How AI Agents Are Forcing Dev Teams to Rethink Their Work

Software teams have spent decades chasing faster delivery cycles—from shipping twice a year in the mainframe era to multiple releases daily in the cloud era. But according to Hannah Foxwell, a prominent voice on engineering culture, the industry has finally unlocked the velocity it always wanted. The new problem? Figuring out what to do with it.

The rise of agentic AI platforms like Cursor—where agent-driven development now accounts for the majority of work—signals a fundamental shift in how development gets done. Agents are handling code generation, testing, deployment, and even production monitoring. For many organizations, this creates an unexpected bottleneck: product teams cannot generate enough well-qualified work to keep developers busy.

The Backpressure Problem#

The traditional software team structure pairs six to eight developers with a single product manager. That ratio emerged under the assumption that engineering capacity was the constraint. Now, with agents handling much of the coding work, product has become the limiting factor.

Some organizations are experimenting with new team shapes to address this. NBIM is testing a 2:1 developer-to-PM ratio, compressing decision-making and feedback loops. Meanwhile, entrepreneur Andrew Ng has proposed the inverse: two product managers per developer, assuming that single engineer orchestrates a fleet of agents across multiple product lines.

A consultancy working across 200 companies found that only two or three have reduced team size in response to agentic coding gains. Most expect increased output instead—raising the question of whether teams will need to expand the scope of domains they own.

Building Products That Matter#

Faster shipping doesn't automatically mean better products. The real risk, Foxwell argues, is "enshitification"—platforms bloated with unvalidated features that do many things poorly instead of one thing well.

The good news: the cost of experimentation has collapsed. When engineering capacity was scarce, teams committed to roadmaps through confirmation bias rather than user validation. Now, prototyping and testing ideas before shipping them to production is economically feasible.

New patterns are emerging:

• Vibe coding PMs: Product managers building quick prototypes to validate ideas before committing them to the product.
• Forward-deployed engineers: Engineers embedded with customers, empowered to solve real problems and shorten feedback loops.
• Product engineers: Engineers who serve as users themselves (common in developer tools), reducing the need for traditional product proxies and enabling direct customer access.