Prompts are the New Code

The Signal

If you were forwarded this newsletter, consider joining over 1,600 weekly readers—some of the smartest GTM founders and operators—by subscribing here:

---

Hey y’all!

It’s Sunday afternoon and it’s raining and my kiddos are napping so I’m tinkering with AI.

Lately, I’ve been going deeper on prompting. I think it’s underrated and incredibly powerful.

In the interest of “learning in public” (and, selfishly, because I believe that good writing requires clear thinking), I figured I’d jot down some thoughts today around prompting.

Here’s what I cover in this post:

• How I’m using prompts today

• How to get better at prompting

• Prompts as critical IP in the future

---

Prompts are the New Code

I’ve been trying to learn prompting better.

“Prompt engineering” was a hot term a few years ago—peaking two years ago, interestingly, and downward trend since then:

Google Trends Link

But, I think “prompt engineering” is incredibly powerful in modern go-to-market, sales technology building and GTM Engineering. (Maybe across all software, but that’s too big brained for me to predict)

I’ve said this before, but I’ll say it again: AI makes creativity the bottleneck.

When there is no longer a “technical person” required to ship a GTM experiment, or code an application, you quickly realize that the limitation is not code, but creativity. All you need to know is what you want to do, and then write it in natural language. An LLM can do the rest now (or, very close to it).

I decided to improve my prompting skills as I came across areas of consulting that were redundant and I thought could be good candidates for an LLM (instead of me, manually doing the ‘thinking’).

How I’m using prompts today

Here are a few of the ways I’m using prompts to onboard customers:

• Company’s ICP (OpenAI)

• Target Account List (Perplexity + Clay)

• Key Buyer Personas (OpenAI)

• Ideate 5-7 Micro-campaigns (OpenAI)

• Ideate 3-5 Offers (OpenAI)

• Build messaging (Anthropic)

  • Signal + self-referencing email templates

I think lists like this for GTM teams will get longer, not shorter, in the coming years.

Instead of a swipe file of templates, I am now building out a library of prompts. Similar to how Tomasz Tunguz talked about his little library. Here’s an excerpt from that post:

I didn’t notice it at first but there in the back corner of my laptop, I’ve been assembling a little library.

The library doesn’t contain books, but scraps of text that explain what an AI should do…

…That little library will contain the secrets to how we each do our jobs. We will need software to create them, share them, update them, measure them, & run them.

The better our little libraries become, the more effective we will be at work.

I also started playing with Replit’s agent for coding, and have some of these built into a web application that I hacked together in less than two hours (mind you, I cannot write a line of code… we are living in wild times).

Or, I can have several columns in a Clay table, and drop a prompt that points to previous columns in the table and points to an LLM end-point (OpenAI, Anthropic, etc.). It’s a magical moment when you see these linked prompts start autonomously working together.

How to get better at prompting

I found an amazing YouTube video of “Prompt Engineers” at Anthropic talking about Prompt Engineering. So, what better source could there be?

It’s a really fascinating listen. I recommend it (if you’re trying to get better at prompting).

One of my biggest ‘hacks’ that I learned in this video, is to take a prompt that you’ve built, and before asking the LLM to execute the task (ask the question), instead, put this at the top of the prompt:

Below is a prompt that I am going to ask you. Do not answer the prompt. Instead, tell me how I can improve the prompt for you. What is unclear? What is missing? What is confusing? Again, don’t answer the prompt, but tell me how it could be better, and rewrite it for me. Here is the prompt:
[Insert prompt here]

The response to this question gives SO much insight into how an LLM works.

It’s slowly retraining my brain to work better with machines (for better, or for worse). I start to think about how to phrase a question when asking an LLM, or what information is needed to give it the context required, etc.

I see this as learning right now. That may take (literally) years. But, as the old saying goes: “the best time to plant a tree learn AI is 20 years ago, the second best time is today.”

Prompts as critical IP in the future

A sales tech founder asked me what prompt I used for this LinkedIn post I put up last week about finding a company’s ICP and Key Personas and micro-campaigns, using only their domain.

It felt like I was giving up a secret when I told him the answer. This was the prompt (see what I did there) to write this post.

I felt like I should protect a) the prompt and b) which LLM I used to ask the prompt. Like I was giving him access to the “source code” of one of the ways I create value in the world (and charge money for said value).

I think that the best prompts will be valuable IP that will be as critical as the code bases of the last few decades.

The 10x software engineer of the future may be the one who manipulates AI to produce the best output with the best prompt.

Will she keep it to herself?
-Tomasz Tonguz
AI Prompts as PRDs : Why Prompts Will Become Important IP Assets

As digital coworkers become a thing (they will be, imho), we’ll need ways to better interact with AI agents via prompts. And will have to give these AI agents access to data (like CRM, data warehouse, stripe, notion, slack, etc.). Basically, give them an Okta with permissions, just like you would a full-time employee or (different permissions) a contractor.

The future is going to get weird. I’m leaning in. Slowly, but surely.

---

Thank you for exploring—things like, how AI and GTM intersect—with me.

Your attention is greatly appreciated.

See you next time,
Brendan 🫡