Sponsored by
Hey folks,
Anthropic had quite a week. On Monday it shipped Claude Sonnet 5, pitched as a cheaper way to run agents and now the default model for free users. On Wednesday, Fable 5 came back online after 19 days in government-ordered timeout. Two models, opposite ends of the price list, sold on the same word: agents.
Every AI company is shouting agents these days; you brief it once, and it turns up with finished work on a schedule.
That's this week's skill. You'll pick 1 recurring chore, write a proper brief, and have your first agent running in about 15 minutes.
Let's dive in..
First, the comeback: Fable 5 is back online
3 weeks ago I wrote about Fable 5 getting pulled offline by the US government. Here's how it ended: the Commerce Department lifted the export controls, and Fable 5 is back worldwide. The fix was a new safety filter that blocks the jailbreak with 99% accuracy. The catch: more false positives on ordinary debugging work.
So what do you do with it now? Fable 5 is the delegation model. It's built for the job you'd hand a contractor for a week:
Tear through a folder of customer interviews and produce the report.
Run a multi-day research project across dozens of sources.
Turn messy raw data into a deliverable you review rather than rewrite.
Rebuild and audit your existing codebase.
The honest limits: it costs 2x Opus, and roughly 1 in 20 sessions gets quietly rerouted to Opus when the safety classifiers wake up. If your job is a recurring briefing, you don't need it.
Which is convenient. The skill that makes Fable worth its price is the same one this issue teaches on something much smaller: writing a brief good enough to hand off.
A scheduled agent is a prompt that runs itself. Write the instructions once, pick a time, and the AI delivers the output without you asking. All 3 major tools now have a version:
What you'll need (any one of these):
ChatGPT: Tasks. Lives in the "Scheduled" page in the sidebar.
Gemini: Scheduled actions. You prompt for the schedule.
Claude: Scheduled tasks in Cowork, via the desktop app.
What's still rough:
ChatGPT tasks are capped at once per hour and can't use your files or custom GPTs. Think reminders and monitoring, not heavy lifting.
Claude's scheduled tasks only run while your computer is awake with the desktop app open. Asleep at run time means it catches up when you're back.
Gemini caps you at 10 active scheduled actions and needs activity history switched on.
None of these will run your business unattended. They will delete the 20-minute chores you repeat every week. That's the win we're after.
How AI-Era Pricing Is Reshaping Finance Operations
Usage-based and hybrid pricing models are changing how B2B companies generate revenue — and creating new headaches for the finance teams behind them.
Tabs co-founder Rebecca Schwartz and PwC Partner Amit Dhir sat down to unpack exactly what that means in practice: how pricing model decisions ripple into revenue recognition, forecasting, and financial ops — and what it takes to scale without piling on manual work.
Watch the on-demand recording to get practical frameworks, real-world examples, and a clear path to operationalizing usage-based revenue — including a forward-looking take on how AI will reshape financial workflows. If your team is navigating pricing complexity heading into the back half of the year, this is worth an hour.
The walkthrough
Step 1: pick a chore an agent can actually do
Most people get this backwards. They pick the most impressive-sounding job (analyse my industry!) and get mush back. The right first chore is boring. It passes 3 tests:
It recurs. Same job, same day, every week.
The steps never change. You could write them on an index card.
You read the output, you don't edit it. If you'd rewrite half of it anyway, the agent saves you nothing.
Good first agents: a Monday briefing of your calendar and priorities, a Friday summary of what shipped this week, a weekly digest of news on 1 narrow topic, a monthly reminder that assembles your invoice list.
Bad first agents: anything with judgment calls, anything client-facing, anything you'd be embarrassed to have wrong.
Watch out: don't schedule an agent to send anything. Have it prepare drafts and briefings for you to review. You stay the last step.
Step 2: write the brief like a contractor brief
The difference between a useless agent and a good one is almost never the model. It's the brief. You're writing instructions for someone who will do this job 50 times without ever asking a follow-up question, so ambiguity compounds.
Here's the template I use, filled in with a real example: the brief behind the agent that scouts topics for this newsletter (the one in the screenshot above).
You are my newsletter topic scout for Beyond the Buzz Weekly, a weekly AI newsletter for product people and knowledge workers. Every Thursday at 8am, do this job: prepare topic options for next Monday's issue.
Pull from: my "AI newsletters" Gmail label (past 7 days); aireleasetracker.com to verify every model name and release date; my topic backlog file; and the titles of my last 6 published issues so nothing repeats.
Output: a 1-line summary of the week's dominant story, then 3 ranked topic options. For each: the angle in 1 line, why this week, the format it fits, and a suggested subject line in sentence case. End with "My pick: [option] because [reason]."
