How frontier teams are reinventing AI-native development

Overview

AWS says the largest gains from AI in software development come from teams that redesign how they work around AI agents, not from teams that treat AI as a faster way to type. Across more than 50 internal teams studied, the 25 groups that adopted both new tools and new practices reached a 4.5x median productivity gain, and some passed 10x in deployment velocity. AWS calls these its frontier teams and lays out five habits behind the results.

Key Takeaways

AWS argues the bottleneck has shifted from how fast code gets written to how well AI agents can access knowledge and how willing a team is to reorganize its work.
Teams that paired new AI tools with new working practices reached a 4.5x median productivity gain, while teams that added AI tools alone saw far less.
One Amazon Bedrock team of 6 engineers finished in 76 days a project first scoped for 30 developers over 12 to 18 months.
On some teams, engineers now handwrite only 1 to 2 percent of the code, with agents producing the rest under human direction.
AWS names five repeatable practices, including investing in agent context and making intent explicit before any code is written.
The tools cited in this work include Kiro, Amazon Q, and Amazon Bedrock.

Stats & Key Facts

#4.5x median productivity gain across the teams that adopted both new tools and new practices
#More than 50 internal teams studied, with 25 that changed both tools and practices outperforming the rest
#Some teams passed 10x improvement in normalized deployment velocity
#One Bedrock team of 6 engineers replaced a plan for 30 developers over 12 to 18 months, finishing in 76 days
#Individual commit velocity on that team rose from 2 per week to 40 per week, roughly 20x
#A Prime Video group produced 556 commits against a baseline of 96 and cut a 90-week estimate to 24 weeks

How frontier teams are reinventing AI-native development

Why AWS Says Workflow Change Beats Faster Typing

The central claim separates two ways of adopting AI.

AWS draws a sharp line between teams that use AI to type code faster and teams that rebuild how software gets made. The first group sees modest gains. The second group, which AWS calls frontier teams, sees the work reorganize around AI agents and the results compound.

The post argues the difference is not the underlying model. It is the set of habits a team builds around the model. AWS frames this as a move from coding faster to redesigning the whole process of building software.

The 4.5x Median Gain Across More Than 50 Teams

The headline figure comes from a study inside Amazon Stores.

›More than 50 teams were observed on real projects, not lab tests.
›The 25 teams that adopted both new tools and new practices reached a 4.5x median productivity gain.
›Some teams passed 10x in normalized deployment velocity.
›AWS measured this as features deployed per sprint, compared against each team's own historical baseline.
›Teams that added AI tools without changing their practices saw much smaller gains.

The Bedrock Team That Replaced 30 Developers With 6

The standout example shows the scale of the shift.

An Amazon Bedrock team of 6 engineers completed in 76 days a project first scoped for 30 developers working 12 to 18 months. AWS reports the team shipped more production code in 5 months than in the prior 10 years.

Weekly commits for individual developers on that team rose from 2 to 40. AWS describes this as roughly a 20x rise in individual output, driven by the way the team fed work to agents rather than the speed of any one person.

Prime Video and Amazon Stores Results

Smaller wins appear across other groups.

›A Prime Video financial systems group produced 556 commits against a baseline of 96, cutting a 90-week estimate to 24 weeks, about 6x throughput and 4x faster delivery.
›A Perfect Order Experience team shipped features in an afternoon instead of two weeks.
›A WW Grocery group wrote design documents in a few hours rather than five days.
›On some teams, engineers now handwrite only 1 to 2 percent of the code.

The Five Practices Behind Frontier Teams

AWS names a repeatable set of habits, not one trick.

›Invest in agent context through steering files and documentation so agents understand the codebase.
›Slow down to speed up, accepting a slower start before the gains compound.
›Feed agents instead of babysitting them, keeping well-scoped backlogs and running several agents in parallel.
›Make intent explicit before any code is written.
›Shift testing earlier with local integration tests and agent self-correction so problems surface sooner.

Swami Sivasubramanian on the Three Multiplying Factors

AWS leadership frames the math behind the gains.

Swami Sivasubramanian, Vice President for Agentic AI at AWS, ties the results to three factors that multiply together: AI handling routine low-judgment work, engineers keeping focus on the hard decisions, and instant access to domain knowledge. The post describes this as roughly a 1.5x effect on each axis, which compounds rather than adds.

He frames frontier teams as optimizing for the rate at which correct, production-ready software reaches customers, rather than the raw speed of generating code. The tools cited in this work include Kiro, Amazon Q, and Amazon Bedrock.

What This Means for Business Readers

The practical takeaway is about organization, not software.

The message for non-technical leaders is that the limiting factor has moved. AWS says speed of code generation is no longer the constraint. The constraint is now how well agents can reach a team's knowledge and how willing the team is to change its habits.

AWS warns the gap between teams that build these habits and teams that do not is widening fast. For a business, the lesson is less about buying a tool and more about restructuring how a team plans, documents, and tests its work.

Frequently Asked Questions

What is a frontier team?

AWS uses the term for its highest-performing internal engineering groups that rebuild their workflows around AI agents rather than using AI only to write code faster. These teams adopt both new tools and new practices together.

How was the 4.5x productivity gain measured?

AWS measured features deployed per sprint, normalized against each team's own historical baseline, across more than 50 teams in Amazon Stores. The 25 teams that changed both tools and practices reached a 4.5x median gain.

Did 6 engineers really replace 30 developers?

AWS reports an Amazon Bedrock team of 6 engineers completed in 76 days a project first scoped for 30 developers over 12 to 18 months. The same team's individual commit rate rose from 2 to 40 per week.

What are the five practices AWS recommends?

Invest in agent context, slow down to speed up, feed agents instead of babysitting them, make intent explicit before writing code, and shift testing earlier. AWS says these habits, not the model alone, drive the gains.

Which AWS tools are involved?

The post cites Kiro, Amazon Q, and Amazon Bedrock as the tools used in this work. AWS stresses the practices matter more than any single tool.

AWS argues the real advantage now belongs to teams willing to reorganize how they build software around AI agents, not to those that simply add an AI tool. As the gap between the two groups widens, the company positions workflow change as the deciding factor.