Operational mindset shifts on a GenAI team
Unlearning what I learned about PM in my second rotation
Hey, hi, ho! We’re three months into my second rotation, but I still don’t know what I’m doing! And that’s okay! Because I’m learning, and that’s what really matters.
In this rotation, I’m charting a new land of PM. One lush with ambiguity, sprawling with dependencies, and lurking with pivots. This is the land of generative AI, the 21st-century California gold rush into uncharted product territory.
Throughout a series of posts, I’ll be mapping out that land acre by acre as I learn how to wield the power of the latest technology, starting with new ways of operating.
Mindset shifts
#1: Learn first, launch later
Throughout the course of a product’s lifecycle, product managers will typically spend time in one of these phases:
In my first rotation, I fluttered mostly between building and releasing a capability feature that could apply to several products across the ecosystem. Build the feature for X offering, ship. Build the offering for Y product, ship. Occasionally, I dabbled in the Design phase to create a design that could scale with the number of use cases we were launching the capability for.
It was a Cinderella slipper, perfectly sized, fitting what I was looking for at the time. (In my summer internship before this role, I never saw my feature see the light of day. So you can imagine how eager I was to see the culmination of my work live in the product being used by real(!) customers.) The team already validated the customer problem and proved the technology worked. The research & planning and design stage were complete, and all I had to do was take the baton and run with it.
Then generative AI entered the picture.
I switched to the generative AI team when we were just falling from the Peak of Inflated Expectations into the Trough of Disillusionment of the Gartner Hype Cycle.
The new technology is like the Tesseract from Marvel. It glows and radiates and teases the power to do incredible things, and everyone wants their hands on it.
But what does it do? What exactly can it do (well)? And when should we wield its power without imploding the universe and pulverizing everyone into dust?!
After learning it can’t do that much very well in our product space, we drew a Go back 3 spaces card and entered learning mode.
Our team research spanned across the customer, the technology, the processes, the metrics— pretty much everything. Things were changing so rapidly that there was just too much to learn overnight for one individual to handle. To handle this, one of the team’s rituals was sharing what we hoped to learn at the beginning of every week and what we learned at every standup. Whatever we learned individually was shared with the rest of the team to expedite every stage of the product development phases we were in.
Generative AI ushered in a new way of operating. In this world, launching takes a backseat to learning until we feel confident about what types of problems the technology is world-class at solving.
#2: Building new norms
Onboarding onto this team, I often found myself in different situations asking “What does the team typically do?” only for my manager to reply “There’s nothing typical yet.”
You’re telling me that this team hasn’t yet figured out all the secrets to working with generative AI?!
I assumed they had all the answers neatly laid out in front of us, everything figured out. Alas, that was far from reality. It took me a bit to internalize that we were working with technology so fundamentally new and different that it confounds almost everyone in tech.
We are an early-stage product working on a mature product with millions of users. Naturally, I expected existing practices and a few “this is how we’ve been doing it on the team” to learn from. I was awaiting tried-and-true steps to take and patterns to follow because they’ve been known to work in the past. But there weren’t, not with generative AI. We were trained and experienced in making product iterations and incremental improvements, not launching new products altogether and focusing on the minimum viable product let alone an AI-MVP. The new norm on the team became having no norms at all.
No norms to work with can be a pretty exciting place to be. A blank canvas means having the unfettered freedom to build new norms and heavily shape the team culture. As PMs, we get to be the ones to build unencumbered by any pre-existing traditions or engrained practices.
I’ve never worked at a startup, but this intuitively feels like one with its embryonic team rituals. I like it.
#3: Timely escalations as a habit
Generative AI poses an existential threat to many large tech incumbents. After all, it's quite powerful and anyone can harness that power if they take two seconds to search ‘How to create an LLM.’ Being on an AI team, we’re reminded of this threat on our heels quite frequently and are constantly prodded to move fast to thwart it.
Move fast. A phrase notoriously inculcated by startups where it’s easier to achieve due to a lack of bureaucracy and tech debt. At a big company that has more than its fair share of both of those, it looks like quick escalations.
Take this scenario:
Me: I’ve validated a feature with a customer and know I want to build it.
*Goes to Person A, B, and C from Team Apple, Blueberry, and Cantaloupe who I know work on components that my feature will need*
Me: Ok, I also know I have to work with Team Dewberry since they own the security component of my feature but I don’t know who Person D from that team should be.
Now I’m two days into building out this feature and Person G and H from Team Guava and Honeydew emerge from the woodwork and unveil that their teams own the identity and motion components of my feature and that they’re already working on it.
Tada! Suddenly, we have two similar but slightly different versions to reconcile since there are platform teams we didn’t know about moving in different directions.
Me: Ok. Which components do we keep and which do we replace?
*Churn, churn, and churn… oh, do I see butter forming?*
Platform dependencies are important to building one cohesive and seamless product experience and reducing redundancy, but they do so at the cost of velocity. Adding teams can stretch out the time it takes to align on decisions and prolong the time it takes to build when the platform team has competing priorities. When teams aren’t prioritized, they may go off and build their duplicate version of the solution. Later on, these versions must be reconciled and the decision churn renews.
When you encounter a knot of platform dependencies that your team alone can’t untangle in a reasonably swift manner, lean on your leaders to step in and help reconcile. As a team expected to build up defenses against generative AI-breathing dragons, we’ve been encouraged to employ this skill whenever we experience undue churn. It turns out that what takes you days or maybe even weeks to do can be done in an hour-long meeting by leadership. Who knew?
Escalate as necessary to get the teams aligned and marching back together in the same direction.
Mindset constant: customer needs first
Our job as PMs is to advocate for our customers, understand their needs, and solve their problems. It doesn’t matter to the customer what technology we use to do this. What matters most is what technology can do for them. How does technology solve their problems better than existing solutions or workarounds?
It’s all too easy to get swept up by the hype of new technologies. As exciting as generative AI is, we need to remember that technology remains just a means to an end. Our focus should always remain on the individuals we’re building solutions for: our intrepid customers.