Doing things that scale - Using AI for actual work
The idea is simple.
The main framing I’m looking for:
You need good ingredients to effectively use AI
Set frameworks for refining what you put into an AI
Offer a solution that leads into my next essay: (Constructing well drafted inputs using airtable, Google Tables or Notion)
One of the most common types of advice I get when playing league of legends is to just be better. A lot of beginners believe that skill is paramount to how well you can enjoy a game. You need to be better than the guy across the table, jump into a match, and if you’ve truly mastered your skill, people will bow down before your incredible aptitude at the game. Or they destroy you, in which case you must be an absolute little shitter.
Actually, success seems to occur by how well you pilot things to the end. There may be a handful of teams that just have better hands, but it usually takes some sort of push to get a team to that victory screen. A good metaphor would be a chef that is given an assortment of food to cook compared to a beginner. Once the chef has a plan for the dinner, a team can take it and run with it, but there is an entirely separate and complex process to get that recipe to the end of service.
Ingredients
The most common thing that runs into AI problems is mindfulness. Nearly all AI tools seem to run into this base-level issue. You can’t just have it do the work for you. You have to go out and get them to run in the right way to align with your own business objectives.
Copy.ai or Jasper are some of the more successful tools in the AI sphere currently, and the problem they solve is a very common one - we’ll write out your marketing material and your business problems will be solved.
However, it doesn’t seem to have evolved in that way. Marketing is still led by conscientious objectors who demand high-touch and high-effort copy. Sales teams that can activate customers based on specific product features remain paramount.
Once you realize that existing conventions about an AI golden bullet don't exist – that shit ingredients lead to shit outputs – you’re actually much more free to explore how to properly run a dinner service.
Experience
I was trying to think of a phrase to convey how extreme your attention to the process should be, and I realized that michelin restaurants have already done it. They focus on quality of execution to a degree that would be considered pathological.
You actually see this trend by how we’re describing current AI tools. We’re not just taking our hands off the wheel and letting this nebulous AI do whatever it wants. Tools pop up that are named “co-pilot” or “assistant.”
— It’s dark magic, but even magic still needs a wizard to cast the spell—
So what needs to happen is an executive chef who can guide the work to the appropriate end. They need to have the experience of knowing what the ideal end result is. A chef-de-cuisine who can taste the output and take the extreme precautions to ensure that whatever is plated has considered the experience that a customer might receive.
Shit, we can even take the analogy once step further and just say that we don’t serve shit.
Service
In the best case, both good experience and ingredients help contribute to your work: the AI work you have to do to get started are not merely a necessary evil, but change the company permanently for the better.
If you have to be cognizant about ingredients when you're small, you'll probably still be cognizant about your inputs when you're big. If you have to pilot your own outputs, or experience things on your user's behalf, you'll learn things you couldn't have learned otherwise. And most importantly, if you have to work hard to delight people from the very beginning, you'll keep doing it when you have too much to keep track of.
I have some ongoing theories of how to best do this:
Run any inputs through a google table so people have to systematically answer questions so they know what they want:
For example: If you handed this release note to a family member or friend with little to know knowledge of the products, would they be able to understand the update at a high level.
Craft well formatted prompts for our expected end result:
For example: A release note, describe it with as much detail as possible, write like Paul Graham, ensure that there is no redundant information, describe the development of the api, make it sound smart:
The world has gotten more complicated: it seems that most dangerous traps now are choosing the easiest path. And the worst thing is, they're not even fun. People have the time of their life at really good restaurants. Let’s do our best to not compromise on that.
Thanks to (you?) for reading this. I’d love to get some feedback to make these essays better! Feel free to email me at layjchan[@]gmail.com and I’d love to buy you a coffee and chat about it.