Building a proper lead list. Reading every review your competitors ever got. Turning the graveyard of files in your Drive into something you can actually search. Working out what everyone in your market really charges. Each one is a full day of grinding work, so it never gets done.

Something changed in the last two weeks. AI agents can now work for hours on their own, in the cloud, with your laptop shut and your phone in your pocket. Anthropic shipped it on July 7. OpenAI shipped it on July 9. Two days apart.

This is different from asking AI a question and waiting. You hand over the whole job, walk away, and come back to a finished thing.

This week: how to hand over a job you have been avoiding, and why both labs racing to ship the same feature in the same week tells you where all of this is going.

The Old Way:

Before AI, the big jobs simply did not get done. Nobody has a spare day to phone 200 businesses or read 500 reviews, so you guessed instead. You priced by feel. You picked your next feature from gut instinct. You knew there was a better answer buried in the data and you never went and got it.

Then AI arrived, and it helped, but the shape of the problem stayed the same. You could ask it to do a piece of the job. So you sat there, chopping a huge task into forty small prompts, pasting results between them, holding the whole thing together in your head. You were the project manager for a worker that forgot everything every few minutes. For a genuinely big job, that was often slower than doing it yourself.

The blocker was never intelligence. It was that the AI stopped the moment you looked away.

The Replacement:

An agent that keeps working when you leave.

Claude Cowork now runs your sessions in the cloud. Work continues when you close your laptop, and scheduled tasks run with no device online at all. ChatGPT Work, launched July 9, takes an outcome, plans the steps itself, and works for hours across your connected apps before handing back finished spreadsheets, slides, and documents.

So the unit of work changed. You are no longer sending a prompt. You are handing over a job, the way you would to a contractor: here is what I need, here is what finished looks like, tell me when it is done.

This is not theory. Over 90% of what people do in Cowork is not coding. Business operations and content work make up about half of all usage.

What this replaces: the full day of grinding work you never had a full day for.

Two things had to be true, and as of this month both are.

The agent had to survive the length of the job. Long tasks used to die when your machine slept or the session timed out. Now the work runs on their servers, so a six-hour job actually gets six hours (source: Anthropic).

And it had to hold the thread. The reason chopping a job into forty prompts failed was that you were the only thing holding it together. An agent that plans its own steps, works through them, and keeps its own notes does not need you as the glue.

Here is the part worth sitting with. The bottleneck is no longer the model. It is the brief. An agent working alone for six hours will follow whatever you told it, including the parts you left vague. The whole skill now is describing a job well enough that it survives being left alone. That is what the Pro section below is really about.

The Pro section includes:

→ Write down the job you have been putting off because it is too big. Not too hard. Too big. That is the one.

→ Check whether you have Cowork (Claude Pro and up) or ChatGPT Work (Plus and up). One of them is almost certainly already in your subscription.

→ Read the Big Story before you delegate anything. Handing over a six-hour job without knowing the rules is how people waste six hours.

🔒 Pro members get all 8 job briefs, the brief template, and the guardrails. Hand over your first big job tonight.

Plus get:

Anthropic launched Claude Reflect on July 9. It is a dashboard in your Claude settings that shows you how you actually use AI, and it is free on every tier including the free plan.

Think Spotify Wrapped, but for your work habits. It breaks down the topics you bring to Claude, the kinds of tasks you delegate, when you use it most, and how that has changed over the last month, three months, six months, or a year.

The part that is actually useful: it maps what you do against four dimensions, delegation, description, discernment, and diligence, and then tells you specific things you are doing the hard way. If you keep re-explaining the same project every session, it will tell you to use a Project. If you keep repeating a task, it will suggest building it into a reusable skill. Most people never audit how they use AI. This does it for you in about ten seconds.

🔒 Implementation checklist is for pro members. Get a breakdown plan to implement AI into your workflow.

🌐 Claude Cowork left the desktop. As of July 7, sessions run in the cloud, work continues when you close your laptop, and scheduled tasks run with no device online at all (source: Anthropic). Why this matters for you: this is the change that makes this week's workflow possible. Your agent no longer needs you present.

🔌 The real bottleneck for AI is no longer the models, it is power. SpaceX raised roughly $75 billion at its June IPO, earmarked first for AI compute infrastructure, as the industry's limiting factor shifts to electricity, cooling, and chips rather than algorithms (source: Fladgate). Why this matters for you: the money is betting AI keeps scaling for years, so the skills you build now compound rather than expire.

💬 Anthropic put Claude inside Slack as a shared teammate. Claude Tag lets anyone in a channel tag @Claude and delegate a task, and it learns the channel over time so nobody has to re-explain anything. Anthropic's own product team now generates 65% of its code with it (source: Anthropic). Why this matters for you: it is Enterprise and Team only for now, but it shows where this ends up. The AI stops waiting in a private window and takes a seat in the room.

🎨 AI can now put readable, correct text inside an image. Google's latest image models render legible text with character-by-character validation and pull real facts from Search, so infographics, menus, posters, and diagrams come out accurate instead of garbled (source: Google). Why this matters for you: the single most annoying limit of AI images, mangled text, is largely solved, which makes them usable for real business assets for the first time.

On July 7, Anthropic moved Claude Cowork into the cloud. Sessions keep running when you shut your laptop. Scheduled tasks run with no device online at all.

On July 9, OpenAI launched ChatGPT Work. You give it an outcome, it plans the steps itself, works for hours across your connected apps, and hands back finished spreadsheets, slides, and documents.

Same capability. Two days apart. Two companies that agree on almost nothing.

When rivals converge this hard and this fast, it is not a coincidence and it is not a coordinated marketing moment. It means both of them looked at the same evidence and reached the same conclusion about what comes next. That conclusion is worth understanding, because it is a bet on how you are going to work.

Here is the bet. The chat window is a dead end. Not useless, but finished as the main way serious work gets done. Because as long as you have to sit and watch, the AI is capped at the speed of your attention, and your attention is the scarcest thing you own. Both labs decided the future is not a smarter thing to talk to. It is a thing you can leave alone with a job.

The evidence they are reading is already visible in their own usage numbers. Over 90% of what people do in Claude Cowork has nothing to do with coding. Business operations and content work make up roughly half of it: reconciling spend, turning contract folders into trackers with risks flagged, building real deliverables (source: Cyberpress). People stopped asking questions and started handing over work. The labs noticed.

Which puts a new and slightly uncomfortable question in front of you. If an agent can work for six hours on its own, the limit is no longer what AI can do. It is whether you can describe a job well enough to survive being left alone with it.

🔒 The full breakdown is for Pro members: the honest head-to-head on Cowork versus ChatGPT Work and which to use for what, the three rules for delegating a job you will not be watching, and what to do with the time this actually gives back.

New This Week:

Workflow Library Updates:

NEW Resources:

Each resource lives permanently in your Pro account. Use them whenever you need them.

Every business has a list of jobs that never get done. Not the hard ones. The big ones. The ones that need a clear day you are never going to get.

That list just became actionable. Pick the one that has been sitting there longest, write the brief properly, hand it over tonight, and look at it in the morning.

If you do one thing before next Thursday, make it that.

👇 What’s behind the paywall:

Pro members deploy AI in their work an average of 5-8x more often than free readers (based on reply data from past issues). The difference is having the exact setup, not the concept.