Krzysztof Jankowski

2024-11

Prepared and presented AI infrastructure budget planning to senior leadership — detailed breakdown of LLM inference costs, vector database hosting, compute, and staffing for the coming year.

Challenge: LLM costs are highly variable and hard to forecast — usage spikes with new product launches and is sensitive to model pricing changes. Introduced per-team token budgets with monthly alerts and a shared 'innovation reserve' budget for experiments.

Learned: Budget presentations for AI infrastructure require separating fixed costs (compute, hosting) from variable costs (LLM tokens). Leadership needs to understand that LLM costs scale with adoption — this is a feature, not a bug.

2024-07

Led multi-tenant platform launch planning — 3 pilot customer organizations, phased rollout plan, go/no-go criteria for each phase, and risk register.

Challenge: Security review took 8 weeks instead of the planned 4 — IT security team was under-resourced and the AI platform review required new assessment criteria they had not developed before.

Learned: Security review timelines for AI platforms are consistently underestimated. Now build in 10 weeks as baseline assumption for enterprise AI platform security assessment, with escalation path if blocked.

2024-02

Implemented OKR framework for the AI Center of Excellence — quarterly objectives, weekly key result check-ins, and end-of-quarter scoring sessions.

Challenge: Initial OKRs with date-based KRs (e.g., 'deploy X by March 31') created pressure to ship incomplete features on time rather than right. Restructured KRs to be outcome-based (e.g., 'X is used by 100 engineers per week by quarter end').

Learned: OKRs for AI teams work best when KRs measure usage and impact, not delivery dates. Date-based KRs optimize for shipping; outcome-based KRs optimize for value.

2023-10

Facilitated Q2-Q3 AI project retrospectives across 3 teams. Synthesized findings and identified systemic blockers. Top finding: integration issues between AI services and enterprise systems consumed 30% of development time.

Challenge: Teams were reluctant to surface problems in retrospectives for fear of appearing unproductive. Created anonymous pre-retrospective surveys to surface real issues before the session.

Learned: Psychological safety in retrospectives requires active facilitation. Anonymous input before the session surfaces issues that would otherwise remain hidden, making the retrospective significantly more actionable.

2023-06

Coordinated APQP Phase 3 cross-functional program for new HPT component — synchronized design, quality, manufacturing, and supply chain milestones across 4 teams in 3 locations.

Challenge: Design freeze was delayed 6 weeks due to unresolved thermal stress analysis issues. Rebuilt the schedule with parallel workstreams to recover 4 weeks of the slip, accepting reduced testing time as a calculated risk.

Learned: When a critical milestone slips, don't just slide the entire schedule. Immediately identify which downstream activities can be parallelized and which can tolerate reduced scope. Recovering partial slip is much better than accepting the full delay.

2023-02

Ran Q1 AI initiative roadmap planning workshop — 8 teams, 23 proposed initiatives. Facilitated prioritization using a value vs. effort scoring matrix with input from both engineering leads and business sponsors.

Challenge: Engineering and business stakeholders had fundamentally incompatible priority rankings — engineers wanted foundational infrastructure, business wanted immediate customer-facing features. Resolved by creating two parallel tracks with a shared dependency schedule.

Learned: AI roadmap planning requires separating 'platform' and 'product' tracks explicitly. Mixing them in a single backlog creates permanent priority conflicts that derail planning sessions.