Beth Galetti, Amazon's senior vice president of People Experience and Technology, published a first-person account on the company's news site on August 25, 2026 describing how she wrote more than 100,000 lines of code with AI tools after a 25-year gap, and tying that account to a $2.5 billion training commitment aimed at at least 50 million people.

The post runs to roughly four minutes of reading and carries Galetti's own byline rather than a corporate communications attribution. It is unusual for that reason alone. Amazon's HR organisation is the function that has absorbed and communicated the largest white-collar workforce reductions in the company's history over the preceding ten months, and the executive who signed those memos is the same executive now describing the pleasure of building software again.

An HR executive returns to code

Galetti's technical background is not incidental to the story she tells. According to the post, she began her career as a software developer in the transportation industry, building package tracking and international systems. Over nearly two decades she moved into senior leadership, running transportation businesses as chief information officer before taking responsibility for full air and ground networks spanning engineering and operations.

The pivot into human resources came in 2013. According to Galetti, she was interviewing at Amazon for a transportation role when the recruiter asked an unexpected question: "What would you think about doing a job in HR?" The interviewers, she writes, had noticed that she became animated when the conversation turned to developing people. She took the role and now leads Amazon's global HR organisation.

The gap between running engineering teams and writing code is the hinge of the account. "Throughout my career, I understood how software was built and led teams that built it. I just wasn't the one writing it anymore," Galetti wrote.

The restart was deliberate and small. According to the post, she decided a few months ago to test what would happen if she started building again with AI assistance, choosing a project outside work: a family calendaring app to keep household schedules aligned. She describes being nervous before starting and having something functional within a couple of hours.

The tool stack behind the claim

Three internal systems appear in Galetti's account, and the distinctions between them are worth setting out precisely because they map to three different categories of enterprise AI deployment.

Kiro is the development tool. According to Amazon, a user describes the desired outcome in plain language and Kiro helps plan the work, write the code, and test it. Galetti reports opening Kiro on a flight back to Seattle and landing with a working submission application. The interaction model is a plain-language prompt rather than a specification document, which is the design decision that determines who can use the product at all.

Amazon Quick is the general work assistant. Teams use it to query data, run research that previously took weeks, and construct automated workflows, according to the company. PPC Land documented the Quick desktop application launch on April 28, 2026, which added a persistent personal knowledge graph, background monitoring, and native connectors spanning Google Workspace, Zoom, Airtable, Dropbox and Microsoft Teams. That release also connected Quick to developer tooling including the Kiro command-line interface and Anthropic's Claude Code, and Amazon stated at the time that internal users had built applications and deployed them to thousands of colleagues without a development process.

Aza is the workplace assistant, and it is the narrowest of the three. According to Amazon, it answers questions about benefits, workplace accommodations, compensation and IT, and handles transactional tasks: booking meeting rooms, applying for leave, searching open roles. Galetti reports that Aza was the entry point for her own project, directing her to the right tools before she opened Kiro.

Both Quick and Kiro are available outside Amazon as commercial products. Aza is not. The post does not disclose pricing, seat counts, or internal adoption figures for any of the three.

The structural claim Amazon makes about this stack is a claim about queueing. According to the company, a non-technical employee who wanted something built previously needed a technical team and a slot on that team's roadmap. Removing the roadmap dependency, in Amazon's framing, frees engineering capacity rather than substituting for it.

What 1,500 employees built

The submission application Galetti built on the flight served an internal contest called "Everyone Can Build." The brief was open: build an app or tool that solves a real problem in work or everyday life using AI, with no technical background required.

According to Galetti, the competition was opened to every employee on her team, and more than 1,500 people were building within a few weeks.

Two submissions appear in the post. One employee built an application that converted a manual process taking nearly four hours into a one-second automated check, which Amazon says saves tens of thousands of hours annually. Another built a desktop assistant named Momo that keeps tasks, reminders and shortcuts within a single click.

"I couldn't believe what they built," Galetti wrote. Elsewhere in the post she frames the participation threshold rather than the output: "You don't need perfection to start. You just need to dive in."

Since the contest, according to the post, Galetti has written more than 100,000 lines of code with AI assistance.

