Work where the tools have no integration
Chapter five connected the systems that offer a connection. This one is about everything else, which in this industry is most of it.
Every media team runs on a set of tools that will never have an integration. The partner portal that shows delivery on screen and offers no export. The booking system with a form somebody fills in by hand every Tuesday. The publisher dashboard where the only way to get a number out is to read it and retype it.
Connectors do not help here, because there is nothing to connect to. The browser is the interface, and for a long time that meant a human had to be the one clicking.
That is the gap this chapter is about. Working inside the page itself: reading what is on screen, clicking, typing, moving between pages and filling in forms, using the logins you already have.
What it is genuinely good for
The honest framing is that this is not a better way to do work you can already do easily. It is the only way to do a specific category of work that has always been manual.
The test is simple: if you would have opened the tab and done it yourself with your own login, it is a candidate.
The rule before you point it at anything
This is the one part of the playbook where I would ask you to slow down.
Working in your browser means acting with your sessions and your permissions. Whatever you are logged into, it is logged into. Treat it the way you would treat handing your screen to somebody competent but new: fine for the reporting portal, not fine for anything with a payment step or a client-facing send button.
There are safeguards. It will generally stop and check with you before consequential actions, things like submitting a form, sending a message or downloading a file. Take that seriously as a checkpoint rather than a formality, and read what it is about to do before approving it. There is also a genuine risk that a page contains instructions aimed at the model rather than at you, which is a good reason to start on tools you already trust rather than on the open web.
Reporting portals, internal dashboards and read-heavy work: yes. Anything that spends money, signs something, or sends to a client: do it yourself. Not because it cannot, but because the review step is the entire value of a human being in that loop, and it costs you thirty seconds.
Why this is the last chapter
Because it is the most capable thing here and the least useful on its own.
Everything in this playbook has been building the same thing in different forms: context the model can act on. Instructions so it knows who you are. Projects so it knows the account. Staged builds so it works the way you work. Connectors and skills so it reaches your material and follows your steps. Browser work is what you get when all of that is in place and you finally point it at the one system nobody ever integrated.
Do it first, with none of the rest, and you have an expensive way to click buttons badly.
Where to start, if you start anywhere
Go back to chapter one and write the instructions box. It takes ten minutes, it is the least impressive item on the list, and it improves every single thing that follows it. The people who get real value out of these tools are almost never the ones who found a clever prompt. They are the ones who did the boring setup and then changed how they work.
Name the portal task you do by hand every week, the one with no export and no integration. Do that one, watching every step. If it works, you have just automated something that no amount of prompting could have touched.