Template: 1on1 Agenda for IC Engineers at Google to Discuss Systemic Impact in AI Reviews

What Is the Actual Purpose of a 1on1 Agenda at Google for AI Impact Discussions?

A 1on1 agenda at Google is not a status report. It is a lobbying document for your promotion packet.

In the Google AI Reviews process—formally the Systemic Impact Review for AI/ML launches—Individual Contributors face a structural trap. Your promotion to Senior (L5) or Staff (L6) requires demonstrated cross-organizational impact. Yet the AI Review process, run by the Responsible AI team with final sign-off from the Central AI Review Board (CAIRB), operates as a black box to most ICs.

I sat in a Q3 2023 debrief for a Google Cloud L6 ML engineer where the hiring manager, a 12-year veteran of the Search rankings team, put it this way: "She did the work. She shipped the model. But her promotion packet reads like she was a passenger." The candidate had spent 14 months on a Gemini-powered feature, passed three AI Reviews, and had zero mention in the packet of how she influenced the process, changed the standard, or mentored others through it.

The 1on1 agenda is your mechanism to surface this narrative before it goes to the Promotion Committee. Not after. Not during packet writing. Before.

Insight 1: The "Invisible Work" Tax

Google's AI Review process, hardened after the 2018 Maven controversy and the 2020 Timnit Gebru departure, now requires extensive documentation: Model Cards, Fairness Indicators, impact assessments. ICs who merely comply—who tick boxes—become invisible. The L6 who advances is the one who changes the box. In a 2024 conversation with a Google Research director overseeing the PaLM team, he described the difference: "One engineer gives me a Model Card. Another gives me a revised Model Card template that three other teams adopted. I only remember one of them."

Your agenda must force this distinction into the open.

How Do I Structure a 1on1 Agenda When My Manager Doesn't Understand AI Reviews?

Your manager's ignorance is your opportunity to demonstrate leadership upward.

Most Google engineering managers, especially those promoted through the traditional SWE ladder rather than ML-specialized paths, have never sat through a CAIRB review. In a 1on1 I shadowed in April 2024 between a L5 ML engineer and her manager on the Google Ads Prediction team, the manager opened with: "So how's the AI Review going?" The engineer spent 12 minutes describing documentation steps. The manager nodded. The meeting ended. Nothing in the promotion packet changed.

The corrected agenda—delivered the following quarter after intervention from a senior staff engineer—looked like this:

Section Content Time
Systemic Impact Narrative "I identified that Fairness Indicators for our demographic segment were undefined. I proposed and piloted a new metric now adopted by two sibling teams." 5 min
Process Influence Evidence "My revised Model Card template reduced CAIRB round-trips from 3.2 to 1.1 average. Here is the doc and the three teams using it." 5 min
Mentorship Documentation "I ran four AI Review dry-runs for junior engineers. Their average time-to-first-approval dropped from 6 weeks to 3 weeks." 5 min
Promotion Packet Alignment "Here is how this maps to L6 expectation 'Shapes technical direction for a large team or product area.'" 5 min

The Staff engineer who coached this format, previously at DeepMind before the Google merger, had seen it work in three promotion cycles. "The manager doesn't need to understand AI Reviews," he told me. "They need to understand you own a thing that other people need."

> 📖 Related: Google Analytics vs Mixpanel for PM Data-Driven Decisions: Which Analytics Tool Fits Your Team?

What Specific AI Review Artifacts Should I Bring to Prove Systemic Impact?

Artifacts without adoption stories are dead weight.

Google's AI Review generates numerous documents: Model Cards, Data Cards, Fairness Attestations, Launch Calendars, post-launch monitoring plans. The error—seen in a Q1 2024 debrief for a TensorFlow team L5—was treating these as outputs rather than as evidence of organizational change. The candidate's packet included 47 pages of Model Card documentation. The Promotion Committee's feedback, relayed by the manager: "Impressive thoroughness. No evidence anyone else used this or that it changed how we do AI Reviews."

