What I know I can't do yet
Also in Chinese: 中文版
The resume tells you what I can do. This page tells you what I know I cannot do yet. The dashed target line on the radar is a commitment with acceptance criteria, not a decorative outer ring — and the rule for this page is that every gap below carries a testable definition of “closed.” Without that rule, pages like this one degrade into motivational copy.
Priority one: the traditional face of security and compliance (40 → 65)
The gap. My security work is real but one-sided: it lives on the AI-agent frontier — deny-by-default tool gateways, write attestation, blast-radius-tiered execution gates. The traditional face — IAM design as an architecture discipline, audit frameworks of the SOC 2 class — is absent from my record, and the 40 on the radar says so plainly.
Why it goes first. It is the largest absolute gap on the radar; operating data platforms in fraud-adjacent domains makes compliance a when-question, not an if-question; and it is the axis least closable by reading — audit and IAM work has to be participated in, because the hard parts are organizational, not technical.
What counts as closed. A participation record in one real compliance audit, or an IAM governance project — role-model design plus an operating access-review loop — that I owned end to end and can narrate at the same evidence standard as the rest of this site.
Priority two: SLO and error budget as an operated practice (72 → 82)
The gap. I have the methodology (SLI/SLO layering, symptom-based alerting) and a working alert architecture. What the record does not contain is the sentence “I defined the SLO for this service and operated it for a quarter.” Knowing the theory of error budgets and having spent one are different credentials.
Why it matters. This is the piece that converts incident response from a reactive skill into reliability governance — the difference between putting out fires well and deciding, with a number, how much fire the organization should tolerate. It is also the single highest-leverage item for the incident axis, and the observability axis inherits most of the gain.
What counts as closed. An SLO definition document for a real service, plus at least one documented decision driven by error-budget burn: a release held back, or reliability work reprioritized over features, with the budget arithmetic attached.
Priority three: cross-team influence (52 → 75)
The gap. The influence record is real but local: internal training, an alert-governance initiative, one audit that reversed a leadership decision. What is missing is a complete narrative at the level of “initiated a cross-team effort”: I proposed it, other teams committed people to it, and it shipped.
Why it matters. This is the hard gate above senior. Without it, every other axis caps out at personal excellence — the organization gets a strong engineer instead of a multiplier.
What counts as closed. One project I originated, spanning two or more teams, with committed resources, traceable from proposal to outcome. Not participation in someone else’s initiative; origination.
The direction underneath
The through-line these three investments serve is a deliberate one: SRE evolving into AI infrastructure — agents as operators, platforms that govern them, observability of them, runtimes underneath them — as the long-term direction. I keep that thread on its own page rather than expanding it here. It belongs in this page for one reason only: the three gaps above were chosen to survive that transition, not despite it. Compliance judgment, SLO governance, and cross-team origination are precisely the parts of the SRE role that stay human as investigation itself gets automated.