Engineering teams are moving faster than ever. Why isn't productivity keeping pace?
AI has made software engineers dramatically more productive. But it has done almost nothing to make product decisions easier. If anything, it has made the question of what to build more pressing.
Previously on ER¹, I argued that AI transformation is fundamentally a systems challenge, and that true productivity gains will remain elusive until we fix the underlying system as a whole.
After a sustained period of accelerated delivery, engineering teams are increasingly running into the same challenge. Just because you can doesn't mean you should - deciding what to build remains the hardest part of software delivery, and no amount of acceleration elsewhere in the development lifecycle will increase throughput if the product challenge is not addressed.
So, what can teams do about this? Can the business ever keep pace? When should they lean hard into AI, and what work must remain uniquely human? Or, are there natural constraints within product management that mean there will always be a limit on how quickly it can feed the machine?
Where product direction comes from
The purpose of product management is to build products that solve real customer problems. Product direction isn't shaped by executive whim or personal intuition, but by evidence, feedback and judgement. In simple terms, it’s a 3-step process:
- Gather information - customer conversations, metrics, observation.
- Synthesise evidence - analysis, validate assumptions, recognise patterns
- Decide direction - prioritisation, strategy, trade-offs, roadmap
Product managers balance their time across all three, but the best expend significant effort gathering evidence, speaking to customers and understanding their needs. These conversations provide context that rarely appears in any metrics dashboard or report - frustrations, workarounds, motivations and unmet needs that only come to light when you ask the right question.
Customer conversations are central to the feedback and learning mantra of continuous delivery. Usage data might tell you that your click-through rate on a newly positioned button has dropped 5% in the past quarter, but a customer conversation will reveal that the underlying workflow is completely broken. Metrics tell you what is happening. Customer conversations tell you why.
Understanding, the human constraint
The human skill is knowing what questions to ask. It is mastered through years of hard-won experience, building trust with the customer and developing intuition for the market. It lies in steering conversations through genuine rapport and empathy. Of course, AI can help, particularly with preparation, transcription and analysis, but the conversations are best done person to person.
AI might also be superb at spotting patterns and churning through metrics data, but turning those insights into product direction takes considerable thought. Shaping is inherently a team sport, involving multiple stakeholders, engineers, designers and users. It requires balancing customer needs against feasibility, budget, time-to-market and a host of competing priorities.
The role of the PM is to navigate this path. Metrics might inform many decisions, but the role itself is intensely people-centred - verging on political, even - involving influence, consensus-building and decision-making. These uniquely human capabilities don’t just happen overnight. They take years to develop.
No amount of AI will short-circuit that effort. The limiting factor isn't the speed at which information can be processed; it’s the speed at which consensus and understanding can be developed. If anything, we are the constraint - our openness (or lack thereof), our willingness to engage, our quirks and biases, our endless inconsistencies, the ambiguities of how we communicate. Deciphering people is a job in itself.
Administrative work is machine work
None of this is to argue against automation. Quite the opposite. Product managers should be striving to automate as much as possible. The question is no longer where should AI be used, but where should we not use it.
AI is freeing product managers from all the administrative drudgery that ate up so much of their time, allowing them to focus on higher-value work - inferring insights from data, making sound product decisions, and steering the team in the right direction. Spending less time on the mundane and more on what’s important feels like a pretty decent trade-off.
That said, writing remains a vital skill, but primarily as a tool to sharpen thinking. Much of the routine toil of producing product collateral can and should now be automated. Drafting user stories, documenting acceptance criteria, transcribing interviews, synthesising findings, preparing stakeholder updates are all increasingly machine work.
Codifying for the common good
The more interesting opportunity lies in codifying these product practices into reusable assets. The modern product manager needs to be as conversant and comfortable with AI as any engineer. The institutional knowledge that once lived in their head should now be shared across teams, iterated on and improved over time.
The role of the product manager is therefore changing. Fluency with AI is becoming as important as fluency with traditional product practices. The function is no longer just about setting direction; it's about building and evolving an AI ecosystem that enables faster feedback, reduces administrative overhead, and produces more consistent, higher-quality artefacts for downstream consumption.
Shared ownership, better outcomes
As AI has democratised more parts of the delivery lifecycle, traditional role boundaries have started to blur. Engineers and designers are increasingly being asked to think beyond implementation, to challenge assumptions, contribute to product decisions and take responsibility for outcomes rather than outputs.
This is a good thing. The shift is encouraging co-ownership, team autonomy and agency. Accountability is a collective responsibility after all, and teams that operate as one are better equipped to understand problems, explore options and learn together.
By removing internal siloes and making decisions informed by evidence, understanding and feedback, teams end up delivering higher-quality products. The true measure of productivity is not output, but impact - happier users, better outcomes, less rework.
An existential moment?
But it has led to an existential question for the product function. If AI can automate large parts of the role, and product thinking is now shared across the team, what is left for product managers?
As it turns out, more than ever.
Product managers are ultimately responsible for product direction. Their role was never simply about translating business needs into engineering requirements. It was about developing customer understanding, navigating competing priorities and helping teams make better decisions in the face of uncertainty. None of this has gone away.
They do the hard yards of building relationships and understanding the motivations behind customer behaviours. They help balance competing demands from users, stakeholders and the business. They act as facilitators, advisors and navigators, with a foot in every camp - product, engineering, design, even sales and marketing.
The difference is that AI enables product managers to spend more time on the parts of the role that were underserved in the past (because they spent so much of it on ‘menial’ work). It is enabling them to be better at their role.
PMs who built their identity around being the keeper of requirements or the translator between business and engineering will find this transition uncomfortable. Those who were always drawn to the harder questions – why does this matter, what problem are we actually solving, what should we build next – will thrive.
Final thoughts
When one part of a delivery system speeds up but productivity as a whole doesn’t, you need to examine why. Engineering might be flying, but it can only continue to do so if it works tightly with business and product, simpatico, as a single team.
Product management has never been about the grind of transcribing or documenting. It is about vision, direction and learning. It’s about making decisions that address real business value. That problem has not gone away.
The objective is not to speed things up just for the sake of it. It is to reduce the time between learning something and acting upon it. It’s to improve the quality of decisions, not to maximise the number of decisions being made.
Product management has never been more vital. What’s changed is that product managers can now focus on the higher-order activities that matter most.
¹ Bonus points for recognising the 90s TV trope :)
