When we opened our Substack Live on Friday, Cory Blumenfeld had used AI seconds earlier to help answer a potential customer while he joined the call. Joel Salinas had it running in the background, reading a client email and attachment, while Philip Hofmacher had already checked in with his AI running coach about his ultramarathon training.
We’d gone live to talk about how much of our work AI should actually do, while all four of us were already giving it jobs that would have taken up our time a year or two ago. What became more interesting was how differently we treated those jobs once AI started doing them without us. 👇🏻
Start with the work you can check
Philip runs much of his business through Notion and now uses AI to help build pages and set up automations without learning every click himself. Cory uses it for repetitive work throughout his day, documenting conversations, updating his CRM, preparing emails and finding people he’s forgotten to reply to.
Cory is comfortable handing over work where he can quickly tell whether it’s been done properly. Summaries and moving information around are easy for him to check. Once the job needs his judgement, he gets involved.
Joel pushed further with some jobs. When he publishes on Substack, AI can turn the original piece into a LinkedIn post and carousel, then publish both without him seeing them first. He spent months testing the process and adjusting the rules before he was comfortable letting it run alone.
Trust has to be earned
Joel wrote the original material, knows what he expects from the LinkedIn version and has spent enough time testing the process to recognise where it can go wrong.
Yet when he was short on time with another post and let AI write the introduction, he published it, read it again and deleted the whole post about five minutes later. It had dealt with the SEO and keywords perfectly well, but it didn’t sound like him.
He understands the first job well enough to tell when AI has done it properly. That becomes much harder when you’re handing over work you don’t understand particularly well yourself.
Your name is still attached
Philip tried letting AI reply to his Substack messages early on and stopped within a week. Nobody complained. He simply didn’t like people believing they were having a conversation with him when they weren’t.
Joel draws a line around email and comments. AI can read the background, prepare a reply in Gmail and leave it there, but he still checks what goes out. He pointed out that you can spend years building trust and lose it quickly after a bad post or piece of advice goes out under your name.
The model isn’t dealing with the customer afterwards if an email gets something wrong, and it isn’t explaining bad advice to a client. You’re still the one who has to deal with it.
AI changes who you can hire
Philip raised an idea I hadn’t given enough thought to. He’d spent most of his time asking what AI could do for him, until hiring another virtual assistant made him consider what somebody else could now handle with AI helping them.
Previously, they’d struggled to give a VA a wider range of work because the company had too many moving parts to pick up quickly. With good documentation and the AI tools they’ve built, a new VA can take on more much sooner without knowing every corner of the business first.
That raised the question I kept coming back to during the conversation: what happens when AI helps you do work before you’ve learnt enough to judge the result?
The polished answer can still be wrong
If you don’t understand good writing, you’ll struggle to spot a weak draft. The same problem follows when you ask AI to communicate with customers or make decisions for you, especially now that the output is polished enough for a mistake in judgement to sit underneath perfectly reasonable wording.
Cory disagreed with part of that. Some of his early LinkedIn writing with AI was awful, he said, but using it got him writing at all. Once he’d started, he paid more attention to other writers, asked better questions and gradually improved. AI helped him try work he might never have taken on otherwise.
AI can help you try something before you’re particularly good at it, and that can become part of how you learn. The problem comes if you never develop enough judgement to stop accepting whatever it gives you.
That matters as you remove yourself from more of the work. I’m already giving AI jobs I wouldn’t have handed over a year ago, and I expect that list will keep growing.
Before I stop checking another one, I want to know I’d still recognise the bad answer.
