Developers were the first function to capture AI's productivity gains in software delivery.

Coding tasks can now be completed significantly faster with AI, which is why most early software AI investment went into engineering: code generation, test creation, code review, and developer tooling. The impact is measurable. Teams are shipping more changes, reducing cycle times, and increasing throughput.

Jellyfish's April 2026 AI Engineering Trends report, analyzing 37 million pull requests, found that top AI-adopting engineering teams achieve roughly 2x the pull request throughput of lower adopters, with even modest adopters seeing 30–60% gains.

Engineering found the first wave of AI leverage.

The next opportunity sits one step earlier: defining the work that enters engineering.

Where Delivery Breaks Down

The constraint in software delivery has rarely been only the speed of writing code.

EY (2026) cites a 59% failure rate for waterfall projects between 2013 and 2020, attributed to the Standish Group's Chaos Report. A significant portion of those failures traced back to unclear requirements and breakdowns between business intent and engineering execution.

AI does not remove that constraint. It accelerates teams into it faster.

A requirement that appears complete in a ticket may still contain unresolved decisions:

• An edge case nobody defined.
• A dependency nobody evaluated.
• A scope boundary nobody agreed on.
• An assumption about the workflow that has never been validated.

When those questions surface during development, faster coding does not reduce rework. It increases the cost of discovering ambiguity later.

This is why some organizations are beginning to apply AI earlier in the delivery lifecycle. InfoQ (2026) reported that Uber uses an AI-assisted first-pass review process to evaluate product requirements for clarity, completeness, and execution risk before engineering begins.

The goal is not to automate product judgment. It is to make sure product decisions are clearer before engineering capacity is applied.

What AI-Assisted Product Work Looks Like

Product teams are already adopting AI quickly.

Productboard's October 2025 survey of 379 enterprise product professionals found that 100% use AI tools, with 96% using them consistently. However, only 65% reported having documented policies for how AI should be used.

The opportunity is moving from individual experimentation to a repeatable operating model.

McKinsey's 2024 study of 40 product managers using AI measured a 5% faster time to market and a 40% productivity improvement. Product functions have significant AI opportunity, but much of the value remains unrealized because the workflows, standards, and decision processes around AI are still developing.

AI can accelerate product work in areas such as:

• Drafting and refining requirements.
• Synthesizing customer feedback.
• Identifying gaps and edge cases.
• Exploring solution options.

But the highest-value product decisions remain human decisions: defining the problem, understanding the customer, and determining what should be built.

EY's 2026 internal case study illustrates the potential. An AI-assisted prototyping tool created a working product version in two days compared with roughly ten weeks using traditional methods. The majority of that difference came from accelerating specification, documentation, and alignment — not simply writing code faster.

Product and Engineering Belong in One Delivery Motion

As AI accelerates engineering, the connection between product definition and engineering execution becomes increasingly important.

V.Two's Pace Car model embeds engineers directly into client delivery teams, bringing AI-assisted workflows across the lifecycle — from story definition and backlog structure through implementation and testing.

The Assessment phase establishes where AI can create immediate value, which decisions require human review, and where workflow changes are needed before work enters a sprint.

When product and engineering operate through one AI-assisted delivery motion, requirements are evaluated against real technical context before implementation begins.

The result is less rework, clearer decisions, and more effective use of AI throughout the delivery process.

Anthropic's 2026 Agentic Coding Trends report describes a similar shift: as AI takes on more tactical implementation work, human expertise increasingly moves toward defining the problems worth solving.

That shift changes the role of product.

The next delivery advantage will not come only from generating code faster.

It will come from defining better work before the code begins.

‍