Point AI backward.
Count what already happened.

What actually changed

• Parrott used feedback analysis, a Peter Drucker practice that compares what someone expected with what actually happened.
• She supplied AI with the quarter's goals and the complete record of essays, projects, and experiments.
• In about ten minutes, the audit showed that she had reached goals she could not feel she had reached.
• The process also counted work that mattered even when it was added after the original plan.

Try this: Run a feedback-analysis pass

• Paste the goals you wrote for the quarter.
• Add the full, messy list of what you actually completed, started, changed, and stopped.
• Ask AI what you hit, what you missed, and what unplanned work still counts.
• Ask which one unfinished item is most worth carrying into the next quarter.

My take

I see this constantly, in therapy and in business both: people can list everything they did this quarter and still walk away feeling like they did nothing.

A review like this gives you what your memory will not, especially when you are tired: evidence. You let AI compare the plan with what happened, including the work that was added along the way.

You might still decide something needs to change next quarter. At least you will make that call from the full record instead of from the feeling that you somehow did none of it.

What actually changed

• The TNO study included roughly 60,000 workers.
• About one third expected AI to make their work mentally harder.
• When automation absorbed easier routine tasks, the work left for people required more sustained judgment and concentration.
• The small pauses once created by routine work could disappear along with the routine work itself.

Try this: Put one recovery break back

• Identify the routine task AI removed from one part of your day.
• Notice what dense decision-making now sits immediately before and after it.
• Add a short transition before the next judgment-heavy task instead of filling the saved minutes with more output.
• Treat the recovery time as part of the workflow for one week and see whether the end of the day feels different.

My take

So it is not just you. When AI takes the easy, routine tasks, what is left can be dense, judgment-heavy work, back to back, with none of the little breaks the easy work used to give you.

As a clinician, I would call that a cognitive-load problem. Adding more tools will not fix it. What helps is guarding the recovery time you quietly lost.

If you have felt more fried since AI showed up in your day, you are reading your own experience correctly.

What actually changed

• Flodesk launched Studio on June 30 as a free beta AI email-design tool.
• A user described the email and received a branded design in about five minutes.
• The finished design exported as clean HTML.
• The HTML could be placed into another email provider rather than requiring a move to Flodesk.

Try this: Design one email and keep your current sender

• Use the next newsletter or promotion you already need to send.
• Describe the audience, purpose, brand style, and required sections to Studio.
• Export the result and place it inside the email provider you already use.
• Check the links and mobile layout before sending. Do not move your list just to test the design tool.

My take

This was a real one, and it was free at publication. You typed what the email needed to say, and Studio designed a branded version instead of making you lose an hour to fiddling with blocks.

The part I liked was the clean HTML export. Even if you used Kit or Mailchimp instead of Flodesk, you could use Studio as the designer and paste the result wherever you already sent email.

What actually changed

• Gemini in Google Sheets analyzed a formula error and the surrounding data.
• It explained the problem in plain language and offered a corrected formula.
• The feature covered simple arithmetic as well as more complicated nested formulas.
• Higher usage limits through July 15 applied to paid Google plans, not free consumer Sheets.

Try this: Fix the sheet you have been avoiding

• Open the sheet with the formula error blocking real work.
• Ask Gemini why that cell is returning an error and what the corrected formula should be.
• Read the explanation before accepting the repair.
• Ask for a one-sentence explanation of why the fix worked so the correction is easier to recognize next time.

My take

I put off spreadsheets longer than I should, so this was one I would actually use. Open the sheet, ask why the formula is broken, let Gemini explain the fix, and move on.

You do not need a whole new spreadsheet system or an afternoon learning formulas. Use it for the one small technical thing blocking work you already need to finish.

What actually changed

• Kajabi introduced a context-aware assistant built into the account where the user's products and audience already lived.
• It asked about expertise, audience, and goals before suggesting product formats and pricing.
• It mapped a go-to-market sequence and guided the user through the work inside Kajabi.
• The useful advantage was access to current business context, not a universally better planning model.

Try this: Recreate the useful part in your current AI

• Write one page with your audience, current offers, prices, strongest customer questions, and delivery limits.
• Give that context file to Claude or ChatGPT before asking what to build next.
• Ask for three options tied to real customer evidence, not a generic launch plan.
• Check the recommendation against actual sales, member questions, and what you want to deliver.

My take

Of course Kajabi's advice felt sharper. It already had your products, prices, audience, and current offers in front of it.

That was the part worth noticing. You could get most of the way to the same quality in Claude or ChatGPT by giving it a current context file first.

I would still treat the recommendation as a starting point and check it against real customer questions, actual sales, and what you wanted to deliver.

What actually changed

• Fable 5 returned on July 1 after an eighteen-day suspension.
• Anthropic said the security issue that prompted the interruption had been patched.
• The return closed the June launch-and-pull cycle.
• Everyday users did not need to move a workflow or reorganize their tools.

My take

Remember the model the government pulled a couple of weeks earlier? It was back, the security hole was patched, and the entire to-do list for readers was nothing.

Tools come and go, sometimes overnight. The people who stayed calm were the ones keeping their knowledge in their own files instead of chasing whatever had just launched.

You did not miss anything by skipping Fable 5 in June. You still had not.

What actually changed

• Anthropic Academy released AI Capabilities and Limitations as a free course.
• It covered hallucinations, bias, prediction, knowledge boundaries, memory limits, and steerability in plain language.
• The material was useful beyond Claude because the limitations applied across generative AI tools.
• Access required a free Skilljar sign-up rather than an Anthropic account.

Try this: Leave with one verification rule

• Complete the short course once.
• Start with the section explaining confident wrong answers.
• Choose one rule you can apply every time, such as verifying every name, date, number, and source before sending.
• Add that rule to the workflow where a wrong answer would matter most.

My take

Most of us were never taught what AI is doing when it hands us an answer, so we are trying to judge the output without a useful frame.

This course looked like a solid place to get that frame without hours of technical material. Go through it once, choose one verification rule you will use, and add it to how you already work.

That one rule can be the difference between catching a wrong answer and sending it to a client.