AI Strategy Index

Amazon AI Strategy: Small Teams, Built Around Agents

How Amazon uses AI on its own work: small teams built around agents, a 76-day rebuild with Kiro, corporate job cuts and a hard lesson in AI metrics. Updated November 2026.

Company
Amazon
Industry
Retail & consumer goods
Headquarters
Seattle, Washington, United States
Region
US
AI strategy
Part of corporate strategy
Index group
S&P 100

Online and physical retail, third-party seller services, advertising, subscriptions and Amazon Web Services (AWS)

Where the AI strategy is set out

AI runs through Andy Jassy’s 2025 letter to shareholders (April 2026), which calls it a multiplier across the company. No standalone AI strategy.

Read the company source

Where Amazon Is Going with AI

Amazon is rebuilding its teams around AI: smaller groups, a flatter organization and agents doing much of the building. Six engineers rebuilt a core AWS engine in 76 days, work Amazon says would normally take 40 people a year. CEO Andy Jassy said in 2025 that AI would shrink the corporate workforce, and Amazon has since cut about 30,000 corporate jobs, while saying AI was not the reason for most of them. It is also unusually open about where its own AI metrics went wrong.

A multiplier, not a project. Jassy wrote in his April 2026 letter to shareholders: “AI is not a standalone initiative—it’s a multiplier” and “You have to be willing to reimagine not only every customer experience, but also how you organize and get work done” (Amazon, April 2026). It is Amazon’s path to an AI-native organization.

Fewer people in some jobs. In June 2025, Jassy told staff: “we expect that this will reduce our total corporate workforce as we get efficiency gains from using AI extensively across the company” (CBS News).

The focus has three parts.

  • Software engineering: Kiro, Amazon’s agentic coding service, used to rebuild core systems with small teams (Amazon)
  • Knowledge work: Amazon Quick, a desktop app that connects email, calendar, Slack and documents with AI search and task execution (GeekWire, 2026)
  • Team design: projects that once needed 30 to 40 people now staffed by six to eight, with product managers writing code (GeekWire)

What It Takes: Smaller Teams, Engineering Rules and a Leaner Corporate Workforce

Small teams and a flatter organization. Jassy wrote that Amazon has “worked to flatten our organization the last year” and wants teammates who “move fast” and act “like true owners” (Amazon). In AWS’s agentic AI division, led by Vice President Swami Sivasubramanian, dozens of small teams start some low-risk projects with a working prototype instead of Amazon’s traditional written proposal, and define specifications and tests before agents write code (GeekWire). It is a working model of hybrid teams of people and AI agents.

Rules for engineers. A revised guideline for software engineers, circulated in September 2026, says that if AI does not help with a piece of work, that is “a finding, not a failing,” and that “AI will be at the center of software engineering practice.” Leaders are expected to have used AI on production work. It calls usage metrics “deeply flawed” as targets, though managers still review the teams with the highest and lowest adoption (Reworked, October 2026).

Governance. AWS sets out eight dimensions of responsible AI, including fairness, explainability, privacy and security, safety, controllability and transparency (AWS). Inside the agentic AI division, AI token spending is tracked as an operating cost, and agents check their own output against tests before code reaches production (GeekWire).

Corporate cuts. Amazon cut about 14,000 corporate jobs in October 2025 and about 16,000 in January 2026, out of about 350,000 corporate staff (Business Insider via AOL). It said AI was “not the reason behind the vast majority of these reductions”, which were about “reducing layers, increasing ownership” (AP via WTOP).

AI fluency, coded against the Lead with AI Fluency Matrix (September 2026 edition):

  • Fluency floor: Not published for all staff. Software engineers are expected to put AI at the center of their work and to explain where it does not help, and leaders are expected to have used it on production work, but there is no company-wide requirement.
  • Highest level evidenced: Level 3 (Integrating), company-claimed. Agentic coding is built into how teams rebuild core systems, with measured time savings. People still write the specifications and own the result, so Level 4 is not evidenced.
  • Portfolio design: Team-led. Amazon redesigns teams around agents, smaller and flatter, and treats tools without restructuring as the weaker path.

How Amazon Measures Results

Amazon reports results by team, with a warning about metrics. All figures below are company-claimed:

  • Bedrock: six engineers built a new inference engine in 76 days with Kiro, work that “might take a team of 40 people about a year” (Amazon)
  • Restructured teams: a median 4.5 times productivity gain, and above 10 times for some, against smaller gains for teams that only added AI tools (GeekWire)
  • Amazon Quick: 10,000 internal users within ten weeks (GeekWire)

A lesson in what not to count. In May 2026, Amazon shut down KiroRank, an internal leaderboard of activity on Kiro, after staff assigned agents to needless tasks to climb it (Reworked). Its engineering guideline now warns against using token use or adoption as targets. Amazon has published no company-wide AI value figure; linking team results to cost and customer outcomes is the step from AI adoption to AI value.

Next markers: a company-wide measure of AI value, how the engineering guideline changes reviews, further corporate cuts, and whether small-team results spread beyond AWS.

Change log

  • June 2025: Jassy says AI will reduce the corporate workforce over time.
  • October 2025: Amazon cuts about 14,000 corporate jobs.
  • January 2026: Amazon cuts about 16,000 corporate jobs and says AI was not the reason for most of them.
  • April 2026: Jassy’s letter calls AI a multiplier and describes the 76-day Bedrock rebuild; Amazon Quick ships.
  • May 2026: Amazon shuts down the KiroRank leaderboard.
  • September 2026: A revised engineering guideline puts AI at the center of software engineering.

