First 90 Days EM at Amazon: 1:1 Setup Use Case for Large Teams
The new engineering manager who survives week three without a 1:1 system already has attrition happening invisibly. I watched a seven-year Amazon veteran walk out of a post-mortem in 2022 because his skip-level had no mechanism to surface that two of his best engineers were interviewing elsewhere until their offers were already signed.
What 1:1 Structure Does Amazon Actually Expect in the First 90 Days?
Amazon does not hand you a 1:1 template. The expectation is that you build operational cadence yourself, and the bar for "good enough" is higher than at peer companies because of the sheer surface area of a typical EM role. A new EM at AWS typically inherits 12 to 18 direct reports, sometimes more if the team is mid-restructure. The first 90 days are judged not by whether you ship features, but by whether you stop the bleeding of organizational debt.
In a Q3 2023 debrief for an AWS Storage EM role, the hiring manager explicitly rejected a candidate who had managed 15 people at Microsoft but described 1:1s as "mostly status updates and career conversations." The hiring manager's write-up: "Does not understand Amazon's operational model. 1:1s are control planes." That phrase—control planes—stuck with me. At Amazon, 1:1s are not relationship-building exercises. They are the primary mechanism for detecting variance in a system that otherwise operates through written documents and sparse synchronous communication.
The counter-intuitive truth is this: the more senior your team, the more structured your 1:1s need to be. Not less. New college hires need coaching. Principal engineers need escalation detection, cross-organization alignment verification, and subtle political intelligence gathering. The 1:1 is where you learn that a principal engineer's "quick question" about another team's roadmap actually signals an impending architecture dispute that will consume six weeks if unaddressed.
My first 1:1 system at Amazon, inherited from a director who had managed through the 2015 growth spike, had four buckets: operational health (on-call, sprint velocity, blockers), career trajectory (promotion readiness, mobility interest), organizational intelligence (cross-team friction, leadership perception), and personal state (burnout indicators, life events). Each had explicit prompts. Each was documented in a running Quip that both parties could edit. The discipline was not the conversation itself. It was the signal extraction from noise.
How Do You Scale 1:1s When You Have 15+ Direct Reports?
You cannot run 15 personalized 1:1s without a system. The managers who try end up with generic conversations that fail both the control plane and relationship functions. The solution is tiering by risk and investment, not by personal preference.
In early 2023, I consulted with an EM at Alexa Shopping who inherited 19 reports after a reorganization. She had 30-minute 1:1s weekly with everyone for the first month. By week five, she was exhausted and her notes were empty of actionable content.
In the skip-level, her director identified that three of her highest-performing engineers had quietly disengaged because the 1:1s had become transactional—she was so focused on covering everyone that she covered no one deeply. The fix was a triage model: weekly 30-minute with high-risk or high-investment reports, biweekly 45-minute with stable performers, monthly with tenured principals who needed minimal touch. The risk categorization was explicit and shared: "You're in the weekly bucket because we're navigating a promotion cycle" or "You're monthly because you're operating autonomously and I trust the signal."
The specific structure that survived Amazon's review cycles had three non-negotiable elements. First, a shared running document in Quip or Amazon's internal wiki, populated before the meeting by the report, not the manager. The prompt: "What do I need to know that I don't know to ask?" This inverted the preparation burden and surfaced issues the manager could not predict.
Second, a hard stop at 80% of scheduled time. The remaining buffer was for the manager's own notes and for any emergency escalation that emerged. Third, a monthly synthesis into a private manager-only document tracking patterns: who mentioned on-call fatigue, who referenced conflicts with specific other teams, whose career conversations were accelerating or stalling.
The mistake most new EMs make is treating all 1:1s as equal. They are not. The problem is not that you have too many reports. It is that you have not made the hard decision about where to allocate scarce managerial attention.
What Does the First 1:1 with Each Report Actually Look Like?
The first 1:1 is a calibration exercise, not a get-to-know-you. The goal is to understand how this person uses you, not to imprint your style on them. I learned this from a principal engineer in EC2 who refused to meet with his new manager for three weeks because the previous EM had used first 1:1s to announce process changes. The relationship never recovered.
My first 1:1 script, refined across two teams and verified with a senior manager in Prime Video, had five questions. One: "What does a good week look like for you?" This revealed whether they measured themselves by output, recognition, learning, or stability—categories that required entirely different managerial support. Two: "When have you been most frustrated in the last six months?" This was not venting.
It was a diagnostic for systemic failures the previous manager had missed or caused. Three: "What do you need from me that you're not getting?" The silence after this question was often more informative than the answer. Four: "How do you want to receive critical feedback?" Some wanted it immediate and verbal; others needed written reflection time.
Violating this preference degraded trust permanently. Five: "What would make you leave?" This was not optional. In a 2022 loop for a senior EM role at Amazon, a candidate described asking this and then acting on the answer by restructuring on-call rotation. The hiring committee vote was unanimous yes. Another candidate in the same loop had never asked it. The debate lasted 22 minutes before a no-hire.
The specific outputs of these first 1:1s were documented in a template with timestamps. Not for HR. For the manager's own calibration six months later, when it became clear whether the initial assessment matched reality. The drift was itself a signal. If I rated someone as "high growth trajectory, needs stretch assignment" in month one and they were stagnant in month six, the error was usually mine—in the initial diagnosis or in the follow-through.
