Table of Contents
It is no longer surprising to receive a completely AI-generated email. It is upsetting, but depending on who is sending it, I might skim it or ignore it.
The larger problem emerges when AI-generated work becomes part of an entire workflow.
The AI-generated content chain
Let me explain what I have been experiencing:
- A sales representative invites me to a meeting with a client. To prepare for it, they send me an AI-generated summary of a few other meetings I did not attend.
- I am tasked with joining the meeting, taking notes, and then using a Claude skill to generate a document from it. (More on this later.)
- The sales representative receives my AI-generated document and, of course, instead of reading and understanding it, generates a summary using Claude.
- That AI-generated summary then becomes the basis for an AI-generated business proposal.
- They use AI to generate the email that sends the AI-generated proposal back to the client.
- The client reads the proposal and has follow-up questions about the infrastructure.
- The sales representative creates an AI-generated deck explaining the infrastructure, despite knowing nothing about it.
- A few weeks later, they tell me that they have no idea what the deck means. I try to figure it out, but I cannot change it because it has already been shown to the client.
- The client accepts the proposal. Since they have no software developers, I assume they either did their best to understand it or accepted it without understanding it because it looked nice.
The chain is now complete.
This is insane.
My survival strategy when AI use is mandatory
What happens when you are tasked with working like this? What happens when your manager literally tells you:
Use this Claude skill that I created. Just feed it all this information, and it will generate the document.
Well, over the years, I have learned that doing too much can be as bad as doing too little. Even though I despise it, I still use the silly skill.
My workaround was to build my own Claude skills to augment my capabilities and use them to properly do my job before running the mandated skill. Here is how the three approaches compare in terms of time to first output and my extremely scientific rating:
The time trade-off
| Process | Average time to first output | Extremely scientific rating |
|---|---|---|
| Using only the mandated skill | 5 minutes | ★☆☆☆☆ (1/5) |
| Doing everything manually | Around 1 full day | ★★★☆☆ (3/5) |
| Using my own skills before the mandated skill | Around 2 hours | ★★★★☆ (4/5) |
The rating reflects my overall confidence in the result, my involvement in creating it, and the time it takes. I know the table seems silly, but it perfectly reflects how I work and how I feel about my work.
- Using only the mandated Claude skill: It tries to do everything in one shot. When it gets something wrong, which it always does, everything that follows is also wrong. It takes a lot of time to read and validate the output, repeatedly going back through the results. It’s just bad. It’s trash.
- Everything manual: It’s true that I don’t enjoy documentation; however, when I’m in charge of the whole process, the result ends up being quite good. Typing everything manually gives me the strongest connection to the result, but the time it requires, thanks in part to my exaggerated aversion to documentation, makes it difficult to justify. That trade-off is why it gets 3/5.
- Using my own Claude skills: They guide me through the same steps I would follow manually. They allow me to understand the material before producing a polished output. The 4/5 rating applies to the complete workflow: I use my own skills first and then run the mandated skill. This saves most of the time required by the manual process while keeping me involved. The process still lacks the human touch that comes from typing everything manually.
Using my own Claude skills takes longer, but I remain directly involved in producing, reviewing, and understanding the final output. The AI assists me instead of replacing me.
I don’t really care whether the whole content chain is weak. My link is strong, and that’s all I need.