How AI is reshaping the product manager role

Executive overview

The core PM value proposition — de-risking product delivery while maximising business value — hasn't changed. What has changed is the speed of the build cycle, which now outpaces how PMs traditionally work. PMs who adapt will spend far more time on sharp problem definition and customer insight; those who don't will become the bottleneck.

The framework Oji and Ezinne Udezue offer centres on three shifts: moving from static to living software, building in shipyard pods rather than siloed functions, and developing a hands-on technical fluency that goes well beyond prompt engineering.

The companies winning with AI aren't sprinkling it at the edges — they're rebuilding around it as the core.

What's changing in the PM role

• Faster build cycles mean PMs must confirm customer insights faster; traditional PRD writing is no longer sufficient
• Software is becoming living, breathing — PMs must design feedback loops and data flows, not just ship features
• Data literacy is now table stakes: where data goes, how it's organised, how it feeds model learning
• PMs must set guardrails to keep AI-powered products behaving as intended
• The PM-to-engineer ratio is under pressure; PMs who don't adapt risk being the bottleneck

What stays the same

• De-risking the product delivery process while maximising business value from investments
• Understanding what customers truly need — ethnographic observation still beats interview transcripts
• The importance of a sharp problem: old, frequent, painful enough that a 3–10x improvement makes people say "take my money"
• Simplicity as a design principle — avoid giving users too many options because you lack the courage to make a decision

The shipyard model

• A shipyard team evokes controlled chaos — apparent disorder underlies orchestrated progress
• Core capabilities (not headcount): PM, engineering, design, user research, data/ML, product marketing
• Teams should communicate and collaborate near-continuously, not in weekly standups
• Tendrils to sales, support, and customers act as the "skin" — the shipyard team is the brain
• Example: including a support manager in design reviews catches usability failures that engineers and PMs miss

Five skills most important for PMs going forward

• Curiosity + humility — willingness to take a junior AI course even as a CPO; teachability is survivability
• Agency — acting like a thermostat, not a thermometer; change the room temperature rather than just measuring it
• Data literacy — understanding how data is organised and leveraged in an AI-first stack
• Evals — writing structured tests to verify model outputs, constrain hallucination, and compare models
• Hands-on technical fluency — calling APIs with Postman, building prototypes, running local models, understanding quantisation and fine-tuning

What separates companies succeeding with AI

• They treat AI as a core capability, not an edge add-on — the code base may actually shrink as the LLM becomes load-bearing
• They specialise first, then build a connective tissue layer across specialised models rather than trying to build one model to do everything
• They take turns on user experience — the chat interface is not the final form; dynamic, personalised UIs are emerging
• They are navigating the "innovator's dilemma" deliberately: using LLMs in the old product while racing to rebuild the new one before competitors do
• They lead on ethics — treating AI capabilities as ordnance-level responsibility, not just a feature flag

AI at the core vs. AI at the edge

• AI at the edge: legacy code base remains; LLMs inserted at GUI intersections or to accelerate discrete tasks
• AI at the core: problem space and workflows are redesigned around LLM capabilities; the model is a first-class part of the solution
• Companies with AI at the core are the ones taking on adjacent problems and expanding scope — e.g. a form tool that became a pre-sales agent

Getting hands-on with the tech

• Write code; code is now "architecture and English"
• Convert PRDs into prototypes; call API interfaces yourself with tools like Postman
• Subscribe to multiple AI tools as a cost of learning to understand what each model is good at
• Pick a passion project that forces you to touch the things you need to learn — motivation sustains the depth of learning
• Experiment with local models (Ollama, LM Studio), MCPs, fine-tuning, and quantisation

Biggest lessons from 50+ years in product

• Sharp problems first — the problem you focus on is more predictive of success than execution quality; build on old, persistent needs re-imagined with new technology
• Simplicity is hard and underrated — especially for distracted users in 2025; shipping a simple, opinionated experience and being wrong beats shipping every option and learning nothing
• Communicate the why relentlessly — strategy fails in execution when people don't understand the reason; treat strategy adoption like product adoption, with early adopters and laggards
• Talk to customers, but observe more than you ask — what people do reveals more than what they say; ethnographic observation remains the gold standard for true customer insight
• Intention drives careers — holding a clear mental image of where you want to go and chasing it with discipline is more powerful than luck or opportunity alone

Ethics and responsibility

• PMs direct large developer teams and shape products used by millions — the responsibility is proportional
• Social media was built without adequate ethical consideration; AI carries higher stakes
• The best companies are beginning to lead on this; most are not
• PMs should ask the hard questions before shipping, not after