The AI-Native Product Manager

How AI has disproportionately impacted PMs, and the implications that has on the product function

The role within modern tech companies that I think has most disproportionately benefitted from recent advances in AI is the product manager. It’s not a stretch to say that a good PM today who knows how to use AI tools well is probably 10-20x more leveraged than they were just a few years ago. This takes shape in a few ways:

AI-assisted coding

Tools like Cursor/Cognition/etc have created a true step function change in what a technically proficient PM can achieve on their own. It is now within reason for a PM to create v0 prototypes of many features or new ideas without any need for external engineering resources.

While it is of course true that many good PMs could write basic software on their own before AI copilot tooling, in my experience the juice was often not worth the squeeze given that even the most technical PMs are typically not proficient software engineers.

However, AI software tools have so dramatically lowered the cost/effort/time to ship functional prototypes and have so raised the ceiling on what a semi-technical person can achieve, that I think this paradigm has fundamentally flipped. Good PMs will start to prototype all new features/ideas themselves before handing off to engineers, and I think this will allow PMs to be almost completely independent for a lot of zero-to-one discovery and validation work.

I suspect that as this plays out, you will see certain coding tasks more fully delegated to product managers. A good example of this is product instrumentation. PMs are typically the primary stakeholder who cares about this, the nature of instrumentation code changes are typically very simple/straightforward, and in my experience engineers typically don’t like working on PRs for it. It wouldn’t surprise me if this sort of task became more fully the responsibility of the product organization.

AI-assisted design

Most good product managers dabble as “part-time” UX designers - utilizing tools like Figma to riff on user workflows or to mock up simple changes. Advances in AI-assisted design will similarly up-level what is possible for a product manager to achieve on their own in this domain. Examples of this include:

1. v0, which makes it easy to create sample frontends

2. Tooling like Figma AI, which will make it easy to create high fidelity visual layouts of common components/workflows/etc (especially as design systems become more common and these AI products can just compose your design system building blocks)

3. LLMs are going to more broadly minimize the gap between the production assets and design files - making it easy to “bootstrap” a design file which mirrors the current production app precisely. This will make it much more accessible for PMs to do many smaller design changes/iterations themselves, as they will not need to either hand-compose the UX from scratch or ask the designer where the canonical design file is.

Similar to my point around how this might allow product instrumentation to shift from engineers > PMs, I ultimately think these sorts of AI-assisted workflows will enable UX designers to spend less time on low-creativity, low-inspiration UX design tasks that need to be done (since the PM can now own those themselves), and focus more on the complex UX design work they specialize in.

AI-assisted customer feedback analysis

The most critical job of a PM is to interface with the market and structure requirements & feedback into a plan for the engineering & design organization. The degree to which LLMs have structurally enabled this job to be done differently is unbelievable.

It is now within reason for a PM to basically analyze every single piece of customer feedback & input across all channels. Emails, bug reports, sales calls, online user sentiment on platforms like X, and similar can all be synthesized with LLMs systems and interrogated by the PM - see platforms like Enterpret and ProductBoard AI as intriguing early examples of this.

Traditionally, PMs had to disintermediate themselves from a lot of this feedback, and rely on, for example, the customer support team’s aggregation of Zendesk feedback to influence product priorities & roadmap. The degree to which a small number of PMs can now get source-of-truth data about all customer input has fundamentally changed.

This will also allow prioritization and roadmap planning to be done differently. LLM tooling should be able to easily reconcile discrepancies between the full universe of customer feedback relative to roadmap prioritization - and probably provide very high quality “first pass” roadmaps. I can imagine something like Granola but for product planning, where the top level strategic goals are input by the product team, and all the other “details” get filled out from there based on user feedback.

AI-assisted communication, alignment, & process management

Finally, LLMs are starting to play an influential role in how PMs communicate & align their teams. The most obvious breakout version of this right now is ChatPRD, which is a copilot for PMs to write high quality product requirement documents. PRDs are the core artifact by which most PMs communicate, and LLMs can clearly play an elegant role in helping to draft them more quickly & improve their quality and rigor.

