Table of Contents
Do not index
Agents can now sign into your tools and keep working while your laptop is closed, which makes the delegation question urgent rather than theoretical. Agency owners keep asking me the same thing: what should I actually hand to an agent, and what do I keep? The line is not technical difficulty. It is reversibility. Give agents the work you can undo. Keep the work that becomes your reputation the moment it ships.
The trigger is xAI's Grok Bot, launched in early beta on August 11 and covered in The Agile Brand Guide's roundup. Each agent gets its own cloud computer that can sign into a user's tools, navigate their interfaces, run multi-step tasks and hand work off to other agents. The named use cases are lead research, invoice management, email and CRM. Per that reporting, "Because their environment runs in the cloud, Bots can continue working even when the user's computer is turned off." It is bundled with xAI's top subscription tier, reported at $120 per month.
Notice what is on that use case list. Research, invoicing, inbox, CRM. Not a single creative task. The vendor has already told you where the value sits, and the industry is going to spend the next year ignoring it in favor of demos about content generation.
This is for agency owners between $200k and $2M in revenue who are personally still doing admin they cannot bill for, ghostwriters charging $5k to $30k per month who run client operations out of a spreadsheet and their own memory, and founders with a team of three to ten where nobody formally owns operations. If you are the bottleneck on invoicing and onboarding and also the person who writes, this applies directly to your week.
This is not for solo operators with two clients. At that scale the overhead is small enough that supervising an agent costs more than the admin does. Skip this if your systems are not documented, because an agent inherits your process and an undocumented process is not a process. And if what you are looking for is something to write the content itself, this is not that article, and I am not going to pretend the two problems are the same.
What to automate first in an agency workflow
The sort I use is what I call the Reversibility Sort. Every recurring task goes into one of two piles. If the agent gets it wrong and you can fix it in under ten minutes with no outside party ever noticing, it is reversible and it is a candidate. If getting it wrong means a client reads something in your voice that you did not approve, or a number goes out that you now have to walk back, it is irreversible and it stays with a human.
Run your week through that filter and the results are unsurprising in the best way. Pulling company research before a call is reversible. Chasing an unpaid invoice through three follow-ups is reversible. Updating CRM stages after a call, formatting a proposal from an approved template, compiling last month's numbers into a client report, all reversible. Publishing a post, replying to a prospect in a founder's voice, deciding what a client's position on an industry story should be, none of those are.
For a 3 person agency, the reversible pile is usually eight to twelve hours a week per person, and it is the pile that has never been billable. That is the actual return on an agent. Not better content. Recovered capacity for the one thing you can only do slowly.
Why the creative half should stay manual longer than feels comfortable
Because the input is what you sell. A content operation is only defensible when the point of view inside it came from a real person, which is why the strategy layer has to be decided before any of it gets systematized. An agent running a documented workflow you designed is leverage. An agent deciding what your client believes is a liability wearing the client's name.
There is a second-order problem with cloud agents specifically. Work that continues while your computer is off is work you are not watching. That is exactly right for invoice follow-ups and exactly wrong for anything that touches an audience, because the failure mode is not an error message, it is something published in your name that you read for the first time after a client does.
The trajectory question is where your margin comes from in two years. Agencies that absorb agents into operations get a structural cost advantage on the unbillable half of the business and keep charging for judgment. Agencies that point agents at the creative half get a temporary speed advantage on the only work clients cannot get anywhere else, and then compete on price with everyone who did the same thing six weeks later. The sequencing is the whole strategy.
