hard decision. The output looked fine and I learned nothing from it.

Where is your line? And here is the real question: what did you take back from AI after trying to outsource it?

Today's teaching topic comes from Dario Amodei's brand new essay, Policy on the AI Exponential.

His central image stuck with me. He compares our institutions to Treebeard, the ancient tree from Lord of the Rings who is wise but so slow it takes him a full day just to say good morning. Meanwhile the forest is being cut down around him. AI is moving at the pace of axes. Our institutions move at the pace of seasons.

Higher education is a Treebeard. A new program can take two or three years to go from idea to first cohort, while the job market it prepares students for rewrites itself every few months. And the part of the essay that hit me hardest was not about regulation at all. It was his claim that the hardest problem in an AI economy is not money, it is helping people find meaning, purpose, and agency. That is education's job description.

So here is the question I am sitting with, and I want yours: if you could redesign one thing about how we teach or train people for an AI world, and you actually had the green light, what would you change first?

I also built a free four-phase toolkit to help you move from worrying about this to designing it: https://www.harvestkernel.com/design-future-of-learning/

Quick gut check for today. When you use AI to fact check or summarize, do you form your own answer first, or do you take the model's word and move on?

A new MIT study suggests the second habit quietly erodes your own ability to catch bad information. So I am curious where this community lands. What is one thing you still do manually on purpose, even though AI could do it for you?

Anthropic just asked the world to keep the option to pause frontier AI, in a report called "When AI Builds Itself." In that same report: Claude now writes 80%+ of their own code, and AI's success on hard open-ended problems jumped from about 25% to 76% in six months.

They wrote "we are not there yet." I think that's the scariest line in the whole thing, not the most reassuring one.

So I'm handing you the mic today:

When you read "we are not there yet," do you hear reassurance, or a countdown?

And the harder one: is a pause that only works if EVERY lab and EVERY country agrees and verifies it even possible anymore, or is that already a fantasy?

No wrong answers. I want your honest read. I even built a free four-phase Pause Debate Kit to help you form one in under 90 minutes, and I'm dropping it in the SeedStacking Resources space this week. Watch for it.