The Bottom Line

  • Disclosure: 4 of 4. Amazon answers all four questions, with results by team and no company-wide total.
  • The team is the unit of change: six to eight people doing work that once took 30 to 40.
  • Restructuring beats tools. Amazon’s own figures show far bigger gains where teams were rebuilt around AI.
  • It has learned from bad metrics, shutting a leaderboard that rewarded activity over results.
  • The say-do gap is the workforce. The CEO said AI would shrink the corporate workforce, yet Amazon says AI did not drive most of its 30,000 corporate cuts.
  • Watch the spread. The test is whether AWS’s small-team results show up across retail and operations.

Revenue, Headcount and Revenue per Employee

What the numbers show: Revenue rose 39% from 2022 to 2025 while headcount rose 2%, so revenue per employee rose 36%, to about $455k. Most of Amazon’s 1.6 million people work in warehouses and delivery; its cuts of about 30,000 corporate jobs since October 2025 fall mostly after this period. These figures do not show that AI caused any change.

Revenue, headcount and revenue per employee, FY2022 to FY2025

  • FY2022: $514.0bn revenue, 1,541,000 employees, $334k per employee
  • FY2023: $574.8bn revenue, 1,525,000 employees, $377k per employee
  • FY2024: $638.0bn revenue, 1,556,000 employees, $410k per employee
  • FY2025: $716.9bn revenue, 1,576,000 employees, $455k per employee

Change, FY2022 to FY2025: revenue up 39%, headcount up 2%, revenue per employee up 36%.

Notes: Revenue is total net sales, as reported. Headcount is approximate full-time and part-time employees at year-end, including warehouse and delivery staff. About 350,000 of Amazon’s employees work in corporate roles. Amazon cut about 14,000 corporate jobs in October 2025 and about 16,000 in January 2026.

These figures give context on scale and productivity. They do not show that AI caused any change. Sources: Amazon 10-K filings.

How Amazon Compares with Retail and Consumer Goods Peers

We looked at how Amazon compares with other retail and consumer goods companies in the index. Here is how they stack up.

CompanyDisclosureAI ownerFluency floorGovernanceAI value reportedRevenue per employee
Amazon4 of 4CEO; AWS agentic AI division builds the toolsNot published; AI central to engineeringEight responsible AI dimensions (AWS)Results by team, no total$455k
L’Oréal4 of 4Executive for AI, data and shared servicesLevel 1 in practice, about 73,000 trainedPrinciples stated in one lineResults by use case, no total$510k
Starbucks2 of 4COO, for the store rolloutNot disclosedA design rule, no frameworkPilot results only$98k
Unilever4 of 4No single ownerNot disclosed; 23,000+ trainedEvery AI use reviewed, a named owner accountableResults by use case, no total$571k
Walmart4 of 4EVP of AI accelerationNot disclosedPublished Responsible AI PledgeProduct figures, no dollar value$340k
Target
Notable
3 of 4Chief AI Officer (from August 2026)Not disclosedNot publishedResults by feature$252k

Disclosure counts how many of the four strategy questions (where it is going, where it will focus, what it will take and how it will know it is working) a company answers in public. Fluency floor is the lowest level of AI fluency a company expects of all its people, on the five levels of the Lead with AI Fluency Matrix: Avoiding (0), Chatting (1), Collaborating (2), Integrating (3) and Transforming (4). The other columns summarize each company’s profile. Revenue per employee is for each company’s latest fiscal year, in US dollars. Rows marked AI Pacesetter or Notable are companies outside the S&P 100 and STOXX Europe 50. They are shown for contrast and are not counted in any index figure. See all companies in the AI Strategy Index.

Frequently Asked Questions

What is Amazon’s AI strategy for its own work?

Amazon treats AI as a multiplier across the company and is reorganizing into smaller, flatter teams that build with AI agents, starting with software engineering in AWS.

‍

Who leads AI at Amazon?

CEO Andy Jassy sets the direction. Within AWS, Vice President Swami Sivasubramanian leads the agentic AI division, which builds tools such as Kiro and Amazon Quick and uses them on its own work.

‍

What is Kiro?

Kiro is Amazon’s agentic coding service. Amazon says six engineers used it to rebuild a core Bedrock engine in 76 days, work that would normally take 40 people about a year.

‍

Is Amazon cutting jobs because of AI?

Amazon cut about 30,000 corporate jobs between October 2025 and January 2026. Jassy said in 2025 that AI would shrink the corporate workforce over time, but Amazon said AI was not the reason for most of these cuts.

‍

How does Amazon measure AI use?

It tracks adoption but warns against using it as a target. It shut down an internal Kiro leaderboard in 2026 after staff gamed it, and its engineering guideline says usage metrics are deeply flawed.

‍

How is this profile put together?

Lead with AI profiles use public sources only, such as annual reports, earnings materials, filings and on-record interviews, dated at the time of writing. AI fluency is coded against the Lead with AI Fluency Matrix. Company figures are marked as company-claimed unless they have been independently verified.

About the AI Strategy Index

The Lead with AI AI Strategy Index profiles how the largest companies in the US and Europe approach AI, starting with the S&P 100 and the STOXX Europe 50. Each profile answers four questions: where the company is going with AI, where it will focus, what it will take and how it will know it is working. AI Pacesetters and notable companies outside the two indexes are profiled for contrast and are not counted. Profiles use public sources only, and AI fluency is coded against the Lead with AI Fluency Matrix.

See all company profiles