Swami's Boomerang List: 2,896 Ex-Amazon ML Engineers to Reach First

Business Insider got the receipts: Amazon recruiters are emailing laid-off AI/ML staff under what one of them called "Swami's Boomerang Reengagement Initiative," pitching expedited interview loops back into the AWS Agentic AI org. If you source AI/ML talent for anyone other than Amazon, this is a starter pistol. Amazon just prequalified a target list for you and told you the deadline is "before the loop closes."

What Amazon just told the market

Amazon is racing to rehire the AI/ML engineers it cut, which means the market has priced the risk on this cohort at zero and any competitor matching base pay wins. The internal program is attributed to Swami Sivasubramanian, VP of Agentic AI at AWS, and targets alumni from the 30,000-person corporate cut spread across October 2025 and January 2026.

The facts, as reported:

• 30,000 corporate layoffs total: 14,000 in October 2025 and 16,000 in January 2026.
• The January 2026 round alone was roughly 4.5% of Amazon's 350,000-person corporate workforce.
• October 2025 cuts cost Amazon $1.8 billion in severance.
• A fresh, undisclosed round hit the AGI unit on July 22, 2026.
• Named divisions in scope: AWS Professional Services, Alexa, Prime Video, advertising, Applied AI Solutions, AGI.
• Impacted geographies on the Colleen Aubrey email: US, Canada, Costa Rica.

Amazon spokesperson Haley Silva told Business Insider that boomerang hiring is "a normal and longstanding recruiting practice" and denied the program specifically targets RTO-leavers. Treat both statements as confirmation. Fast-track loops do not exist unless internal demand is desperate and validated, and PR departments do not deny things that are not happening.

30,000

The 2,896-profile addressable pool, split two ways

The addressable pool of Amazon-tagged US ML talent is roughly 2,896 profiles, and it splits cleanly into two keyword slices that most sourcers merge by mistake. In Refolk's index of professional profiles, searching for US-based Applied Scientist and ML Engineer titles with "Amazon" in the headline returns 1,797 people. Swapping that keyword to "AWS" returns a different, smaller 1,099-person cohort that identifies by business unit rather than parent brand.

The two slices overlap but are not duplicates. A sourcer who runs only "Amazon" misses roughly 1,100 AWS-tagged ML profiles. A sourcer who runs only "AWS" misses the Alexa and AGI people, who tend to brand themselves "Amazon." If you are doing this in a LinkedIn Recruiter seat, you need two saved searches, two projects, and two message cadences. If you are doing it in Refolk, you describe the person in plain English ("US applied scientists who worked on Alexa or AWS Agentic AI in the last three years") and get the merged, deduped shortlist back ranked by recency.

Why the AWS-tagged cohort is the warmer lead

The "AWS" slice is more poachable than the "Amazon" slice because it has already churned at a higher rate. In Refolk's sample, only 44% of "AWS" ML profiles are still at AWS today, versus 68% for the "Amazon" slice. That gap is not noise. It is the mechanism.

The AWS-brand identifiers skew toward Solutions Architects, Technical Account Managers, and Professional Services staff, which is exactly the title set the January 2026 cuts hit hardest. These people were already more mobile than core research scientists before the layoffs, because their work is customer-facing and their skill set ports cleanly to any cloud or data vendor. Now they are 10 to 14 months out of role, severance is thin, and Swami's recruiters are the first warm touch many of them have had.

Counter-position accordingly:

1. Lead with location flexibility. Amazon's own spokesperson had to deny the program targets RTO-leavers, which tells you RTO is the open wound.
2. Match or beat base, not equity. Alumni discount Amazon RSU grants after watching two years of stock chop.
3. Skip a loop stage. If Amazon is expediting, matching the expedite is table stakes.
4. Pitch the team, not the mission. The recruiter email pitch ("there's some really compelling work happening across AIML right now") is deliberately vague because the org chart is in flux. Name your tech lead.

Fast-track loops do not exist unless internal demand is desperate and validated. Match the expedite or lose the candidate.

Four named sub-pools worth separate shortlists

Treat "ex-Amazon AI/ML" as four distinct markets, not one, because the orgs cut different work and the alumni cluster around different competitors. The research confirms four named sub-pools in scope across the 2025-2026 cuts.

• AWS Professional Services. Solutions Architects and Technical Account Managers. Natural destinations: Databricks field eng, Snowflake SA org, any Series C selling into Fortune 500.
• Alexa AI. Speech, NLU, on-device inference. Natural destinations: Anthropic, Perplexity, voice-agent startups.
• AWS Agentic AI / Applied AI Solutions. The Colleen Aubrey org. Agent frameworks, Bedrock tooling. Natural destinations: Anthropic, Microsoft Azure AI's agent group.
• AGI unit (July 22, 2026 cut). Freshest pool, count undisclosed, largely un-pitched because it landed after most layoff trackers froze. Natural destinations: frontier labs, stealth-mode founder bets.

