AI Accelerates Shipping, But What's Worth Building?

Claire Vo noted that product management has undergone a complete transformation since the last Lenny's Summit, a change she previously anticipated by jokingly declaring product management 'dead' to ensure she remained honest about its evolution.

She observes that the role has become significantly different, signaling the traditional roadmap as her next target for critical examination, acknowledging its past importance while foreseeing its decline.

Historically, roadmaps defined direction, sequence, and priority for teams, serving as a defining artifact for careers within the product industry.

Vo suggests that many professionals are likely on the verge of writing their last traditional roadmap due to fundamental shifts in how product teams conceive and build their offerings.

Vo confirms that she is now shipping more software than ever before, attributing this acceleration to the proliferation of AI agents, advanced developer tools, and sophisticated models.

Despite having these powerful building blocks, she finds herself 'out of good ideas,' indicating a shift in the primary bottleneck from 'what can I build' to 'what is actually worth building.'

Historically, engineering capacity served as the main constraint, forcing product managers to prioritize rigorously and decline many requests.

This abundance of execution capacity has not translated proportionally into revenue growth, highlighting a disconnect between increased output and actual market impact.

Vo described a hypothetical 'Product Graph' project, designed to create chat-based Product Requirements Documents (PRDs) by ingesting all company and customer data for processing by AI agents.

She presented this as an example of a technically feasible idea that nonetheless raises significant questions about its true impact and whether it genuinely solves a valuable problem.

Vo expressed a personal sentiment of running out of genuinely good ideas for new products to build, even with advanced capabilities at hand.

Claire Vo recounted building a complex semantic product graph and an auto-wiki tool using AI agents, achieving a functional 'oneshot' success that matched competitors' feature sets.

Despite its technical viability and market alignment, Vo felt the project lacked true differentiation, describing it as 'code to nowhere' that should belong in the trash.

She questioned the return on investment and the interface choice, pondering if an agent-first approach would have yielded a more meaningful outcome.

AI automation has dramatically accelerated development, allowing for rapid resolution of customer issues and efficient code refactoring, a process referred to as 'chat purity.'

Historically, the scarcity of engineering resources served as a crucial filter, eliminating less promising ideas below a certain 'cut line,' but today's nearly limitless execution capacity means companies can ship even their worst ideas.

Vo contends that cheap execution does not obviate the need for human judgment; without it, companies risk engaging in aimless productivity, merely shipping for shipping's sake.

Vo identifies three key traps in the AI-accelerated roadmap: the Backlog Trap, where clearing requests does not equate to meaningful progress against business or customer problems.

The Parity Trap arises from competitors using identical insights, leading to homogenized products that lack true differentiation.

Lastly, the Churn Trap allows teams to rapidly abandon features, preventing the compounding of product learning and sustained impact.

Vo warns that these combined factors accelerate a company's trajectory toward mediocrity, essentially accelerating its path to 'mid'.

Roadmap Zero signifies a new state in product development where every visible feature on a backlog becomes both plausible and technically buildable.

This shift means traditional prioritization based on engineering constraints is no longer applicable.

Companies must therefore develop novel methods to determine which products are worth building when capacity ceases to be the limiting factor.

RICE prioritization, a common method for evaluating feature impact, is losing its meaning because effort has become a less useful proxy for value in the age of AI-driven development.

The combination of an AI 'factory' producing features at machine speed and an outdated roadmap rapidly leads companies into traps of bad ideas and weak judgments.

Vo argues that the true bottleneck in product development is no longer the code itself, but rather obtaining reliable truth and insights directly from the market.

AI, despite its capabilities, cannot transform untested assumptions into facts, regardless of adversarial reviews or internal debates.

Teams must now prioritize real-world testing, data-driven insights, and a clear point of view to validate ideas.

The primary goal is to achieve faster contact with reality, which imposes a higher obligation to actively listen to and incorporate market feedback.

Engineering scarcity once made feature-based roadmaps viable, but that model is no longer effective in the current landscape.

Vo advocates for building durable convictions about the future rather than relying on rigid three-to-nine-month feature lists.

Teams need to define upfront what evidence will prove a conviction true or when it's time to stop pursuing an idea.

The goal is to maintain a high-speed development 'factory' that is used specifically to test convictions, not merely to execute tasks.

Customers are becoming accustomed to product churn, shifting their trust from individual features to the broader vision and the competence of the development team.

Vo explains that good stubbornness involves staying committed to solving a core problem while being willing to revise the solution based on new insights.

Conversely, bad stubbornness is characterized by moving the goalpost simply because the team has the capacity ('tokens') to continue shipping, rather than because the underlying problem is being addressed effectively.

Not every shipped item should be treated as a definitive promise; some must be embraced as iterative experiments, acknowledging their temporary nature.

Traditional roadmaps often compelled teams to make unrealistic promises about specific features and delivery dates, creating a burden of absolute commitment.

Internal and external communication now requires a clear distinction between exploratory 'probes,' structured 'experiments,' and genuinely 'durable promises' that customers can rely on.

Probes are small bets designed to explore new ideas, while promises indicate features that customers can confidently build their own operations upon, reflecting a higher level of conviction.

The true value of this roadmap shift lies in communicating the level of conviction behind a product initiative, rather than merely its estimated impact or build difficulty.

Product managers must recognize that while the ability to produce code is abundant, customer trust is a finite and precious resource.

Building and shipping features without full conviction risks eroding long-term customer relationships and loyalty.

Features should ideally be feature-flagged until they meet a clear threshold of conviction and proven value.

Roadmaps, therefore, need to evolve from simple lists of features into more dynamic statements of ambition and strategic intent.

The past 12 to 18 months have been characterized by a focus on 'velocity,' leveraging AI to rapidly ship prototypes and pull requests.

Vo suggests that the coming year should be defined by an 'ambition game,' encouraging large, high-conviction swings and daring experiments.

Teams should now aim to execute complex builds in a matter of weeks, tasks that previously would have required a full year.

Success metrics for this new approach should emphasize the number of high-impact experiments launched, accepting that many of these ambitious undertakings will inevitably fail.

Traditional spreadsheet-based roadmaps, which lock in dates and features, are no longer effective in the current dynamic environment.

Product development should instead aim to build a 'factory' that can generate unexpected innovations and surprise the team with its results.

It is acceptable to discard software if it does not meet a high bar of quality or prove its value.

AI will frequently generate better ideas than the human team; the objective is to leverage this potential through an adaptable system rather than adhering to rigid, pre-planned lists.