Intensive hands-on workshops, cohorts of 10. Your people become expert-level Claude Code trainers. They train everyone else.
Enterprise AI rollouts follow a predictable pattern: deploy the tools, commit the budget, then watch adoption stall. The infrastructure works. The models work. But the last mile — getting the workforce from "tool deployed" to "tool productive" — is where the investment breaks down. Recent industry research puts a number on it: among employees who received structured AI training, 64% reported productivity gains of 20% or more. Among those with no training, 36%. That's a 28-point gap, and it isn't coming from better models or faster infrastructure.
Most organizations know this. Over 60% of executives have formally linked AI spending to ROI targets. But the same discipline hasn't reached training: 61% rely on workers to learn on the job, fewer than half provide dedicated time for that learning, and over half of employees are seeking training from outside sources at their own expense. More speed shouldn't mean more problems — but without structured skilling, that's exactly what happens.
This is not going to happen organically. Workforce AI competency requires the same deliberate design that the AI systems themselves required. The metric that matters is time-to-competencyThe elapsed time from tool deployment to the point where a worker uses it well enough to justify the investment. The metric that matters for training ROI — not hours of training delivered, but how fast productivity gains appear. — not how many hours of training were delivered, but how fast your people reach the point where the tool pays for itself.
This is not a lecture series. It's engineering, not magic — every session is hands-on: short concept introduction, then exercises that your trainers work through in real time with real projects. Learning happens in the flow of work, not in a separate classroom. By the end of each day, they've done the thing, not just heard about it.
Exercise: Each trainer configures a CLAUDE.mdA project configuration file that tells Claude Code how to behave in a specific repository. Like .editorconfig for formatting rules, but for AI behavior — coding standards, cost limits, domain-specific instructions, what to never do. for an actual project — encoding coding standards, cost rules, and domain-specific instructions. They set up permission systems, governance boundaries, and observability. They run sessions with and without configuration and measure the difference. By end of day, every trainer has a strong foundation: a working project config they built themselves.
Exercise: Trainers take the same coding task and complete it two ways: naive and optimized. Token managementControlling how much text the AI processes per request. Tokens are the unit of AI consumption — roughly ¾ of a word. Inefficient prompting wastes tokens like inefficient queries waste database IOPS. Optimization means doing the same work with fewer tokens., model routingSelecting the right AI model for each subtask. Like choosing between a forklift and a hand truck — the expensive model for judgment calls, the cheap model for mechanical work. Can reduce costs 75% with no quality loss., grep-before-read, parallel tool calls, session chunking. They see 75% cost reduction in their own usage, not a slide deck claiming it. The goal: throughput, quality, and consistency — not just speed. Techniques become muscle memory through repetition.
Exercise: Trainers run the Triad patternA role-separation architecture where one AI instance plans and another implements, with neither seeing the other's raw output. Like the auditor/developer split in SOX compliance — the separation is the control. Prevents self-validation drift. live — separate Claude instances for coding and auditing, directional information flow, neither seeing the other's raw output. They experience the failure mode first (single agent self-validates, misses drift) then the fix. This is a deliberate design discipline, not a productivity hack — the pattern clicks when you've watched it catch something you missed.
Exercise: Multi-agent orchestration on a real codebase. Automated workflows, specification-driven development, structural debugging. Trainers design a workflow, run it, break it, diagnose why it broke, and fix it — all in the session. They learn the real strengths and weaknesses of the tool, not the marketing version. Context hygiene for long sessions: they'll hit the wall and learn to manage it.
Exercise: Each trainer designs and delivers a 30-minute teaching segment to the cohort, covering one pattern from Days 1–4. Peer review and instructor feedback. They leave with a living curriculum — a tested lesson plan they've already delivered once, built to evolve as the tool does. Not notes they hope to turn into a class later. Measurable impact: can your trainer teach it cold on Monday morning?
I've used Claude Code as my primary development tool since its launch — daily, across production systems, for over a year. Not demos. Not prototypes. Shipping work.
Don't take my word for it. Before we discuss scope, pricing, or scheduling — test me. Give me a coding problem, a configuration challenge, a workflow design question, whatever your best people would find hard. I welcome any evaluation you want to run, no commitment required. Claims are cheap; demonstrated capability isn't.
| Delivery | Rate | Details |
|---|---|---|
| Remote | $5,000 / day | Live sessions via video. Full curriculum. Timezone-flexible scheduling. Significant expenses additional. |
| On-site | $10,000 / day | In-person delivery at your location. All travel, accommodation, and significant expenses additional. |
| Follow-up support | $250 / hour | Live consultation, curriculum review, troubleshooting. 30-day window per cohort. Routine email questions included at no charge. |
The model: Cohorts of 10 — small enough that every trainer gets individual attention, works exercises at their own pace, and gets direct feedback. This is a hands-on workshop, not a webinar. To train 50 people, run 5 cohorts.
Why 10: Hands-on AI training fails at scale. When someone is stuck on a configuration problem or a workflow that isn't clicking, they need the instructor's attention now, not after 49 other people. The exercises are the point; the class size protects them.
This program doesn't just teach AI-assisted engineering — it runs on it. Alongside the instructor, each participant works with a dedicated, course-trained AI teaching assistant for real-time, one-to-one coaching. The routine "why isn't this working?" moments get unblocked instantly, so the instructor's attention goes where it matters most — the hard problems. It's the same workflow the course teaches, demonstrated in the room. A signature feature of the program, tailored per cohort.
Adjustable scope: The 5-day structure is a baseline. Programs can be compressed to 3 days (essentials only) or extended with role-specific tracks (security, DevOps, data science) for enterprise-scale workforce transformation. My objective is that your trainers succeed — quick email questions are always welcome; substantive follow-up work is billed hourly.