Things can likely be taken much further than that, however. Some larger companies like Google have built internal infrastructure which use AI to auto-convert PRDs into Jira/Linear tasks - decomposing the product scope into its sub features. You can also imagine AI systems which analyze PRs and preview environments relative to the PRD, acting as a “first pass” product review without the PM having to manually review every single major feature that enters staging.

More broadly, so much of triaging, routing, and assigning bugs and tickets falls to product managers in many organizations, but could almost certainly be done more automatically in a world of AI. Why can’t an LLM triage incoming bug requests relative to the roadmap and determine which should go to the backlog vs. which should be prioritized? The more that AI is integrated into the project management cadence, the more PMs can focus on actually valuable strategic work.

What are the implications of all of this?

Assuming all of this is true, the next question is how do product organizations commensurately evolve?

Organizationally, I suspect the ratio of product managers to engineers & designers goes down. It will be more common to have a very small set of extremely senior PMs, who are ridiculously levered thanks to all of this AI tooling. The key driver of this will be that so much of the “grunt work” that can suck up a huge amount of a PMs time can now be heavily automated or accelerated with AI - whether analyzing user feedback or triaging issues. Correspondingly, more day-to-day product work will be owned by engineering & design teams, who are empowered with AI tooling to act more like product owners.

Procedurally, I think the product development lifecycle will change. PMs should be able to be much more self sufficient in doing early market discovery + solution validation thanks to being able to do much more end-to-end prototyping on their own, only handing off to engineering & design when it is very clear that something should be built.

The relationship between product, engineering, and design will probably also change. PMs should hopefully be able to take on a lot of the “low hanging fruit” tasks often assigned to engineers and designers (e.g. product instrumentation, simple UX changes for minor features). Engineering & design should will likely not require as much input/feedback from PMs if it becomes much easier to write extremely high quality, rigorous PRDs (thanks to tooling like ChatPRD) and if agents can act as a “first pass” PM review or feedback channel in many situations.

I suspect this may also lead to the opportunity for a next generation product management platform - ala Pendo/Productboard 2.0. So much is changing in terms of how PMs work and what will be expected of them that I think you can likely redesign the system of engagement/system of record for product work. Imagine a tool that enabled the following integrated workflow:

1. Synthesize/search/query all the current user feedback across all channels via Enterpret-esque LLM analysis, helping you create a ranked priority list of key things to address

2. Rapidly create extremely detailed PRDs for the top N priorities via ChatPRD style copilot system, contextualized by the prioritized user feedback

3. Bootstrap UX mocks and mini-app prototypes for each PRD, which the PM can use to quickly do first-pass customer validation. LLMs could even assist with determining which users to run the idea by (and potentially help run UXR style survey-style workflows to collect feedback on the solution)

4. AI-assisted auto-refinement of PRD based on market feedback

5. Once the PM feels the feature is worth implementing, the tool can do a first pass on auto-decomposing it into Jira tickets and subtasks

6. Finally, an LLM agent can act as “product guardrails” as the new feature gets implemented, reviewing incremental artifacts (e.g. PRs for each subtask, preview environments for subtasks, etc), providing immediate feedback to eng/design and helping determining what to bubble up to the PM

Something like this would have been totally inconceivable to build until recently, but now I think it is quite possible. Furthermore, while the above touches mostly on the core product iteration loop above, I think LLMs can also address additional issues in product management - such as quantifying product velocity and aggregating product progress across a larger organization.

A tool like this will need to consider engineering, design, & product all as core stakeholders. AI will make everyone in the organization more of a participant in product work - shepherded by a few very senior, AI-enabled product leads - and correspondingly product management tooling which primarily serves PMs becomes less relevant. Almost all engineers will become “product engineers”. This will require a re-envisioning of what product tooling looks like and how it is bundled.

I’d be very curious for input/feedback from product managers on these ideas and observations - especially product managers who have tried to really embrace AI-tooling in their job function. Shoot me a note at davis @ innovationendeavors.com