The corrected approach, executed by a L6 on the Google Ads team who made Staff in 2023:

  1. The Before/After Metric: "CAIRB feedback cycles averaged 4.2 per launch. My standardized pre-submit checklist reduced this to 1.8. Here is the sheet, the 12 launches using it, and the CAIRB chair's email acknowledging the improvement."
  1. The Cross-Team Diffusion Story: "My Model Card section on temporal drift was copied into three other teams' templates. Here are the CLs."
  1. The Institutional Memory Creation: "I discovered no documented guidance existed for multimodal model reviews. I authored and got approved the first internal wiki page, now the top result for 'multimodal CAIRB' in Google internal search."

In a November 2023 conversation, the CAIRB chair for Google Cloud's AI products described what separates promotable ICs: "They don't just survive the process. They make the process better for the next person. And they document that the next person actually used it."

How Do I Time These 1on1s Relative to Promotion Cycles and AI Review Milestones?

Timing determines whether your impact is narrative or noise.

Google's promotion cycles run twice yearly: "H1" (packets due March, decisions June) and "H2" (packets due September, decisions December). AI Reviews, however, operate on launch timelines—unpredictable, often compressed. The IC who waits until packet season to retrospectively construct impact stories fails. The successful IC, observed in a 2024 study of 23 promotion packets by a senior staff engineer in Google Research, begins agenda construction at least two quarters before packet submission.

The specific timeline from a L6→L7 promotion case in Google Search (2023):

Date 1on1 Agenda Focus Artifact
T-6 months "I will lead the AI Review for launch X. Here is my plan for making this review produce reusable assets." Commitment document
T-4 months "The review is complete. Here is what I learned and what I will change for the next team." Retrospective with metrics
T-3 months "Here is the reusable asset and the three teams piloting it. I need your help validating adoption." Manager check-in with named teams
T-2 months "Here is the promotion packet draft. This section maps to systemic impact. Does this narrative hold?" Packet section for review

The manager in this case, a 15-year Google veteran now at L8, remarked in a 2024 panel: "The engineers who surprised me were the ones who treated me as a resource, not an obstacle. They didn't dump a packet on me. They built the story with me over quarters."

> 📖 Related: Software Engineer Interview Playbook vs Alex Xu System Design for Google L5 Coding

When Should I Escalate Beyond My Manager in the AI Review Process?

Escalation is not betrayal. Failed escalation is.

Google's matrix structure means AI Reviews touch multiple reporting lines: your engineering manager, the product manager, the Responsible AI partner, and potentially a Tech Lead or Staff engineer with CAIRB experience. A critical 1on1 agenda function is identifying when your manager cannot alone validate your systemic impact—and pre-negotiating that escalation.

In a Q2 2024 case from the YouTube Recommendations team, a L5 engineer's AI Review stalled for 11 weeks over an unresolved fairness metric. The manager, new to ML, repeatedly advised "just address the feedback." The engineer's 1on1 agenda, finally, included: "I believe we need a second opinion from [named Staff engineer] who led the 2022 CAIRB policy update. I have drafted the specific question for him. Can you introduce us, or should I reach out directly?"

The difference: the engineer did not ask permission to escalate. She presented the escalation as inevitable and asked only about protocol. The manager agreed to the introduction. The review unblocked in two weeks. The promotion packet included: "Identified systemic blocker, engaged cross-functional senior leadership, resolved 11-week delay."

A senior director in Google AI, speaking at an internal 2023 conference, noted: "The ICs we fast-track are not the ones who never need help. They are the ones who know exactly whose help they need and make it easy to say yes."

