The Death of Product Development as We Know it

Goodbye three-legged stool, 2-pizza teams, and "managers"

Much has been written about what AI’s magical powers can enable—whether engineering, design or documentation. But I believe something just as interesting is happening with how teams build in the era of AI.

This post was co-created with gems of insight from Henry Modisett, head of design at Perplexity. I admire Perplexity for many reasons, one of which is how they are a cutting-edge applied AI company at the forefront of a new way of working.

---

i. The EPD stool and the two-pizza team is dead

When I was a kid, I loved Captain Planet—that ’90s cartoon where five teens with magic rings (Earth! Fire! Wind! Water! Heart!) teamed up to fight eco-villains.

At the time I thought the series was so incredibly original. (I mean, the planet personified in an aqua dude wearing tights?!)

Little did I know I was looking at a well-known archetype emulated later by the likes of Power Rangers, Teen Titans, Avengers, and tech product teams across the globe.

For decades, the traditional tech team looked something like this: Engineers (3-5 of them) + PM + Designer = good stuff built. Sometimes, we’d add more “rings” (Sales! Data! Marketing! Customer Success! User Research!) to fight bigger villains while trying to abide by the best practice that 2 pizzas should be able to feed this group.

In the era of AI, all this breaks. It’s time for a new group of heroes to take the stage.

---

ii: The death of mocks and docs

Traditional product development has been best practice-ized until it resembles an Ikea instruction manual:

1. The Idea (conceived in enthusiastic brainstorming chats deemed “groundbreaking” and “innovative”)

2. → Mocks (polished until it’s a sparkling good storyboard for stakeholder alignment)

3. → PRD (a beautifully formatted 12-page novella admired by many but read closely by precisely zero)

4. → Code (with deadlines extended as frequently as the MCU)

5. → Launch (with ample hopes and dreams and craft beer and pizza)

6. → Surprise! (most of the time, the surprise is why users reacted with a collective shrug rather than thunderous applause)

This worked when software moved in cycles of months. But now that we have AI at our disposal (who is today an insanely productive, cheap yet junior talent that must be aggressively micromanaged), that tidy linear path is going the way of the dinosaurs. Ask yourself:

• Why brainstorm for hours when AI can generate 50 ideas (and roast the bad ones) in minutes?

• Why show static mocks when AI can build a version of the damn thing for your stakeholders to play with?

• How much effort should you put into your PRD when AI changes the rules of what’s possible every other Tuesday?

• Why spend a bunch of time debating engineering flows when you can’t deterministically predict what the technology will do?

• How much more would you achieve if you could learn the Surprise! of your launch in a fraction of the time?

---

iii. The prototype-and-prune path is the future

Forget IKEA manuals. You don’t want to be stuck in a situation where you follow every step, screw by screw, only to realize halfway through that what you actually need is a ladder.

The future of product development looks like this:

1. The Idea Garden

   Don’t settle prematurely. Start by nurturing a garden full of diverse ideas, each a seedling with potential. Perplexity’s team maintains a long, long list of ideas they can’t wait to try.

2. → Prototype-and-prune
   The bulk of future product development will be spent researching whether an idea has legs. Say goodbye to cycles of mock and doc feedback, and say hello to code-first prototypes with constant iteration (on prompts, models, etc) to answer three key questions. If the answer to any of the below is no, the idea is pruned.

   1. Capability: can this idea be done with today’s technology?

   2. Accuracy: Will the results consistently meet the user’s expectations?

   3. Speed: Can it perform smoothly and quickly enough for real-world use?

3. → Polish-and-productionize
   Once an idea demonstrates clear promise, it can now get refined for real customer launch, whether to a small beta group or beyond. This is the time for optimizing, aligning with brand vibes, and coordinating release logistics and communication.

4. → Launch
   Still with plenty of hopes and dreams and craft beer and pizza

5. → Further pruning
   Selectively prune ideas that don’t thrive in real-world testing, swiftly retiring unsuccessful experiments to free up space for fresh seeds.

---

iv: AI favors the individual contributor

Imagine you're assembling a team for an escape room challenge. Would you rather have six managers debating strategies or two sharp thinkers who rapidly test every combination to unlock the door?

Welcome to the era of AI-native companies, where individual contributors (ICs) reign supreme.

Prototype-and-prune calls for teams with the following 5 characteristics:

1. Tiny teams working in parallel — teams with a handful of people (think 1-3) can simultaneously explore a wider range of ideas to see if they work.

2. Blurred functional roles: why do you need a stool with three (or more) legs when a classically-trained engineer can use AI tools to iterate on mocks, a classically-trained designer can use AI tools to iterate on marketing copy, or a classically trained product manager can use AI tools to get insights from their data?

3. Reduction of “management” roles: the cost of communication overhead shrinks drastically when you go from 7-10 people to 3. The PM-as-coordinator, or manager-as-coordinator role is no longer needed until (maybe) the polish-and-productionize step.

4. Good taste: taste is the ability to distinguish excellence from mediocrity. The hand can never consistently produce better than the eye can discern.

5. High agency: AI favors those who bias towards action and see themselves as problem solvers with the power to get shit done. Those who prefer structure, consensus, and waiting to be told what to do will find themselves struggling.

---

v: How to Survive (and Thrive)

For leaders:

• Hire curious generalists, not specialists.

• Track speed of feature development: ensure that it’s going up.

• Reward doers over sayers: look for the proof in the problem solved, not in the discussion had

For ICs:

• Identify as a problem-solver: not an engineer / designer / pm.

• Make AI your constant collaborator: experiment, experiment, experiment with doing things you would have previously relied on a teammate for. Experiment with doing whatever you know how to do well even faster.

• Just do it: try stuff out, don’t wait around for permission.

For Everyone:

• Embrace the chaos. The future isn’t built on plans—it’s grown in gardens.

---

---

Other essays in the Looking Glass AI series