Mid-Level IC Engineer at Google: When AI Performance Reviews Ignore Your Hidden Impact

TL;DR

The AI‑driven performance system at Google systematically down‑weights work that isn’t captured by its metric suite, and the only way to win is to inject manual signals that the model cannot infer.

Mid‑level engineers who rely on “good code” alone will be flagged as low‑impact, even if they are the silent architects of cross‑team reliability.

The decisive judgment: treat the AI review as a data‑point, not a verdict, and build a parallel narrative that forces leadership to see the hidden value.

Who This Is For

If you are a Software Engineer III or IV at Google earning between $185,000 and $215,000 base, have shipped at least two major features, and are staring at a pending performance cycle that will decide your next promotion, this piece is for you. You are likely frustrated that your contributions to latency‑critical services, mentorship, and undocumented tooling are invisible to the AI‑scorecard, and you need a concrete framework to make those contributions count.

Why do AI‑driven performance reviews miss the hidden impact of mid‑level engineers?

The answer is that the model is trained on observable outputs—code commits, bug‑fix counts, and sprint velocity—while ignoring the “invisible work” that senior staff rely on to keep the ecosystem stable. In Q2, during a debrief for the “Reliability Team,” the hiring committee asked why the AI rating for an engineer who had reduced service‑wide latency by 12 % was lower than a peer with more visible feature flags. The committee’s rebuttal was that the model could not quantify the engineer’s “latent‑risk mitigation” because no explicit ticket existed. The judgment: the AI system is a proxy for productivity, not a proxy for impact; you must translate hidden work into the language the model understands.

Counter‑intuitive insight #1 – the problem isn’t the lack of data, it’s the lack of structured data. Engineers who treat their undocumented scripts as personal tools are effectively hiding the very evidence the AI needs. By logging every reliability run as a tracked ticket, even if the ticket never moves to “Done,” you create a data trail that the model can ingest.

Framework – Impact‑Signal Lattice: map each hidden contribution (e.g., “runtime‑guard scripts”) to a visible signal (e.g., “ticket #12345 – performance audit”) and assign a weight based on senior leadership interest (e.g., “Latency < 100 ms” gets higher weight). This lattice forces the AI to see the invisible work as a quantifiable metric.

How can I surface my invisible contributions before the next review cycle?

You must proactively insert manual “signal injections” into the review pipeline, because the AI will never infer impact that you never surface. In a recent Q3 performance debrief, a senior manager pushed back on the AI score for an engineer who had introduced a cross‑team alerting framework. The manager demanded a “visibility packet” that included: (1) a one‑pager linking the alerting framework to the company‑wide SLO‑breach reduction, (2) a dashboard screenshot showing a 30 % drop in incident frequency, and (3) a peer endorsement email. After the packet was attached to the review, the engineer’s rating jumped from “Meets Expectations” to “Exceeds Expectations.”

Counter‑intuitive insight #2 – the problem isn’t the AI’s inability to see your work, it’s your reluctance to package it as a formal deliverable. The “not a deliverable, but a contribution” mindset blinds you to the fact that the review system rewards documentation over intuition.

Practical script: “I’ve compiled a concise impact summary linking my reliability scripts to the Q2 SLO improvements; I’ll attach it to the review portal so the model can factor it into the score.” This line forces the reviewer to treat the hidden work as a formal artifact.

What signals do senior leaders actually weigh when AI grades my output?

Senior leadership looks for three core signals: (1) measurable product outcomes, (2) cross‑team influence, and (3) forward‑looking ownership. In a January debrief, the VP of Engineering asked why the AI rating for a mid‑level engineer with a high commit count was lower than a peer with fewer commits. The answer was that the peer’s commits were tied to a product KPI—user‑engagement growth of 4 %—while the high‑commit engineer’s work was confined to internal tooling. The judgment: the AI model mirrors leadership’s signal hierarchy, so you must align your metrics with those three pillars.

Counter‑intuitive insight #3 – the problem isn’t the AI ignoring “hard work,” it’s the AI rewarding “hard‑to‑measure work.” Your invisible impact becomes valuable only when you can tie it to a KPI that leadership cares about, such as “Mean Time to Recovery (MTTR) reduced by 18 %.”