Rules: every topic must solve a specific, common problem a reader can implement in one sitting. Tool-agnostic, no vendor cheerleading. Skip anything too broad, too niche, theory-heavy, or so news-dependent it dies in a week. Never trust release dates from roundup newsletters; if aireleasetracker.com can't confirm one, leave it out. Flag any backlog topic this week's news suddenly made timely.
If you can't reach a source or something looks unusual, say so at the top. Don't guess.Your job will look different, but the 5-part shape is the template: who you are and when you run, where to pull from, what to output, what rules apply, and what to do when something's off. Swap in your own chore and keep the shape.
That last line matters more than it looks. Agents run unsupervised, and an agent that quietly guesses is worse than no agent. Give it permission to flag problems instead.
Notice how many of my rules are scar tissue. The release-date rule exists because roundup newsletters kept getting dates wrong: one week in June, 4 wrong dates and a model that didn't exist. Your brief will grow the same way. Start simple, add a rule every time the output disappoints you.
Pro tip: write the output format first. When you know exactly what you want to receive, the rest of the brief writes itself.
Step 3: set the schedule in your tool
Each tool takes the same brief. The setup differs slightly:
ChatGPT: open the Scheduled page in the sidebar, create a new task, paste your brief, pick the cadence. Or just type the brief in a normal chat and add "run this every Monday at 7am." Turn on notifications or you'll never see the output.
Gemini: paste the brief at gemini.google.com with the schedule in the prompt ("every Monday at 7am"). Gemini replies with a summary of what it scheduled. Manage or pause it under Settings and help, then Scheduled actions.
Claude: in Cowork on the desktop app, describe the task and cadence. Claude confirms the schedule. Because Cowork can touch your actual files and connected tools, this is the one that can produce finished documents rather than just messages.
Watch out: schedule the run for when your first coffee happens, not midnight. You want to read the output while the fix is cheap, and (for Claude) your laptop needs to be awake anyway.
Step 4: supervise the first run
Before you trust the schedule, run the brief once manually in a normal chat. Read the output like an editor. 3 questions:
Did it look at the right sources?
Is the format what you asked for?
Would you have caught a mistake if there was one?
The first run almost always surfaces a gap in the brief. Mine kept including newsletter promo emails in my inbox digest until I added 1 exclusion line. That's normal. Fix the brief, not the output.
Pro tip: ask the agent to end every run with 1 line: "Sources I checked: X, Y, Z." Cheap insurance against silent failures.
Step 5: review at 2 weeks, then tighten or kill
Put a note in your calendar for 2 weeks out. Ask 1 question: did I actually read the outputs? If yes, tighten the brief with everything you've learned and consider a second agent. If no, kill it without guilt. An agent nobody reads is just warm air in a data centre.
The 2-week review is also where you catch drift. Agents don't get worse, but your needs move, and a brief written in July can be quietly useless by October.
How this all connects
The skill you just practised is going to be everything for the next few years of AI. A scheduled task is the training wheels version of delegation. The same brief structure works when you hand an AI a 1-off big job, and it's why the models launched this week were all sold on agent performance rather than benchmark scores.
It also chains upward. Once your Monday briefing runs itself, the same brief structure scales to bigger jobs, and eventually to the Fable-size ones: week-long projects you hand off entirely and review at the end. The delegation ladder starts with a scheduled summary and ends with an AI you trust with real work. Every rung is the same skill.
And if the "new model every week" noise tempts you to re-platform your agents each launch, don't. I wrote about reading launches without getting played 2 weeks ago. The brief is portable. The habit is the asset, not the model underneath it.
My definitive take: 12 months from now you'll spend more time reading AI output than requesting it. The people who win that shift are the ones who learn to brief, starting with something as small as a Monday morning summary.
Links worth your click
Try this now
Open whichever AI you already pay for and paste this:
You are my Monday briefing assistant. Every Monday at 7:30am, do this job: prepare my week-ahead briefing.
Pull from: my calendar for the next 7 days [adjust to what your tool can access].
Output: 5 bullets. Top 3 commitments, 1 conflict or crunch point, 1 thing I should prepare before Tuesday.
Rules: skip routine recurring meetings. If you can't see my calendar, say so at the top and build the briefing from what I paste below instead.
Run it once now so I can check the format.Read the output, fix 1 thing in the brief, then schedule it. That's your first agent.
👉️ Know someone who'd use this? Forward it to them. When 1 person subscribes from your unique referral link, you unlock the Beyond the Buzz Prompt Vault: 10 copy-paste prompts that do real work.
Hit reply and tell me: what's the 1 recurring chore you'd hand off first? The best answers become worked examples in a future issue. I read every reply.
Cheers, Tim
