You've might've been using AI like a slow intern. One instruction, wait, check, next instruction. That hand-holding is most of the time it was supposed to save you.
As of this week, the tool outgrew the babysitting.
๐ฏ THE STRATEGIC SHIFT
What happened. On June 30, 2026, Anthropic shipped Claude Sonnet 5 and made it the default on the Free and Pro plans, and available across the paid tiers. (Anthropic) It runs close to the quality of the top-tier model at a fraction of the cost. The part that matters for how you work: testers reported it finishes multi-step tasks that earlier versions left half-done, and it reviews its own output without being told to. (TechCrunch)
Why this lands for you. You're the whole company. When the AI only handles one step at a time, you're still standing over its shoulder running the project, and that supervision is where your hour really goes. A model that takes a whole task and catches its own mistakes moves you from operating the tool to editing the result. Same work, a lot less of you inside it.
What I'd do this week. Pick one task you currently walk the AI through step by step. Hand it the entire job in a single brief, tell it to check its own work, and only step in at the end.
๐ ๏ธ THE PLAYBOOK
Today's move: take a task you already do with AI and rewrite it as one outcome brief. About an hour, and you'll feel the difference on the first run.
- Pick the task you babysit most. Turning a newsletter into a week of posts, a call transcript into a follow-up, one offer into a run of emails. Something you already do by feeding the AI one instruction at a time.
- Write the outcome, not the steps. In one message, describe the finished result you want, hand over the inputs, name the format and the limits, and end with this line: "Check your own work before you show me, and flag anything you're unsure about." You're briefing it like a capable teammate, not pressing one button at a time.
- Run it on the free default and time yourself. Read the result once, fix the voice, ship. The metric you can see today: the job that used to eat an afternoon done in one sitting, and the pile of back-and-forth messages you just cut to near zero.
๐ฌ THE EXPERIMENT
This changes how you work from here. Build a reusable job brief for the thing you repeat most, so you stop rewriting the ask every time.
Honest hypothesis: the more you describe the outcome and the limits up front, the less you babysit, and the closer the first draft comes to finished. Not a promise it's perfect. A bet that your review time drops.
- Take a brief you wrote before and save it as a template, with blanks for the inputs that change each time.
- Run it three times this week on real work.
- After each run, note the one thing the model got wrong, and add a line to the brief that prevents it next time.
- By the third run you should have a brief that hands back near-final work you only edit.
Measure: your own review time across the three runs, and whether run three needed fewer fixes than run one. That's visible inside a week. No rankings, no citations, nothing you have to take on faith. Just less time spent and cleaner drafts.
CMO Note
In a marketing org, the reason a senior person costs more was never the hands on keyboard. It's that you hand them an outcome and trust them to run it and check themselves. That is the part that just got cheap. Treat the model like a capable teammate you brief once, not a machine you feed one line at a time, and the time it gives back moves from minutes to hours. On the free plan the spend is closer to zero, so the only real cost is changing the habit.
Have a great weekend.
Tuck
Marketing AI Playbook
โ
โPS. Fable 5 is back for a limited time and in paid accounts but only til July 7. Go check it out.