The AGI cut is the one to prioritize. It is roughly 10 weeks old at the time Business Insider's story broke, and most competitor sourcing teams had already drained their outreach budget on the October and January cohorts. If you want a shortlist you are not sharing with four other recruiters, start there.

The Costa Rica arbitrage

Costa Rica is the geography most sourcers will miss, and it has the shortest severance runway of the three countries named in the leaked Aubrey email. The internal AWS note confirmed layoffs in the US, Canada, and Costa Rica. Every US-focused tracker and every LinkedIn Recruiter search anchored on US zip codes will miss the Costa Rica pool entirely.

Why this matters in practice:

• English fluency is the default inside AWS San Jose.
• Severance norms in Costa Rica deliver 6 to 9 months of runway at the senior IC level, not the 12 to 18 months a US L6 gets.
• Remote-first US and European employers can hire there through an EOR in under three weeks.
• Competition for the cohort is almost entirely intra-Costa Rica right now.

If you run a remote-first team, this is the single highest-leverage move in the whole Swami story. Build a Costa Rica + ex-AWS + applied AI slice, send 40 messages, and you will outbook any US sourcer running the same filters on LinkedIn. Refolk is built for exactly this kind of plain-English query: "ex-AWS applied scientists currently in Costa Rica" returns the slice without you fighting a geo filter that assumes everyone lives in King County.

1.63x

The 30-day outreach plan

Ship cold outreach to this cohort inside 30 days or Swami's recruiters will close the loops first. Expedited interview processes at Amazon's scale run four to six weeks end to end, which means the window to insert a competing offer is now, not after the holidays.

A workable sequence:

1. Week 1. Build two shortlists, "Amazon" and "AWS" keywords, filtered to applied scientist and ML engineer titles, US and Costa Rica, departed in the last 14 months. Dedupe. Expect 1,800 to 2,900 names before scoring.
2. Week 1. Score for the four sub-pools (Pro Services, Alexa, Agentic AI, AGI). Prioritize AGI alumni from the July 22, 2026 cut.
3. Week 2. Send a first-touch message that names the Amazon team they came from, matches on base, and offers remote. Reference the fact that you know expedited loops are in flight. Do not pretend you do not know.
4. Week 3. Second touch, with a specific tech lead or founder name. Alumni do not reply to recruiters, they reply to peers.
5. Week 4. Pull a fresh slice. The AGI unit is still shedding people and the Pro Services wave from January 2026 is now hitting the end of severance.

Alumni talent pool sourcing rewards speed over cleverness, and the Swami story is a 30-day speed test.

What the competitor set should actually say

The pitch that beats Amazon's recruiter email is specific, named, and dated, not inspirational. Amazon's own pitch language ("there's some really compelling work happening across AIML right now") is a tell: the recruiter does not know what the candidate will work on because the org chart is still moving.

Beat it with three sentences:

• Sentence one names the Amazon team the candidate came from and the specific problem that team was solving when they left.
• Sentence two names your tech lead, your model or product, and a dated milestone (shipped, raised, launched).
• Sentence three offers a 30-minute call with that tech lead inside 72 hours, remote, with a comp band attached.

Anthropic, Microsoft Azure AI, Databricks, Snowflake, and Perplexity are all currently hiring the exact title set Amazon just cut. If you are one of them, your advantage is not brand, it is response time. If you are a Series B founder, your advantage is that you can name the tech lead because the tech lead is you.

FAQ

How many ex-Amazon AI/ML engineers are actually available right now?

Refolk's index returns roughly 2,896 US-based Applied Scientist and ML Engineer profiles that self-identify with Amazon or AWS in their headlines, across two keyword slices with some overlap. Not all of them are departed, but the "AWS" slice shows 56% have already left and the "Amazon" slice shows 32%, which puts the addressable departed pool well into the four figures before you add Canada, Costa Rica, or title variations like "research scientist."

Why focus on AWS-tagged profiles over Amazon-tagged ones?

The AWS-tagged cohort has already churned at a higher rate (44% still at AWS versus 68% still at Amazon in the Refolk sample), and skews toward Solutions Architects and Pro Services titles that the January 2026 cuts hit directly. They are warmer leads because they have been in-market longer, severance is thinner, and their skills port to any cloud or data vendor without a research-lab story.

Is boomerang hiring really a competitive threat or just PR?

It is a threat. Amazon spokesperson Haley Silva framed it as routine, but the leaked recruiter language ("Swami's Boomerang Reengagement Initiative") and the expedited interview loops both indicate validated internal demand. Fast-track loops get approved only when a VP is personally unblocking headcount, which is exactly what happens when Swami Sivasubramanian needs to staff Agentic AI from a known-quality alumni pool before competitors reach them.

What is the single highest-leverage move in this story?

Build a Costa Rica slice. The leaked Colleen Aubrey email confirmed Costa Rica was in scope for the Applied AI layoffs, almost no US sourcer runs that geography, severance runway is 6 to 9 months instead of 12 to 18, and English fluency is the default inside AWS San Jose. One well-written plain-English query returns a cohort nobody else is messaging.