Preparation Checklist

  • Map your last four AI Reviews to the Google L5/L6 rubric specifically—"Shapes technical direction for a large team" or "Identifies and solves ambiguous problems"—and identify gaps where your narrative is weak
  • Draft one "before/after" metric for each AI Review artifact you produced: time, accuracy, adoption count, or reduction in review cycles
  • Identify three specific individuals outside your immediate team who used, referenced, or were affected by your work; confirm they will acknowledge this if asked
  • Prepare a one-page "promotion narrative" document that your manager could, in theory, submit with minimal editing; bring this to a 1on1 as a discussion anchor
  • Schedule a pre-mortem 1on1: "If my promotion packet were rejected for insufficient systemic impact, what would the committee say?" Run this conversation six months before submission, not after
  • Work through a structured preparation system (the PM Interview Playbook covers Google-specific ladder expectations and 1on1 agenda framingRegional
  • Review examples from real Staff engineer promotion packets, available through internal Google tools like "Promotion Packet Library" or peer networks; do not invent formats

Mistakes to Avoid

BAD: "I completed the AI Review documentation thoroughly and on time."

GOOD: "My standardized checklist reduced CAIRB round-trips from 3.2 to 1.1 average across 12 launches, adopted by the Ads and Cloud ML teams per these CLs."

The problem is not your thoroughness. It is your thoroughness without transferable impact.

BAD: "I mentored junior engineers on AI Reviews."

GOOD: "Four engineers I mentored achieved first-time CAIRB approval; before my mentorship, first-time approval rate was 23%. Here are their names and launch dates."

The problem is not that you mentored. It is that you cannot prove the mentorship changed outcomes.

BAD: "My manager doesn't understand AI Reviews, so I can't get good guidance."

GOOD: "I will use my 1on1 to educate my manager upward, bringing artifacts and metrics that make my impact legible without requiring technical AI Review expertise."

The problem is not your manager's ignorance. It is your failure to make your work interpretable by non-specialists.

FAQ

How often should I update my 1on1 agenda for AI Review impact discussions?

Every two weeks minimum. In a 2023 analysis of successful L5→L6 promotions at Google, the average was 2.3 weeks. The failure mode: quarterly updates that retroactively construct narratives.

Promotions committees at Google, especially in AI/ML where skepticism runs high after public controversies, discount reconstructed stories. The engineer who updates biweekly produces contemporaneous documentation: dated agenda items, manager responses, incremental metrics. One Promotion Committee member from Google Cloud, after a 2024 cycle, told me: "I can smell a packet built in the last month. The ones built over quarters have different texture—specific dates, evolving metrics, manager notes that show ongoing conversation."

What if my AI Review work is classified or restricted from broad discussion?

Classified work requires classified impact narratives. In a 2024 case from Google DeepMind, an IC on a restricted project could not name models, launch dates, or even specific Fairness Indicators. Her solution: metricized abstraction. "I led the AI Review process for a launch in [sensitive product area]. The standard review cycle was 6 weeks.

My process innovations reduced this to 3 weeks. Three senior reviewers adopted my approach. Here is the anonymized framework and their anonymized feedback." The Promotion Committee, cleared for the classified material, could map this to reality. The public packet remained sanitized. The key is not less documentation—it is documentation at two levels of abstraction, prepared in advance with legal and compliance review.

Should I ever use my 1on1 to discuss AI Review failures or blockers?

Immediately and specifically. The Google promotion rubric for L6 rewards "Identifies and solves ambiguous problems." A flawless AI Review trajectory suggests you selected easy problems. In a 2023 packet from Google Search that advanced to Staff, the engineer devoted 30% of his systemic impact section to a failed AI Review: "I proposed a novel fairness metric that CAIRB rejected after extensive analysis.

I documented the rejection rationale, shared it with three sibling teams, and all subsequently avoided the same error. The standard was updated based on my case." The Promotion Committee's comment, per the manager: "Rare to see someone own a failure this productively. Strong signal." The failure was not the problem. The failure without learning was the only real failure.

---amazon.com/dp/B0GWWJQ2S3).

Related Reading

What Is the Actual Purpose of a 1on1 Agenda at Google for AI Impact Discussions?