AI saved time.
Go one layer deeper.

What actually changed

• Bozer and Kotte interviewed 12 coaches at one firm that had adopted AI coaching tools.
• The coaches stopped dividing the work into human versus AI and started dividing it into the who and the how.
• AI took the how: structure, frameworks, and the first pass.
• The coach stayed the who: the read on the person and the judgment about what actually mattered.

Think about this

• Where in your work are you the how, and where are you the who? Hand more of the structure and first pass to AI, and protect the human read.
• When the am-I-still-needed feeling appears, name which part of the job it is actually about. It is often the how, and the how was never the only reason people came to you.

My take

This study is small, twelve coaches at one firm, so hold it loosely as evidence. But it names out loud the thing I hear from members constantly: if AI can do this, am I still needed?

What these coaches landed on after sitting in that fear is worth borrowing. They stopped dividing the work into human versus AI and started dividing it into the who and the how. AI is good at the how: the structure, the frameworks, the first pass. You are the who: the read on the person in front of you, the call about what actually matters here. Nobody is automating that.

If the am-I-still-needed question has been sitting in your chest, that is the distinction I would hold it up against.

What actually changed

• Sixty-nine percent said AI saved them time, and adoption rose from 38% to 82% in one year.
• Among frequent users, only about a third used AI for operations and about a third for bookkeeping.
• Women who used AI in finance and operations were about 1.8 times more likely to report higher revenue.
• The report called the gap adopted, not embedded.

Try this: Take one real part of the business one layer deeper

• Pick one job you currently do by hand or dread, such as invoicing, client onboarding, or a bookkeeping tidy-up.
• Hand AI the real version of it, not a toy task. For example, ask it to categorize a messy monthly expense list, flag anything odd, and identify the three largest costs.
• If it saves real time, turn it into a repeatable step with the same inputs, prompt, review, and schedule.

My take

Here is the number that stopped me: 69% of the women business owners in this study said AI saved them time, but only about a third used it for their actual operations, and about a third for their books. The tool was in the door, but parked in the easy jobs.

The women who pushed AI into finance and operations were almost twice as likely to say their revenue went up. I think many of us stay shallow for a plain reason: going one layer deeper feels like it needs a permission we never gave ourselves.

So here is the permission. Pick one real part of your business this week, the invoicing, the client onboarding, or the task you dread, and take AI past the easy version of it.

What actually changed

• Sonnet 5 became the default on Free and Pro on June 30.
• Introductory API pricing was $2 per million input tokens and $10 per million output tokens through August 31, before rising to $3 and $15. This affected API builders, not ordinary app subscriptions.
• A new tokenizer made the change closer to cost-neutral than genuinely cheaper.
• It was the second flagship shift in six weeks after Fable 5 arrived, disappeared, and returned.

Think about this

• How closely do you need to follow model changes to run your business? For many readers, the honest answer was not very closely.
• Some people tracked Fable 5 in real time, some were still waiting, and some shrugged. Which response was the calmest version of you?

My take

If you use Claude, the version you opened this week is quietly a better one. Anthropic made Sonnet 5 the default on the free and Pro plans on June 30, and you did not have to do anything.

The introductory API pricing used a new tokenizer, so the change was closer to cost-neutral than genuinely cheaper. For most people in the app, that did not matter.

Not rushing to keep up is its own kind of skill. The tools will keep churning, and you get to choose how closely you follow along.

What actually changed

• As of June 15, Zapier priced AI steps in three tiers: Standard at 1x, Advanced at 3x, and Premium at 5x.
• New AI steps defaulted to Advanced, so an unwatched automation could cost triple.
• A safeguard paused a Zap at 75 tasks in one run.
• The change affected Professional, Team, and Enterprise plans.

Try this: Check the AI steps

• Open Zaps that draft, summarize, or classify text.
• Check the tier selected on each AI step.
• Test Standard anywhere the output does not need the Advanced model. Keep the cheaper tier only when the result still works.

My take

This was the one to actually check because nobody emailed users about it. Zapier changed how it charged for AI steps, and new steps defaulted to Advanced.

If you were building Zaps without watching the tier, you could pay triple for work Standard handled fine. The useful move was a quick review, not a rebuild.

What actually changed

• The universal connector used one link and took about a minute to set up.
• The assistant could read and write: pull metrics, draft broadcasts and sequences, and tag subscribers.
• Users set the permissions during connection and could revoke them later.
• The connector required a paid Kit plan.

Try this: Start with a read-only question

• Connect Kit and review the permissions before allowing changes.
• Ask a read-only question first, such as which of the last ten broadcasts had the best click rate and what they shared.
• Only then ask for a draft, and review it before anything is sent.

My take

Kit's connector was the cleanest example yet of where this was heading. One link could let an assistant work with your list, offers, and history.

The thing to flag was that it could make real changes. Set the permissions, know how to revoke them, and review before sending.

Connectors turn your own context into something you can hand to an assistant quickly. That context is the part that compounds and travels with you.

What actually changed

• Riverside 2.0 launched June 30 with a rebuilt Studio and AI Editor, and the company moved from riverside.fm to riverside.com.
• In-person multicam, newsletter-from-recording, social scheduling, and motion graphics were live.
• The MCP assistant remained on a waitlist.
• Features were rolling out over roughly three weeks, so they were not in every account yet.

Try this: Turn one recording into more than one thing

• Record or upload one useful talk on a question the audience keeps asking.
• Use the AI editor to cut it down and create a newsletter from the same source.
• Schedule a few short clips from the session instead of exporting them into another tool.

My take

The AI editor and newsletter-from-recording feature were the parts I would try first. The useful promise was fewer file handoffs, not more content for its own sake.

Two caveats mattered: the rollout was gradual, and the assistant was still a waitlist. If distribution was the thing eating your week, it was worth checking. If you did not make audio or video, skip it with a clear conscience.