First, understanding Amazon.

When you start any new job, it’s important to wrap your head around the company, the culture, the people, and the operations.

We’re used to onboarding trainings where someone teaches you how to print something or where to find your healthcare benefits, but the rest of your first 3-4 weeks should really be dedicated to laying your foundation and learning the culture of the company you just joined.

(The book “First 90 Days” has a great walk-through of this, and I’m grateful another mentor Bill recommended it to me. Buy it if you’re ever starting a new job.)

Some questions you should ask in your first few weeks at a new job: what is everyone’s role? What are they goaled on? Where do you fit into the puzzle? What do you think you are goaled on? Does that make sense at first pass? Who are you expecting to collaborate with? Who is glad your role exists? Who is annoyed that your role exists? Where are the dependencies, and can you forecast your blockers or resistance? What are people expecting from you, and what are you expecting to build? Do those ideas overlap or contradict—how?

In my first month (and really my first day), it was obvious to me that my remit was extremely broad. Too broad. But that was part of the job, and actually the best part of the job—they hired someone to not only deliver results but also add structure and vision to an otherwise vague landscape.

Navigating ambiguity is a skillset, and you can absolutely stretch that muscle. There were 400,000 global startups and 100+ different types of AI, and I had to make sense of it all. So, I started where any overwhelmed learner does: I opened a blank note page. In this case, I created folders in OneNote (one for internal conversations, one for external conversations, one for research, and one for takeaways and hypothesis testing that resembled a mad scientist’s notebook).

I spent my first few weeks talking to as many internal folks as possible. Among many, my two most eye-opening conversations were with Matt Wood (VP of Product, AWS) and Eric Heikkila (Director of WW GTM for Databases).

My learnings from part one: lay your foundation, know and earn trust with your stakeholders and collaborators, stay open-minded, be ready to be surprised, and don’t let others sway you.

Second, understanding customers.

After 50+ internal conversations with product managers, VPs, technical leads, solution architects, marketers, sales, and PR, I started phase two: going deep with startups.

I had posted about my new job on LinkedIn and went quite viral (100K+ likes), and between that and friends in the space, I had a healthy pipeline of top ML folks to speak with all over the world. It’s important here to do an 80/20 evaluation—80% of your conversations and interviews should be with the “this is very obviously my customer” group, and 20% of your conversations or interviews should be with the “I bet they impact my customer and would have interesting things to say and shape my wider view” group. With that, 80% of my conversations were with AI founders, with a 20% extra sprinkling of individual contributing ML engineers, AI ethicists, data scientists, research scientists, and professors.

To prevent data fatigue, I devised my own process to get to the insights faster (navigating ambiguity quickly turns into frameworks and consistent discovery processes). Every conversation was its own page, and every conversation included the same questions to more easily compare. (Author’s note: LLMs has made this much much easier, and if I did it today, I would use LLMs in the process).

Those interview questions included: tell me about yourself? What are you building? Why are you building that/what inspires you/what problem are you solving? Why is now the right time for your business? What are you blocked on? What is a wonderful funky surprise in your company that makes people smile—could be a funny mascot or an offsite tradition or a way you’re using AI? How can I help?

I started mapping out the landscape and it became obvious that the market segmentation approach adopted by other teams was wrong. (And every PM knows that if your segmentation is wrong, your solution and strategy is probably wrong too.)

Onto fixing it!

Third, I made sense of the mess.

The biggest finding was that late-stage startups who integrated AI later in their business (like DoorDash) behaved incredibly differently compared to startups that “went deep from day one” into AI (like Scale AI).

It wasn’t just that one was building an AI product and one wasn’t. Everything was different. The founder profile, the attitude, the motivations, the research, the ops, the team skills and structures, the salaries, the raises, the business model, the speed they moved at, and even the cap tables. EVERYTHING.

GPT-2 had just come out with 1.5 billion parameters, ML ops companies like Tecton AI and Fiddler AI had been grooving for just a few months, YC’s AI track had only had a few cohorts…it was still so early.

Most companies were treating these startups as your standard VC-backed fast-moving SaaS companies with a little AI frosting on top, but there was a revolution brewing and 99.99% of people didn’t know it was coming.

Fourth, I packaged the mess.

Your insights are worthless if you can’t tell a good story around it.

It’s correct action on the insight that counts. And at some point, even if you’re a solo founder, you’re going to need to convince someone of your insight.

Coined terminology is a big part of that storytelling—you have to give people the right words and phrases to use when you’re not in the room to defend or explain it. The bigger your org, the more important it is for your strategy doc to speak for itself.

So how did I come up with the business name?

I knew three things:

1. I needed two terms, one for “the startups who are using AI from day one” and “the startups who are adopting AI later in their journey to gain some efficiency”.

2. The market segmentation names would probably be leaked or used publicly at some point, so I couldn’t use terms like “AI-first” (because who the heck wants to be called AI-second?).

3. AWS publicly rebranded in 2019 from ‘AWS AI’ to ‘Amazon Machine Learning’, so I knew I needed to use ‘ML’ in the names.

I stayed up with an ML engineer friend late at night thinking through names. His favorite was ‘ML-native’ (like “digitally native” which had been used for companies like Airbnb and Pinterest). My favorite was ‘ML-core’ because that’s what it was: ML was at the core of every single part of their business. It told the story I needed, it was short, and it was easy to say and understand.

So I went with it.

For the other startups (where they used ML later in life, usually around series B to start and series C in a “big way”), I went with the term ‘ML-supported’. Their business would still exist without AI, but it was supporting everything they did and making them more efficient.

Five, I made a plan.

Final thoughts

I’ll wrap it up with this: I've always believed that before we leap forward, we need to glance back, challenging the assumptions and frameworks laid down by previous leaders but also learning from their incredible work. It sounds strange, but it helps to think of the business as a living organism with a connected and shared learning curve.

Now, it's your turn to carve out your path in the AI age. Grab the knowledge from seasoned subject matter experts in your company, but don't lose sight of what's coming next. It's about finding the perfect blend of the tried-and-true bits and the groundbreaking bits. As you pivot towards starting to use AI and eventually embedding ML at the core of your work like these startups are, remember it's not about keeping up; it's about leading the charge.

Set your foundation, stay customer-centric, and be ready to challenge the status quo; your AI leadership journey starts now.