> 📖 Related: Zero Trust vs Perimeter-Based Security: Amazon Cloud Security Engineer Interview
How Do You Use 1:1s to Survive Amazon's Operational Review Culture?
Amazon's leadership principles are enforced through mechanisms like OP1/OP2 planning, operational excellence reviews, and the infamous six-pagers. The 1:1 is where you discover whether your team can survive these mechanisms intact. The new EM who misses this connection plans for process. The successful EM plans for human capacity under process load.
In a 2023 operational excellence review for an AWS database service, a team missed its availability target for the third quarter running. The root cause, unearthed only in a skip-level 1:1 six weeks prior, was that the on-call engineer was also the primary author of the six-pager for the review—and had been for three cycles.
The manager had distributed work evenly on paper. In practice, the same three people absorbed all invisible load. The 1:1 system had no mechanism to surface this because the manager's questions were about "how are you doing" rather than "walk me through your last two weeks hour by hour."
The specific 1:1 adaptation I used was the "load audit." Every eighth 1:1, the conversation was exclusively about time allocation. The report brought their calendar. We categorized: core work, administrative overhead, work driven by my requests, work driven by other teams, and "invisible work"—the mentorship, document review, and informal coordination that did not appear in any planning system.
The pattern that emerged across a team of 14 was stark. Three people carried 40% of invisible work. Two were burning out. One had already updated their LinkedIn and was interviewing at Stripe.
The counter-intuitive truth: 1:1s at Amazon are not for building trust. They are for verifying that the operational system is not grinding people into dust. Trust is a side effect of accurate load perception. The manager who knows where the pain is and does nothing loses trust faster than the manager who never asks.
Preparation Checklist
- Audit existing 1:1 artifacts before your start date. Request the previous manager's running documents, if they exist, to identify which reports have documented concerns versus stable histories. The absence of documentation is itself signal.
- Design your tiering system before day one. Define criteria for weekly, biweekly, and monthly cadence explicitly, and be prepared to communicate them directly to reports in first conversations.
- Build a private pattern-tracking document. This is not a performance review input. It is your own managerial journal to identify systemic issues before they surface in operational reviews.
- Work through a structured preparation system. The PM Interview Playbook covers Amazon-specific leadership scenarios with real debrief examples, including how EMs are evaluated on operational cadence in loop feedback.
- Schedule your own 1:1 with skip-level and director in week one. The new EM who waits to be invited signals poor self-direction. The specific ask: "What do you wish the previous EM had surfaced earlier?"
> 📖 Related: Bias for Action vs Have Backbone: STAR Story Template for Amazon PM Conflicts in 2026
Mistakes to Avoid
BAD: Running identical 30-minute 1:1s with all 15 reports weekly, focusing on personal check-ins and career aspiration discussions.
GOOD: Tiered cadence with explicit risk criteria, shared pre-read documentation, and 80% time discipline. The personal check-in is one of four structured buckets, not the dominant mode.
BAD: Using 1:1 time to relay information that could be written. "I wanted to tell you about the reorganization directly" burns relationship capital on low-leverage communication.
GOOD: Written async for information, 1:1 time for discovery and calibration. "I sent the reorg details in writing. In our time, I want to understand what this changes for you specifically."
BAD: Treating principal engineers and senior engineers identically. The same question set applied to a new college hire and a tenured principal produces either condescension or missed signal.
GOOD: Adjusted prompt depth. With principals: "What decisions are you making that I don't know about?" With new hires: "Walk me through your current blockers and who you've already asked for help."
FAQ
How do I handle reports who resist structured 1:1s?
The resistance is diagnostic, not an obstacle. In my experience, engineers who push back explicitly—"I don't see the value in weekly meetings"—often have the most to surface once you demonstrate the structure serves them. One principal at Amazon Logistics told me in week two: "I don't need hand-holding." I replied: "The structure is so I don't waste your time.
If nothing surfaces in three sessions, we'll revisit cadence." In session two, he raised a cross-team dependency that had been stalled for six weeks. The issue was not lack of need. It was lack of trust that managerial attention would result in action.
What if my skip-level disagrees with my 1:1 tiering?
Disagreement is common when the skip-level has a different risk model. In a 2023 situation with a new Director of Alexa, I had placed a high-performing senior engineer in monthly 1:1s. The director wanted weekly, citing visibility needs. The resolution was to separate signal from touch: I maintained monthly 1:1s but added weekly written async updates for that report. The director got visibility. The engineer retained autonomy. The key was not defending my system but translating it into the skip-level's actual concern—information flow, not relationship frequency.
How do I know if my 1:1 system is working by day 60?
The signal is not satisfaction. It is predictive accuracy. You should be able to predict, with specific names, who will raise concerns in operational reviews before they happen. You should know which two to three people are at highest flight risk before they update LinkedIn. You should have surfaced at least one systemic issue—load imbalance, process failure, cross-team conflict—that was not visible in any written system. If your 1:1s only confirm what you already knew from documents, they are not working. The test is new signal, not comfortable conversation.amazon.com/dp/B0GWWJQ2S3).
TL;DR
What 1:1 Structure Does Amazon Actually Expect in the First 90 Days?