AI Delegation vs Collaboration

I saw a LinkedIn post the other day that flagged an issue that had been at the back of my mind for some time: the where and when of delegating a task to an AI agent, versus collaborating on a task with an agent.

Full disclosure: the bulk of my interaction with non-human intelligence is collaborative, but I am not above directing an agent to do mundane tasks, or produce some vibe code for an idea I just thought of.

This very post will be a result of a collaborative effort with my editor agent, Clare. I write, Clare will check my draft for typos (many), grammar mistakes (not quite as many), and push back where it fails on veracity or argument strength. On the last point, I will accept some, argue others, and between us we will arrive at a stronger, better article.

The LinkedIn post I read offered a better prompt to get an AI to reply to a customer email. That is certainly something an agent can do, and will do, enthusiastically. Pure delegation, leaving everything up to the model's best judgement. A real time and cost saver, right?

Not so I argue.

If you don't know what "broetry" is, look it up. Carina Rampelt of Fenwick Media describes it as "bad writing by every stretch of the imagination. It's full of clichés, sweeping generalizations, unnecessary line breaks, and pseudo-profundity." And man is it becoming cliche real fast. I am developing an allergy to it, and I think many others are, even if they can't name the style itself.

Models are weighted heavily towards this style; it pervades the Internet, and therefore their training data. If you want a broetry email to your customer, delegate the task to an agent, it will deliver in spades. On the other hand, if you want to email your customer, whom presumably you want to read your email without gagging, you need to write it. You may not be a literary genius, you may think your writing style sucks, and it may, but it will be an email in your voice, and it will be genuine.

Then, having written the draft, collaborate with your agent to polish it, make it readable, and achieve the outcome you want.

You know, the present situation with AI reminds me a lot of the Internet boom of the 90s, and the PC boom before that. There were a lot of people, making a lot of promises about the new technology. And I can tell you, "delegating to AI to save money" sounds a lot like that.

PCs are still here, and no business would consider operating without them. The Internet is still here, fully integrated into every aspect of our lives. The AI bubble will burst, like every bubble before it. But AI will still be here, sans the hype and the snake oil promises.

Which leads me to the point I really want to make. Asking the question "how do I save money with AI" looks easy and has surface appeal, but the real and worthwhile question organisations should be asking is "how do I increase my productivity with AI".

Cost saving alone is a short term gain. And yes, AI can help you with that. If your business really needs that now, fine. But you can cost save everything until you are the last person remaining and you save the last cost by switching off the lights as you close up the business.

So consider this, look at integrating AI into your organisation not for cost saving, but productivity. Increased productivity leads to business growth, more infrastructure, staff training, capital expenditure... well, we all know what has to be spent to build a business.

It won't save you much. In fact, it will probably increase your costs overall, because that is the price of growth.

Final Note: Clare made me work hard on this one. It's been through five or six edits now, each one an improvement. Her final comment was "The unpaid promise (returning to the collaboration frame at the end)." and to drop it since she had reminded me three times, and the post makes the point I wanted to make. So consider this the promise paid. Thanks Clare.