The context the post does not mention

The August 25 post contains no reference to the workforce reductions Galetti communicated to Amazon employees over the preceding ten months. Those reductions are the largest in the company's corporate history and they were announced under her signature.

Amazon announced 14,000 corporate job cuts during the week of October 28, 2025, with Galetti describing artificial intelligence as the most transformative technology the company had seen since the internet. Three months later, Amazon confirmed a further 16,000 corporate positions, completing a reduction of approximately 30,000 roles since October 2025. Galetti addressed the cadence question directly in that January communication: "Some of you might ask if this is the beginning of a new rhythm - where we announce broad reductions every few months. That's not our plan."

That second round also drew allegations from self-identified Amazon software engineers that American workers had been disproportionately affected relative to H-1B visa holders on some teams. United States Citizenship and Immigration Services data placed Amazon first among all United States employers with 13,265 approved H-1B beneficiaries in fiscal year 2025.

The juxtaposition matters for how the August post is read. A narrative about employees gaining the ability to build software themselves sits inside a company that has removed roughly 30,000 corporate roles while doing so. Amazon does not connect the two in the post, and the post does not claim they are connected.

What the evidence says about AI-assisted building

The 100,000-line figure invites a measurement question that Amazon does not answer. Lines of code has been a contested productivity proxy for decades, and the post does not state the period over which the total accumulated, the counting methodology, or whether generated code that was later discarded is included.

The research record on AI-assisted development is also more mixed than a first-person account can convey. Anthropic published a randomised controlled trial involving 52 software engineers learning an unfamiliar Python library, and PPC Land reported that developers using AI assistance scored 17% lower on comprehension tests despite finishing marginally faster. Participants using AI averaged 50% on knowledge quizzes covering concepts they had just applied, against 67% for those coding manually.

Experienced practitioners have been publicly sceptical of the unsupervised variant of this workflow. PPC Land's coverage of why expert developers reject fully AI-driven coding documented a consensus favouring close human supervision over delegation. Separately, Stanford researchers argued in a paper covered by PPC Land that industry productivity claims for agentic AI rest on an evaluation imbalance, with return-on-investment calculations routinely excluding human oversight and error-correction costs.

None of that contradicts Galetti's account. It does establish that a first-person report of volume produced is a different category of evidence from a controlled measurement of value delivered, and Amazon's post makes no attempt to bridge that gap.

Future Ready 2030, restated

The second half of the post shifts from tooling to money. According to Amazon, Future Ready 2030 is a $2.5 billion commitment to expand access to education and skills training and prepare at least 50 million people for the future of work.

The programme spans several existing vehicles. Amazon Future Engineer targets K-12 students. AWS Educate is positioned as an entry route into cloud-based roles. The Mechatronics and Robotics apprenticeship is the one carrying a specific outcome figure: according to Amazon, graduates can earn up to 58% more per year than typical starting wages for entry-level roles.

Within Future Ready 2030, Amazon is directing $1 billion to Career Choice, the education benefit that prepays tuition for degrees and skills development, with a stated goal of upskilling a further 500,000 employees globally. The focus areas named are cybersecurity, software development, logistics and renewable energy.

That $1 billion figure is not new. Amazon announced it at the Delivering the Future event in London on June 4, 2026, alongside a 10 billion euro European fulfilment commitment and a next-generation Proteus robot. PPC Land's coverage of that event recorded that Career Choice sits as a centrepiece of the $2.5 billion Future Ready 2030 initiative and has already been used by more than 300,000 employees globally, including 30,000 in the United Kingdom. The August 25 post restates the commitment rather than extending it.

One element does appear to be specified more precisely here than previously. According to Amazon, AI training is free for every employee, and the company covers the full cost of one AWS AI certification exam per year, with options ranging from foundational concepts to professional-level generative AI credentials. The post does not state pass rates, uptake, or how many employees have taken the exam.

Why the marketing industry is watching

Amazon is not a peripheral actor in advertising, which makes its internal tooling posture commercially relevant rather than merely cultural. The company reported second-quarter 2026 advertising services revenue of $19.8 billion on July 30, 2026, a 26% year-over-year increase and the strongest rate across the six quarters covered by its supplemental disclosures. On the same day, Amazon Ads consolidated DSP advertiser accounts across 34 countries into a single login, a plumbing change published hours after the revenue figure it supports. The company simplifies interfaces for its own buyers on the same principle it applies to its own staff: remove the step that requires a specialist.