Organizational psychology principle – Expectancy Theory: engineers will allocate effort toward outcomes they believe will be recognized. By explicitly stating the expected reward (e.g., “This alerting framework will cut MTTR by X”), you align your effort with the review system’s reward structure.

When does a “good” engineering score turn into a promotion blocker?

A “good” AI score (e.g., 4.2 out of 5) becomes a blocker when the accompanying narrative is silent on strategic impact. In a March promotion committee, an engineer received a 4.3 rating but was denied L5 because the reviewer noted “no evidence of leadership beyond immediate code ownership.” The committee’s decision hinged on the lack of a “leadership narrative” rather than the numeric score. The judgment: a solid numeric rating is insufficient; you must embed a promotion narrative that demonstrates strategic breadth.

Not “good code, but strategic influence.” The AI rating alone does not guarantee promotion; the narrative does.

Script for promotion packet: “Beyond the three shipped features, I led a cross‑team effort to refactor the logging pipeline, resulting in a 15 % reduction in log storage cost and a measurable improvement in downstream analytics latency.” This statement converts a good score into a compelling promotion case.

Which compensation levers can I pull if the AI review undervalues my work?

If the AI review systematically undervalues your hidden impact, you can negotiate compensation by anchoring on objective market data and internal equity. In a recent salary negotiation, an engineer with a 4.0 AI rating leveraged a “compensation audit” that showed comparable L5 engineers in the same office receiving $215,000 base plus $25,000 sign‑on. The engineer framed the request: “Given my documented contributions to latency reduction and cross‑team reliability, I am seeking an adjustment to $210,000 base and 0.04 % equity.” The compensation team approved the increase because the request was tied to concrete impact metrics, not just the AI score.

Not “the AI says I’m average, but I’m high‑impact.” The judgment: you must translate hidden impact into market‑aligned numbers and present them as a compensation correction rather than a dispute with the AI.

Preparation Checklist

Document every reliability‑related script as a ticket in the internal tracking system, even if it never reaches “Done.”
Build an Impact‑Signal Lattice linking each hidden contribution to a visible KPI (e.g., latency, MTTR, cost).
Draft a one‑page impact summary that includes screenshots of dashboards and explicit KPI changes.
Secure at least two peer endorsement emails that reference specific outcomes (e.g., “Your alerting framework reduced incident frequency by 30 %”).
Prepare a promotion narrative that highlights cross‑team influence and forward‑looking ownership.
Align compensation requests with internal equity data; cite concrete base‑salary ranges ($185,000–$215,000) and equity percentages (0.03–0.05 %).
Work through a structured preparation system (the PM Interview Playbook covers impact quantification with real debrief examples, and its templates help you phrase hidden work as measurable outcomes).

Mistakes to Avoid

BAD: Submitting an AI score without any supplemental narrative, assuming the model will “see” your hidden work. GOOD: Adding a concise impact packet that translates invisible contributions into measurable signals before the review deadline.

BAD: Treating internal tooling as “just a side project” and not logging it as a formal ticket, which leaves no data for the AI. GOOD: Recording every internal script as a tracked ticket, then mapping it to an SLO improvement in the Impact‑Signal Lattice.

BAD: Negotiating compensation solely on the basis of “the AI gave me a low rating,” which the compensation team dismisses as a model error. GOOD: Anchoring the negotiation on concrete market data and documented KPI improvements, framing the request as a correction to the performance metric, not a challenge to the AI.

FAQ

How can I prove that my undocumented work actually reduced latency?

Tie the work to a measurable KPI. Create a ticket for the script, attach a dashboard screenshot showing a 12 % latency drop, and request that the reviewer include that metric in the AI score. The judgment: only quantifiable evidence will shift the AI’s weighting.

What if the AI still gives me a mediocre rating after I’ve added all the signals?

Escalate with a promotion narrative that emphasizes strategic influence, not just the numeric rating. The judgment: the narrative can override a modest AI score when it clearly connects your work to company‑wide objectives.

Can I request a higher equity grant based on hidden impact?

Yes, but you must frame the request with market‑aligned numbers and concrete impact data (e.g., “My alerting framework saved $30K annually, justifying a 0.04 % equity increase”). The judgment: equity adjustments are granted when hidden impact is translated into financial terms the compensation team can evaluate.

Ready to build a real interview prep system?

Get the full PM Interview Prep System →

The book is also available on Amazon Kindle.