Kamil Rajtar

I help engineering teams turn Claude from a tool their devs tried once into a daily multiplier across the whole organization. Anthropic's own engineering org reported daily AI use going from 28% to 59% of work in twelve months — and a 67% increase in merged PRs per engineer per day after rollout. (Source: How AI is transforming work at Anthropic, Dec 2025.) That's the upper bound. Most teams I meet sit far below it. The gap isn't access. It's craft. Most teams have Claude licenses but no shared operating practice. One dev is wielding it as a daily harness — scoping work, loading context, steering the agent loop, recovering when it goes sideways. The other is copy-pasting into the web UI. Same access, different week. Four skills separate the operator from the tourist: Plan, Context, Harness, Review. Plus the meta-skill: operating, not (just) asking. I run two formats: → Essentials (4 hours) — for teams just getting started. By the end, every dev in the room can scope, prompt, and steer Claude on real engineering work. Hands on keyboards. On-site or remote. → Deep Dive (2 days) — Essentials plus the harness I actually use, custom skills authoring, agent patterns, and a working setup your team leaves with. On-site preferred. What changes 30 days after a Deep Dive: features ship in days not sprints, PRs arrive with tests and self-reviews already attached, and senior engineers spend more time on architecture than shepherding juniors through repetitive work. A Claude seat costs roughly 1% of a senior dev's loaded monthly cost. The question isn't whether you can afford it — it's what return you need to justify the 1%. Anthropic measured 27% of work that wouldn't have happened at all without Claude. That's the upside finance partners miss. If your team has uneven Claude adoption and you want a consistent operator baseline, let's talk.