The skills question Galetti raises is one the marketing profession has been answering badly by its own account. Brandwatch surveyed 1,028 marketing professionals and found that 84% named AI and automation the most important skills to master in 2026, the highest single result in that survey, with 79% saying management of AI and automation was already among the fastest-growing parts of their daily work. MiQ's research across 3,169 marketers in 16 countries found a 27-percentage-point gap between how many planned to increase AI use and how many felt ready to do so.

The displacement side of the ledger is documented as well. IAB Australia's 2026 talent review found entry-level roles at just 1% of vacancies, with AI fluency treated as a hiring baseline and pressure concentrated in the process-driven positions that historically trained early-career staff. The IAB Tech Lab's chief executive had made a similar observation earlier, reporting that publishers were closing open junior ad operations roles because agentic tooling absorbed the work. Where permanent headcount contracts, contractor arrangements tend to absorb the residual volume.

Policy has begun to move in response. Google backed a package of bipartisan United States bills in April 2026 that included the AI Workforce Training Act, proposing a 30% tax credit for businesses providing AI training to employees. Amazon's $2.5 billion is a private commitment sitting in the same policy space, and it is roughly an order of magnitude larger than any single legislative mechanism currently proposed.

For agency and in-house teams, the operational read is narrower than the headline. The Amazon post describes a specific pattern: an organisation that gave every employee access to build tooling, ran a low-stakes internal competition to generate first attempts, and reported participation in the thousands within weeks. Whether the resulting artefacts survived in production is not addressed.

What the post does not establish

Several gaps limit how far the figures can be read.

Amazon does not disclose the headcount of the People Experience and Technology organisation, so the 1,500 participants cannot be converted into a participation rate. The company reported approximately 350,000 corporate employees before the October 2025 reductions, but the HR function's share of that total is not public.

The four-hour to one-second automation claim is self-reported and unaudited, as is the estimate of tens of thousands of hours saved annually. No baseline methodology accompanies either number.

The post does not indicate how many of the tools built during "Everyone Can Build" remained in use after the competition closed, nor whether any required subsequent engineering support - the cost that determines whether the roadmap-avoidance argument holds.

Finally, the framing of the $2.5 billion is a restatement of commitments already public since June 2026, presented here alongside a personal narrative rather than as a new financial disclosure. Readers tracking Amazon's workforce spending will find no incremental figure in the August 25 post beyond the AWS certification detail.

Timeline

Summary

Who: Beth Galetti, senior vice president of People Experience and Technology at Amazon, writing under her own byline, together with the employees of her organisation who took part in the internal "Everyone Can Build" competition.

What: A first-person account stating that Galetti wrote more than 100,000 lines of code with AI tools after a 25-year gap, that more than 1,500 employees on her team began building within a few weeks of an internal contest, and that Amazon's $2.5 billion Future Ready 2030 programme is intended to prepare at least 50 million people for the future of work, including $1 billion directed to Career Choice and a further 500,000 employees targeted for upskilling.

When: Published August 25, 2026, roughly ten months after the October 28, 2025 announcement of 14,000 corporate job cuts and seven months after a second round of 16,000 that brought the cumulative reduction to approximately 30,000 positions.

Where: Amazon's corporate news site, describing tooling deployed across the company's global workforce, with Kiro and Amazon Quick also sold externally and Aza restricted to internal use. Career Choice operates globally, with more than 300,000 prior participants including 30,000 in the United Kingdom.

Why: The account matters to marketing and advertising professionals because Amazon is simultaneously the operator of a $19.8 billion quarterly advertising business, a supplier of the AI development tooling in question, and the largest single case study in AI-adjacent corporate workforce reduction. It is a document about capability expansion published by the executive responsible for the contraction, and the industry facing the same trade-off - entry-level roles at 1% of Australian ad vacancies, an unresolved readiness gap across thousands of surveyed marketers - has an obvious interest in whether both things can be true at once.